A Reconsideration of The Theory of Non-Linear Scale Effects

Downloaded from https://www.cambridge.org/core.
University of Delhi, on 16 Jan 2019 at 07:37:44, subject to the Cambridge Core

terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555029
Downloaded from https://www.cambridge.org/core. University of Delhi, on 16 Jan 2019 at 07:37:44, subject to the Cambridge Core terms of use
, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555029
Elements in Evolutionary Economics
edited by
John Foster
University of Queensland
Jason Potts
RMIT University
A RECONSIDERATION OF
THE THEORY OF NON-
LINEAR SCALE EFFECTS
The Sources of Varying Returns to,

and Economies of, Scale
Richard G. Lipsey
Simon Fraser University, British Columbia
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781108453097
DOI: 10.1017/9781108555029
© Richard G. Lipsey 2018
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2018
A catalogue record for this publication is available from the British Library.
ISBN 978-1-108-45309-7 Paperback
ISSN 2514-3573 (online)
ISSN 2514-3581 (print)
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
A Reconsideration of the Theory of Non-Linear
Scale Effects
The Sources of Varying Returns to, and Economies of, Scale
Elements in Evolutionary Economics
DOI: 10.1017/9781108555029
First published online: March 2018
Richard G. Lipsey
Simon Fraser University, British Columbia
Abstract: The main thrust of this Element is a critical assessment of the

theory and evidence concerning the sources of scale effects. It is argued
that the analysis of static scale effects is important because scale effects
are embedding in our world, and new technologies associated with an
evolving economy often allow their exploitation when they cannot be
exploited in less technically advanced and smaller economies. So,
although static equilibrium theory is not a good vehicle for studying
economic growth, showing how scale effects operate when output varies
with given technology helps us to understand the scale effects that occur
when output rises as a result of economic growth, even though that is
typically driven by technological change.
The set of production functions that are consistent with Viner’s
treatment of long-run cost curves are distinguished from the single
production function that is found in virtually all modern
microeconomic textbooks. It is argued that the inconsistencies and
ambiguities relating to the use of such a single production function to
cover all possible scales of a firm’s operations are such that it is an
imperfect tool for analysing the scale effects that firms actually face.
The relation between scale effects and the size of the firm are
discussed. It is shown that under certain commonly occurring
circumstances the ability to replicate production facilities is consistent
with short ranges of diseconomies of scale and an indefinite range of
increasing returns.
Next comes a detailed analysis of the sources of positive scale effects
and a critical assessment of the treatment of these in a large sample of
the existing literature. It is argued that the nature of our world, with its
three dimensions, its physical laws and the many random elements in
its behaviour, is such that when the scale of anything changes, we
should always expect to encounter non-linear scale effects. Most
authors list a series of examples of sources that are assumed to give
rise to scale effects but seldom attempt to show in any detail how these
are supposed to work. When we do this, some alleged sources are
found not to give rise to scale effects at all, while others have effects
that differ from what has been assumed. Furthermore, there is seldom
agreement among authors as to whether a particular source is a cause
of varying returns to scale or economies of scale. Most authors argue
that indivisibilities are an important source of scale effects, although
these are seldom well defined, nor are the precise ways in which these
are supposed to work typically analysed. When we do this, we identify
two basic types of indivisibilities, ex post and ex ante, plus several
variations of each of these main types. We then argue that the
discussion of indivisibilities has been confused by use of different
implicit definitions of the term and also that only one of these types of
indivisibility can be a source of scale effects. Constant-returns
production functions are found to be inconsistent with much that is
known about actual production techniques, even when firms expand
by duplicating identical plants. Unless ruled out by definition,
diseconomies of scale are found to be a real possibility in many
circumstances. When these occur in some parts of complex capital
goods or plants, they limit the extent to which economies in other parts
can be exploited by increasing the scale of the whole operation. Finally,
brief consideration is given to the literature concerning the factors that
limit the exploitation of the scale effects that are ubiquitous in the real
world and to the consequences of their exploitation.
Keywords: economies of scale, production function, returns to scale,

indivisibilities, replication, long-run cost curves.
© Richard G. Lipsey 2018

ISBNs: 9781108555029 (OC), 9781108453097 (PB)
ISSNs: 2514-3573 (online), 2514-3581 (print)
Contents
1 Background Conceptual Issues 2
2 The Sources of Positive Scale Effects 22
3 Constant and Variable Returns with Replication 56
4 Sources of DoS 58
5 Conditions Leading to the Exploitation of Scale Effects 64
6 Consequences of Exploiting Scale Effects 68
7 Conclusion and Summary 68
Classified References 81
A Reconsideration of the Theory of Non-Linear Scale Effects 1
We live in a three-dimensional world subject to both physical laws and uncer-

tainty. These cause ubiquitous scale effects related to the world’s geometry, its
physics and the random aspects of its behaviour. We define these effects gener-
ically as anything that affects the firm as a result of changing its scale of
operations. When new firms or industries are developed, old ones are expanded
or new technologies are developed, scale effects are typically encountered. Even
when these are fully exploited in equilibrium, their existence affects the approach
to and the nature of that equilibrium. All of these scale effects are typical of
growing economies, and they interact with economic growth in a relation of
mutual causation. (Different types of scale effects are distinguished later in the
Element, and since the Element contains a large number of terms, some of which
are novel, they are listed along with their definitions in Appendix A.)
Here are three relevant questions about scale effects: (i) What are the sources
of such effects? (ii) What conditions lead to their being exploited? (iii) What are
the consequences of the common existence of scale effects? The purpose of this
Element is to consider in detail the first question, although that inevitably over-
laps to some extent with the second question, which will be considered briefly
near the end of this Element but with no pretence at completeness. The important
question (iii) is already the subject of extensive literature and so is only con-
sidered briefly near the end of this Element.
Most of the analysis in microtheory textbooks is conducted assuming a
constant-returns-to-scale production function because the non-convexities
introduced by scale effects are analytically much more difficult to handle
than the convexities of constant returns. Indeed, when an aggregate production
function is used with labour and capital as the main inputs, scale effects pose
aggregation problems that are intractable for all practical purposes. As a result,
scale effects tend to be treated as exceptions rather than what they actually are:
the rule to which constant returns are the exception. Also, most of the treat-
ments in micro textbooks are highly abstract and readers are seldom asked to
give examples of what some formal result would look like in terms of real-
world examples. Indeed, few authors attempt to show how the sources they list
are supposed to operate in any detail. When we do this, some of the alleged
sources are found to have different effects from what have been assumed, and
others are found not to cause scale effects at all. Furthermore, many authors
argue that indivisibilities are an important source of scale effects, although
there is no agreement as to how these should be defined, nor are the precise
ways in which these are supposed to work typically modelled in any formal
I am indebted to Kenneth Carlaw, Curtis Eaton and Colin McLean for comments and suggestions,
and also to Colin McLean for his indefatigable research efforts.
2 Elements in Evolutionary Economics
way. When we do these things, we find many confusions and mistaken con-
clusions in the existing literature on indivisibilities.
In preparing the Element, I surveyed entries related to the sources of scale
effects in all economic encyclopaedias and dictionaries of economics that I
could locate. I also surveyed a selection of microeconomics textbooks aimed at
various levels and some relevant articles. I stopped covering textbooks when I
felt I had enough to illustrate the treatment of scale effects in the literature so
that little would be gained by surveying more. In Section 2, I investigate the
working of some of these alleged sources both to check on their validity as
sources and to illustrate, without any attempt to be exhaustive, the unexplored
complexity of the effects associated with some of the alleged sources.
After some preliminary ground clearing, the Element discusses the representa-
tion of scale effects through the firm’s cost and production functions. A treatment of
the sources of scale effects, particularly in the reconfiguration of capital goods, leads
to a distinction between the set of production functions that are consistent with the
Viner treatment of the long-run cost curves and the unique production function that
is found in virtually all modern micro textbooks. Problems related to the concept of
a production function that spans the whole of input space are first discussed in the
text and then elaborated in Appendix B. After that, the relation between scale effects
and firm size is discussed. Most of the rest of the Element provides a critical
assessment of, and an elaboration on, the treatment of the sources of scale effects
in the literature. It argues that the nature of our world is such that when the scale of
almost anything is changed, we should expect to encounter scale effects. Surprises
should only occur when such effects are not encountered. A major section on the
confusion over the meaning of indivisibilities, and the manner in which they do and
do not cause scale effects, is followed by a treatment of the sources of scale effects
that are associated with the design of capital goods. Finally, brief consideration is
given to the literature concerning the factors that limit the exploitation of the scale
effects that are ubiquitous in the real world and to the consequences of their
exploitation.
1 Background Conceptual Issues

1.1 The Use of Static Analysis
Most of the study of the sources of scale effects in the literature is done by
comparing equilibrium situations in static models assuming given technologi-
cal possibilities. Although many in the scale effects are associated with eco-
nomic growth, much can be learned from the study of scale effects using
comparative statics, which we do at some length in this Element. The analysis
of static scale effects is useful because scale effects are embedded in our world,
and new technologies associated with an evolving economy often allow their
exploitation when they cannot be exploited in less technically advanced and
smaller economies. Although static equilibrium theory is not a good vehicle for
studying economic growth, it can be used effectively to study both the nature
and sources of these scale effects. By showing how and why they occur when
output varies with present technology, we can better understand the scale
effects that occur when output rises as a result of economic growth, even
though that is typically driven by technological change. For but one example,
the new technology of the transcontinental railway system turned much of the
US economy from a subset of small markets, isolated from each other by high
transport costs, into a single unified market whose size allowed the exploitation
of scale effects that were inherent in the real world and could be shown in static
models but were unavailable to small firms operating in regional markets.
1.2 Issues of Definition

The extensive literature on scale effects that we have surveyed for this Element
contains many different and sometimes conflicting approaches. Scale effects are
defined in three ways in the literature, two explicitly using a production or a cost
function and one implicitly using a dynamic growth concept. Some authors sharply
distinguish between the two explicitly defined concepts and argue that they are not
wholly overlapping, while others switch back and forth between them without
comment, appearing to hold that they are synonymous. Furthermore, there is little
agreement among the various authors as to which of the two concepts applies in
each of the cases studied. The third, the implicitly defined, growth-related concept,
has been a source of confusion in the literature because it is given the same name as
one of the explicitly defined concepts and, although both refer to different and
complimentary phenomena, many authors have argued that one is right and the
other, if not wrong, is both unimportant and misleading.
When a firm’s unit costs change as its scale of operations changes, it is useful
to distinguish between real and purely pecuniary scale effects. There is a real
resource effect when there is a change in the amount of inputs per unit of
output. There is a purely pecuniary effect when there is a change in the price of
one or more of the firm’s inputs with no related reduction in real resources used
per unit of output.1
1
This distinction was once commonly made in the literature. However, of the many authors we
surveyed, only the minority make the distinction between real and purely pecuniary effects.
These are Bain (1968: 492), Bannock et al. (1984: 141–2), Becker (1999: 150), Bohm (2008:
189), Calhoun (2002: 357), Frank (2008: 359), Graaff (1987: 7599), Jackson (1996: 229) and
Pearce (1992: 122). With the exceptions of Becker and Bain, these authors do not comment on
the relative benefits or desirability of separating these two categories.
An example of a purely pecuniary effect is when the growth of an assembly

firm allows it to gain market power over the many small suppliers of its parts. It
can then transfer profits from the parts suppliers to itself with no change in real
resource costs anywhere in the economy. An example of a real cause is when the
increase in the assembler’s size allows its parts suppliers to increase their size and
reap more of their favourable scale effects. (Of course, to have unexploited scale
effects in equilibrium, the suppliers must be price setters, not price takers, which
is the normal case with differentiated products.) Note, however, that although this
will appear as a real effect to the parts suppliers, it will appear, if the saving is
passed on in terms of lower parts prices, as a pecuniary effect to the final goods
producer. This example illustrates a general problem with the real–pecuniary
distinction. Virtually all of the analysis of scale effects in the literature deals with
a single firm. This would pose no problem if the production of raw materials,
capital goods, other intermediate products and final goods were all integrated
within a single firm. As it is, however, production is disintegrated into a large
number of stages. As a result, what originates as a real cause upstream can appear
as a pecuniary cause to downstream producers. We look at this issue in more
detail at the beginning of Section 2.
Scale effects can be divided between those that arise at the level of individual
items, usually capital equipment; the individual production facility, which with
some violence to reality, we call a ‘plant’; the whole firm; the industry; and the
whole economy. This Element is mainly concerned with the first three sorts, all
of which are internal to, i.e. under the control of, the firm. External effects at
levels higher than the firm do exist, and we give some attention to them towards
the end of the Element.
1.3 Approaches to Studying Scale Effects

The literature on scale effects typically considers a single firm that can be
assumed to be producing some final consumers’ good. In dealing with the scale
effects facing this firm, three main approaches can be discerned. First, in
microtheory texts, scale effects are treated almost exclusively in terms of the
characteristics of a production function that relates physical inputs to physical
output. The typical scale exercise is to alter all inputs by some constant multi-
ple, λ, and observe the response of output, about which much more is said
below. The analysis is usually purely formal and the underlying real-world
causes of the characteristics of the production function that give rise to these
effects are usually not considered in any detail, if at all. Then, if the firm’s costs
are required, given input prices allow a cost function to be derived from a
production function.
Unit Cost
SRATC1 SRATC7
SRATC2
SRATC6
AC
SRATC3
LR
SRATC4 SRATC5
c1
0 q1
Output
Figure 1 Short- and long-run costs
Second, in the encyclopaedias, in most introductory textbooks and in some

journal articles, the sources of scale effects are investigated in some detail. Our
main concern in this Element is to follow in this tradition by seeking to explain,
but in more depth than is usual in this literature, the reasons why a firm’s costs
vary in specific ways when its output varies over the long run with constant
technological possibilities and in the very long run when these possibilities can
change.
The third tradition is related to the growth literature and is considered in a
later part of this section.
1.4 Displaying Scale Effects

Static scale effects, those that exist at some point in time, can be shown using
either cost curves or production functions which lead to different but related
definitions.
1.4.1 Through Cost Curves

The original graphic description of scale effects is due to Viner (1931) and is
shown in Figure 1, a similar diagram to his. Input prices are assumed to be
given and the curves then depend on the distinction between the long run, when
all inputs can be varied, and the short run when some inputs are fixed, usually
taken to be the plant and its equipment. The primitives in this treatment are the
family of short-run average (or unit) total cost curves shown as the SRATC
curves in the figure. Each curve relates to a different size of plant and possibly
to a different production technique (as, for example, when mass production is
used instead of craft-style production above some critical level of the firm’s
operations). The derived curve is the long-run average cost curve (LRAC
curve), which is the envelope to the SRATC curves. If size of plant can only
be varied discretely, the LRAC curve follows each SRATC curve until it is
intersected by the curve derived from a larger scale of operations. In the limit, if
techniques can be varied continuously, each point the LRAC curve relates to a
different size of plant.2
The typical textbook LRAC curve for a firm is U-shaped as shown in the
figure. The negative-sloping section indicates what are typically called ‘econo-
mies of scale’ (EoS); the positively sloped section, ‘diseconomies of scale’
(DoS); and the horizontal portion, if one is assumed, the absence of scale
economies (CoS).3 We refer to these collectively as ‘efficiency effects’. The
output at the minimum point on this curve is called the minimum efficient scale
or MES (q1 in the figure); ‘minimum’ to allow for the fact that there may be
more than one output that achieves this lowest unit cost − as shown for example
in Figure 2. Although as already pointed out, many authors who identify
efficiency effects do not distinguish between those that result from real
resource savings and those that result solely from changes in the prices of the
firm’s inputs with no resource savings, we do so when needed.
1.4.2 Through a Production Function
The analysis of scale effects in microtheory textbooks typically uses the

relation between inputs and output as described by a production function for
a firm that, as already mentioned, can be assumed to be producing some final
consumers’ good:
yi ¼ fðs1 ; . . . ; sn ; snþ1 ; . . . ; sm Þ (1.1)
where yi is the ith final good and s1, . . ., sm are the service flows of m individual
inputs. Inputs s1 to sn are those that are held constant in the short run while sn+1
to sm are those that can be varied in both the short and long runs.
The returns effect is then defined as follows:
2
Since it is costly to develop a production facility with a distinct SRATC curve, a firm will only be
presented in practice with a finite number of possible plants and hence SRATC curves.
3
It is worth noting that this modern usage is not quite that of Alfred Marshall’s whose concept of
internal economies ‘is analytically looser than this [the modern usage], but richer in empirical
content and, possibly, in philosophical insight’ (Becattinni 2008: 419).
f ðλs1 ; . . . ; λsm Þ ¼ αλ f ðs1 . . . ; sm Þ; (1.2)
where αλ > λ, αλ = λ and αλ < λ indicates respectively what are called increasing
returns to scale (IRTS), constant returns to scale (CRTS) and decreasing returns
to scale (DRTS).4 We refer to these collectively as ‘returns effects’ and will
have much more to say about them later in the Element. The previously
introduced term ‘scale effects’ thus covers both returns and efficiency effects
and any other changes due to alterations in the firm’s scale of operations.
In the literature surveyed the large amount of disagreement on allocating
sources between the returns and the efficiency categories suggests problems
with the procedure of dividing all sources of scale effects into one or the other
of these classes – problems that I discuss more fully in a subsequent section.
1.4.3 Through Growth
The above two ways of viewing scale effects are through static relations that exist
at any point in time. A different view defines scale effects as something that can
only be observed through changes rather than static relations. Authors in this
tradition often quote Allyn Young (1928) as a precursor. Although he does not
define the term ‘increasing returns’, taking its meaning as obvious, he has a
conventional view of their sources, saying for example (p. 538): ‘In so far as it is
an adjustment to a new situation created by the growth of the market for the final
products of industry the division of labour among industries is a vehicle of
increasing returns.’ His main concern is to argue that the exploitation of scale
effects is permitted by the extension of the market for some industry. This may be
due to economic growth or to a reallocation of resources into that market. Indeed,
the figure in his concluding NOTE (p. 540) implicitly defines increasing returns
as a falling opportunity cost of commodity X (e.g. manufactured goods) in terms
of foregone Y (e.g. agricultural goods) as the production of X increases and that of
Y decreases, which is resource reallocation not a growth phenomenon. So a
discussion of Young really belongs in the later section on the conditions that
lead to scale economies being exploited rather than as a novel concept of scale
effects themselves.
But other authors who are in the Young tradition take a different view. Given
that they are writing at much later dates, it is perhaps surprising that many in
this tradition do not give a clear definition of what they mean by increasing
returns, apparently, as with Young, taking its meaning as self-evident. But it is
4
For example, ‘Definition: a technology exhibits increasing returns to scale if a proportionate
increase in all inputs allows for a more than proportionate increase in outputs; in a single-output
case, this implies a decreasing average cost curve’ (Vassilakas 1987: 4625).
clear that they do not mean the definition of IRTS given above as they stress
changes in both firms and products while rejecting any use of a production
function with highly aggregated inputs. For example, Chandra and Sandilands
write (2006: 201):
At the macro level, supply of inputs is not the main driving force in Young’s
conception of increasing returns; rather it is the size of the market, and the
resulting greater specialisation at firm and industry levels. In this perspec-
tive, the concept of an aggregate production function suffers from a fallacy of
composition: it does not depict the social picture, which may be much
broader than the simple addition of its parts.
Although neither Young, nor Chandra and Sandilands (2006, 2009), nor Grieve
in his reply to them find it necessary to give any clear definition of what they
understand by increasing returns, Grieve (2010:128) does extract a clear
definition of EoS from the Chandra and Sandilands paper. This is ‘any cost
reductions experienced as a firm, responding to increased demand for output,
moves along a given, downward-sloping long run cost curve’. Although not
unlike the standard definition that we use here, this one covers too much as EoS
refers only to the slope of the long-run cost curve and not to where the firm is
currently located on it, nor to why the firm might move from one point on it to
another.
It would appear from the above, and other related, statements that Chandra
and Sandilands, and others in the same tradition, see EoS as a static concept
involving a given long-run cost curve and increasing returns as a dynamic
concept involving shifts in that curve, shifts that are associated with economic
growth in general and increases in the size of the market in particular. We
should also note that one of the causes of the many confusions in the discus-
sions of scale effects in the literature is the failure of economists studying
evolutionary behaviour of growing economies to distinguish their concept of
returns from the static concept found in the majority of the modern literature on
scale effects.5
Although the reciprocal relation between growth and the exploitation of
scale effects is clear (see Section 5), it is not clear that any new, largely implicit,
dynamic definition of increasing returns is needed to study it. The standard
theory of the sources of scale effects as covered in this Element seems to
5
Chandra and Sandilands (2006: 200–202)make a big point about EoS not being important for
increasing returns as they understand the concept. This is fair enough since for them one is a
static concept and the other is a dynamic one. But IRTS as defined both here, and in the literature
on scale effects that we have surveyed, is closely related to EoS because with input prices
constant, the existence of the former implies the existence of the latter.
provide sufficient tools for studying the scale effects of growth in markets that
result from growth in either part or all of the economy.
In this context Setterfield (2001: 489) is the only author in our survey who
overlaps with this dynamic approach by distinguishing between what he terms
static and dynamic scale effects. He defines static IRTS as increasing returns
from an increase in the scale of production at a point in time, and dynamic IRTS
as the technological and/or organisational transformation of the production
process over time as the scale of production increases. Setterfield’s examples of
static IRTS are spatial relationships such as that between surface area and
volume, while his examples of what he calls dynamic IRTS include capital
and labour specialisation and process indivisibilities in the context of demand-
driven expansion. But each of these latter effects can be studied in a static
context. Both can be defined at a point in time as alternative locations on a long-
run cost curve, while to move from one such point to another clearly takes place
over time and will require some incentive such as an expanded market. So once
again a new dynamic definition of scale effects does not seem to be required.
1.4.4 Replication and Reconfiguration
The plant with all of its capital equipment is the main input held constant in the
short run and thus one of the main locations of scale effects over the long run. If
a firm’s scale of operations is to be increased, the capital goods that deliver the
needed capital services must be altered. The firm is then choosing from new
sets of capital goods, all of which embody known technological knowledge.
Two types of choices are possible. Either the firm’s capital may be replicated,
which means creating more units identical to those already in use, or it may be
reconfigured, which means using differently designed capital goods. For exam-
ple, when an airline buys more units of a 150-seat aircraft already in their fleet,
this is replication; when it replaces its fleet of 150-seat aircraft with an off-the-
shelf purchase of 300-seat aircraft, this is reconfiguration.
If an airline reconfigures its aircraft and related capital goods in the long run
in order to increase the scale of its output, the new capital may be more efficient
than the original at delivering its service input to the firm for two distinct
reasons.
First, consider the production of the aircraft. Typically with today’s technol-
ogy, the 300-seat aircraft will have lower costs per passenger-kilometre than the
150-seat aircraft. The reasons for this are investigated at length later in the
Element under the heading ‘Efficiencies of Design’ (Section 1.5). It should be
obvious, however, that we cannot describe the differences in the production of
the two aircraft as being simple multiples of the physical inputs used in the
aircraft manufacturer’s given production function. Each aircraft will have its own
distinct production process that uses some different and some identical physical
inputs the latter in altered proportions.
Second, consider the airline that goes over from a 150- to 300-seat aircraft. Major
structural readjustments in its production process may not be needed, although even
in this simple case, all of the firm’s inputs will not have to be increased in
proportion. The same computer-driven reservation system and the same-sized
cockpit crew may do the job, and, while the cabin crew will have to be increased,
their number will typically not need to be doubled. Moreover, the economies inherit
in larger aircraft will mean that the fuel used will rise by less than in proportion to
the number of passengers carried, and so on.
More generally, when the firm increases its scale of operations, increasing its
output of the number of passenger miles carried, we cannot assume that the types
of capital goods and labour whose services it uses will be nothing more than
variations in the quantities in an unchanged set of generic inputs to its production
function. Instead as scale increases, more specialisation of both labour and capital
goods may occur, requiring different types of labour, capital goods and inter-
mediate inputs of the nuts-and-bolts variety. Often new jobs and new types of
capital will be needed for activities that were not required at a lower and less
specialised production process. In summary, the firm is adopting a new produc-
tion process requiring some new inputs, some different amounts of the existing
inputs and possibly a wholly new organisation of its production processes.
So, each point on the firm’s LRAC curve is defined by a different relation
between inputs and output with at least some different inputs of various types of
the services of both capital and variable factors and in which those inputs that are
used in more than one of these production processes often have different marginal
products in each.
1.4.5 Two Types of Production Functions

The above discussion raises an important question: How are we to describe the
relation between inputs and output in long-run situations when techniques of
production typically vary greatly from one scale of operation to another? To deal
with this question we distinguish two different definitions of the production func-
tion (hereafter PF).
Two PF Definitions
Definition 1: The type-1 PF gives the maximum output that can be produced by
each given bundle of inputs on the assumption that the firm is using a specific
technology of production.
In this case, the inputs can be defined in terms of actual physical quantities, as
they almost always are in the literature. Of course, some level of aggregation is
necessary to prevent the PF having thousands of inputs, but they can be physical
units such as hours worked by various types of labour, units of energy, raw
materials, semi-finished goods and the services of various types of machines
and structures. It follows that no single, type-1 PF can apply to different long
runs where different production techniques require different physical inputs used
in different ways and in different proportions. Instead, as the scale of output rises
progressively, there will be a series of type-1 PFs that are consistent with Viner’s
original analysis. Although he dealt exclusively with cost curves and not with PFs,
each of his short-run cost curves refers to a plant of a different scale.6 As discussed
earlier, the primitives here are the short-run curves, each one referring to a
different plant often using different equipment, types of labour and intermediate
inputs. Although he does not use the term, this implies that the short-run cost
curves are derived from a series of plants each having a different PF. The envelope
long-run cost curve is derived by the exercise of finding the most efficient short-
run curve to adopt for each given output in the long run.
Definition 2: The type-2 PF gives the maximum output that can be produced
by each given bundle of inputs on the assumption that the firm can utilise any
technology of production that is currently available.
As argued above, when the firm changes the scale of its output in the long run, it
is altering its ‘plant’, which often means employing different technologies of
production such as machines that have different capacities and do different jobs,
for example, moving from a craft-style to a mass-production-style production
process. From the point of view of the Viner approach, Definition 2 gives a meta-
PF that covers all these different production techniques in one function. Now the
LRAC curve is not derived from the short-run curves. Instead, given input prices,
it can be derived directly from the PF.
Some of the theorists whom we have studied explicitly make the point that
different production processes are used at different scales of production.7
6
Viner (1931: 205–6) lists three reasons why costs may vary over the long run: operating a given
plant with a different intensity of use, changes in the scale of plant and replication of identical
plants. It is not clear how the first reason differs from the short run situation of varying intensity
of use of a fixed amount of capital, while the third is to be considered later in this Element. This
leaves the second, using different plants having different scales of operation.
7
Baumol (1977: 290), Blaug (1978: 397), DeSerpa (1985: 207), Eaton et al. ( 2012: 185–6),
Kamerschen and Valentine (1977: 232–3), Mansfield (1979: 182), Shone (1981: 158) and Sher
and Pinola (1986: 96) all say that at different scales of output the firm will use different
production techniques. Mahanty (1980: 229) does say that each different short run curve is
derived from a different production function, but he derives each of these from a single type-2 PF
using a different amount of an input that is fixed in the short run and using the same variable
inputs each time.
Others do so implicitly without saying it in so many words.8 Yet others just

relate the LRAC curve to a single PF and do not say explicitly what changes are
occurring between different points on that curve.9 In all cases, however, the
authors use a single PF to describe the firm’s long-term behaviour. Varian alone
among all of those we have studied seems to recognise that when different
techniques are used, his one-production-function analysis cannot be applied.
He writes that when subdividing inputs (1992: 15, italics added): ‘[T]here may
be some minimal scale of operation so that producing output below this scale
involves different techniques. Once the minimal scale of operation is reached,
larger levels of output can be produced by replication.’ He does not analyse
what happens below this minimal scale that is reached by decreasing the
number of possible replicated plants until only one remains. But this is where
the LRAC curve takes on a negative slope because scale effects operate when
different techniques are embodied in plants with different optimal scales of
output.
Morroni (1992: especially 1–3 and 9–11) gives a discussion, with many
references, of why a single production function is a poor tool for analysing firm
behaviour when its scale of operations is altered. While accepting much of the
force of his argument, I follow here the main line of the literature in treating
scale effects within the confines of production function analysis, either PF-1 or
PF-2. I do this for two reasons. First, in order to critique the treatment of scale
effects in the literature, I must follow all those writers who express them in
terms of the characteristics of a production function. Second, I believe that
what matters for this Element, but not necessarily for other lines of investiga-
tion related to economic growth, can be studied using production functions.
Variables in the PF
All writers who are explicit about the variables in the PF state that these are
actual physical quantities. Since this is an important point, we illustrate with a
few examples of what writers say when they are explicit about specific inputs in
the PF (italics added to all quotations): ‘The level of skill of workers (as
compared with their pay), the quality and price of the materials used, the
type of machine, and the ability of management’ (Ammer and Ammer 1984:
368–9); ‘labour, capital, raw materials, etc.’ (Bannock et al. 2003: 311); ‘The
output might be kilowatt hours of electricity per year; [input 1] might be tons of
coal per year; and [input 2] might be maintenance hours per year’ (Smith
8
Nicholson (1979: 188) and Quirk (1987: 181) seem to imply this without stating it explicitly
9
Binger and Hoffman (1988: 258–60), Griffiths and Wall (2000: 163–73), Hirshleifer et al. (2005:
177) and Miller (1978: 200) say nothing about what techniques are being used at different scales
of production.
1968: 512); ‘A mining firm, in order to extract ore, has to employ land . . .
together with labor, buildings, electric power, gasoline, and machines. The
technological relation between such inputs and firm’s output is called the
production function’ (Hirshleifer et al. 2005: 340). More generally, Varian
(1992: 4) Kamerschen and Valentine (1977: 184) and Jehle and Reny (2001:
127) define the PF as an array of physical inputs and a single output and then
leave it at that, while Binger and Hoffman (1988: 231, 258), DeSerpa (1985:
168), Mahanty (1980: 146, 164) and Nicholson (1979: 131–2) define it as
including specific quantities but then use the aggregated capital labour, K-L,
formulation for their detailed analysis. Slightly differently, Shone (1981: 105–
8) uses the general quantity form until he gets to the scale discussion where he
switches without explanation to the K-L formulation. Finally, Griffiths and
Wall (2000: 151), Quirk (1987: 144) and Sher and Pinola (1986: 5) all use the
K-L formulation from the outset.10
There are three serious problems with this common view that the variables in
the PF are real physical inputs. The first relates to the assumption of the
substitutability of inputs in most production function analysis. Many, in some
cases most, inputs at this micro level are complementary to each other and need
to be used in approximately fixed proportions. Thus, although it is possible to
think of substituting between labour and capital defined at a high level of
aggregation, it makes little sense to consider substitution between such specific
physical inputs as needles and thread. Morroni, who discusses this issue of
complementarity among inputs in some detail, observes: ‘It is hard to conceive
of a form of textile production in which yarn could be replaced by machine-hours
or man-hours’ (1992: 29). Probably for this reason, most who use production
functions for specific analytical studies work at high levels of aggregation, such
as labour, capital and materials, where substitution among these inputs can be
considered as a common case. But this is not what they say they are going to do
when they define the inputs as specific physical entities as noted above.
The second problem is that, as already observed, the variables in use alter in
myriad ways from one scale of operation to another. For example, the vast
majority of the tools used in a labour-intensive, craft form of production are
10
All authors are clear that the production function relates physical inputs to physical output so
when they use two aggregate inputs such as capital and labour, this must be assumed to be for
illustration only and that a solution exists to the old capital controversy as to what is meant by
different quantities of capital when they are bundles of very different physical entities. The only
author to make explicit reference to this problem in the present context is Mansfield who writes
(1979: 145), ‘Another important problem is the measurement of capital input. The principle
difficulty stems from the fact that the stock of capital is composed of various types and ages of
machines, buildings and inventory. Combining them into a single measure . . . is a formidable
problem.’
different from those used in Fordist mass-production factory, which in turn

differ from those used in a modern roboticised factory. So are the needles and
finger guards used in a textile factory using foot-operated sewing machines
different from their counterparts used in a mechanised form of production of
the same product. If we are to avoid the PF having a vast number of inputs, most
of which are unused at most scales of production, we must again define the
inputs as quantity-index numbers for broad categories, such as capital (K),
labour (L) and, when needed, materials (M) – the latter can be important as
many modern technologies are much less resource using per unit of output at
large scales than at lower scales of activity. The weighting of individual items
in each such index must use their prices. Although at this level of aggregation
we have macro not microeconomics, it is possible to divide labour into sub-
groups, such as skilled, semi-skilled and unskilled, and subdivide capital in
various ways. However, the definition of each input must not be such as to
include the specific physical items listed by the writers mentioned in the
previous paragraph; they must instead be flexible enough to include these
items in broad generic groups.
The third and more serious problem related to the inputs in the PF being real
physical units concerns the measurement of scale effects. The problem arises
only with the type-2 PF, which covers various scales of output in one function.
As noted above, returns to scale are typically measured by multiplying all
inputs by some constant λ and seeing if the output changes by a multiple that is
more than, less than or the same as λ. But this operation makes no sense when
there is a detailed list of the physical inputs used at any one scale of operations
since, as already noted, when the scale of output changes significantly, firms
typically adopt new production techniques that often use new tools, new kinds
of labour and new produced inputs. So once again we see the need to use broad
quantity indexes such as total labour and total capital.
But even this aggregate usage does not avoid all of the difficulties associated
with the definition of returns effects. Even in the simplest two-input case that
uses indexes of capital and labour, the most efficient capital/labour ratio typically
changes as the scale of output changes with input prices constant. Thus, the equi-
proportionate multiple experiments can be misleading. Doubling the set of
specific physical inputs that many writers refer to is in most cases a nonsense
operation that no engineer would contemplate. Doubling both of the generic
inputs, L and K, that are used at any scale of output may cause output to less than
double (if a larger plant is created using these input ratios) or just double (if
exiting plant is replicated), while varying the input ratios at a doubled total cost
may arrive at an optimum for that cost that more than doubles output – as, for
example, when the firm moves from craft to mass-production techniques.
This critique of the returns to scale definition reveals what should be

obvious: we cannot tell if the firm will encounter favourable scale effects
from knowledge of the PF alone. We need prices of the inputs that allow us
to maximise by calculating the combination of inputs that either minimises total
cost for any given output or maximises output for any given total cost. There is
one exception: if all expansion paths for any given set of relative input prices
are linear through the origin, the optimum proportions in which to combine the
inputs do not vary with the scale of output. In this case the returns definition
will correctly identify the existence of IRTS, CRTS and DRTS. Here, however,
we have something that has little empirical relevance in a world of ubiquitous
non-constant scale effects related to capital goods with differentiated parts. In
all such cases, the optimum ratio in which to use the aggregate inputs will vary
with the scale of output and the returns definition will not necessarily identify
the existence of favourable or unfavourable scale effects arising from the nature
of the PF.
Returns and Efficiency Effect’s in the PF

If everything that confers costs on the firm is included as inputs in the PF, and if
the prices of these inputs are given, we do not have two independent sources of
scale effects, one arising from the PF and one from the cost function. To
determine scale effects from the PF, we need to know the input prices so as
to determine the least-cost way of producing any given output, and when this is
done for all scales of output, we have the cost function.11 But this relation
depends on (1) all things that affect a firm’s costs at all scales of output being
included in its PF and (2) input prices being constant. If things such as selling
costs are not included in the PF, a firm may increase its output by the proportion
λ and encounter CRTS but EoS if input quantities increase by λ while selling
11
In certain circumstances the duality theorem allows one to ‘recover’ a production function from a
given cost function. This is not the place to go into the details of this procedure but suffice it to say
(1) that input prices need to be constant, which, as we will soon see, is not necessarily the case when
the scale effects are located upstream in capital goods producing firms and (2) that problems arise if
firms are price setters, rather than price takers, as are all firms selling differentiated goods and
services. On the latter issue, Jorgenson (2008: 671) states: ‘Under increasing returns and compe-
titive markets for output and all inputs, producer equilibrium is not defined by profit maximization,
since no maximum of profit exists [under perfectly competitive conditions]. However, in regulated
industries the price of output is set by regulatory authority. Given demand for output as a function of
the regulated price, the level of output is exogenous to the producing unit.’ He conducts his
subsequent duality analysis under IRTS with this assumption. Alternatively it can be assumed that
firms all charge the same constant markup over all units of production and all time periods, and then
do the calculation for every firm independent of any aggregation – also very restrictive conditions.
Neither of these special situations covers the cases that are of prime interest here: price setting
monopolists, oligopolists or monopolistic competitors that face EoS that cannot be fully exploited
because of their current equilibrium positions.
costs increase by less. Also, if an increase in a firm’s size increases its market
power over its input suppliers, its input prices can fall, creating an EoS with no
change in the relation been physical inputs and output. More importantly, as we
observed above and will illustrate more fully under the subheading ‘Scale
Effects in Two-Stage Production’ at the beginning of Section 2, real scale
effects achieved by a capital goods producer may be transmitted to the final
goods producer in the form of pecuniary effects.
As we will see later in this Element, when authors discuss the sources of
favourable scale effects, such as more efficient capital or greater division of
labour, they disagree in almost all cases as to whether the source causes an
IRTS or an EoS. This should not surprise us in the light of our arguments both
that the standard definition of returns to scale is deficient and that production
actually takes place in multiple stages.
Indeed, it is not obvious that there is any gain in distinguishing between these
two concepts. However, if we wish to do so, the following definitions may suffice:
returns to scale occur when there is a change in a firm’s unit cost of production that
would have occurred if all input prices had remained constant, while economies or
diseconomies of scale occur if there are changes in a firm’s unit cost including
those that would not have occurred if input prices had remained constant. Thus,
returns to scale are located in the relation between a firm’s inputs and its output and
these give rise to economies of scale which can also arise from sources such as
input price changes due to changes in market power or upstream scale effects.
For the rest of this study we use the type-1 definition where each different
technique of production has its own PF, each one of which may differ from the
others in the nature of inputs and the marginal rates of substitution among them
at various scales of output. Those who wish to continue with the type-2 PF
definition can think of returns to scale when we speak of efficiencies of design
as defined below.
Further Problems with the PF

We note in passing that there are other serious problems connected with both of
the definitions of a PF. To see these, consider the concept of technical effi-
ciency, an engineering concept. Let there be a single product produced by a
bundle of inputs: (1) A bundle of inputs is being used with technical efficiency
if you could not use that bundle to produce more of that product than it is
currently producing; (2) a good is being produced with technical efficiency if
there is no other way to produce it using less of all inputs that are now being
used. Condition (1) looks at the efficiency with which a specific bundle of
inputs is being used while condition (2) looks at the efficiency with which a
given output is being produced.12
Most definitions of the PF consider only condition (1), for example: ‘The
production function for a firm shows the maximum output that can be produced
with specific levels of inputs, given the available technology’ (McAuliffe 1999:
165). Such definitions implicitly assume that there is no problem with condition
(2). That is they assume that when we define the maximum output that can be
produced with each possible input combination, there is no possibility that any
of these will be technically inefficient in the sense that the indicated output
could be produced with fewer of all inputs. If we accept that such inefficiency is
possible, and knowledge of real technologies shows that this is a possibility in
many cases, then some of the points in input space may not be efficient points
and hence not part of the PF because, although they fulfil condition (1), they
violate condition (2). Or to put it another way, the PF will not span the whole of
the input space. This issue is elaborated in the Appendix B under the heading
‘Alternative Factor Combinations at a Given Output’. But since these compli-
cations do not affect our subsequent analysis (but do raise serious issues
concerning the treatment of the PF in theory textbooks) discussion of them is
confined to the appendix.
1.5 Efficiencies of Design

For the reasons outlined earlier, there will typically be scale effects associated
with the firm’s change of scale because different capital goods and different
production process will be used at different scales of output. In our approach
this means that one type-1 PF is substituted for another type-1 PF.
We now define a new concept called ‘efficiencies of design’ and refer to
EoD, DoD and CoD when the replacement of an original production process by
a new one described by a new type-1 PF leads to a fall, a rise or no change in the
unit cost of output – for example, a fall when we move from any of the PFs that
give rise to the SRATC curves labelled 1 to 4 in Figure 1 to a curve that is one
number higher. Efficiencies of design are similar to efficiencies of scale except
that they explicitly refer to movements from one PF-1 to another caused by a
reconfiguration of the capital goods being used or the organisation of produc-
tion, whereas EoS (1) do not specify what type of production function lies
behind the observed cost changes and (2) may refer to changes in costs not due
12
These concepts do not require prices. In contrast economic efficiency, which does not concern
us at this point, looks at costs and defines output to be efficiently produced if there is no other
input combination that would produce that output at a lower cost.
to design changes such as changes in input prices as result of changes in market

power.
The sources of each of these implied variations in efficiency are the subject
of subsequent sections. Note, however, that there is one major exception to the
need to refer to the design effect, which is when the firm increases its scale of
operations by replicating existing plants. No new design is involved; it is a case
of just more of the same. In all other cases, however, there will typically be
design effects when one PF is substituted for another. If one wishes, one can say
that the resulting changes in unit costs are the result of returns to scale in a type-2
PF. But this is merely a description, not an explanation. We still need to ask why
the type-2 PF has the shape that it does, and that takes us to explanations based on
what we call efficiencies of design, which we consider at length in Section 2.
1.6 Scale Effects and Firm Size

Some further ground clearing is needed because in subsequent treatments we
need to consider firms in equilibrium on all sections of their LRAC curves, those
with negative, positive and zero slopes. Although there are many things that
influence the size of firms, what we need to consider here is the relation between
scale effects and size (as well as some related matters). Indeed, two of the most
important applications of knowledge about scale effects is (i) to contribute to
explaining the size of firms in particular industries and (ii) to identify natural,
scale-induced barriers to entry – an absence of which helps to explain the
prevalence of man-made barriers to entry in many industries.13
1.6.1 Implications of MES Relative to Market Size

As is well known, if a firm’s MES is at a very large output relative to the market
demand at the corresponding market price, there will be room for only one firm
producing at or near its minimum efficient scale – a natural monopoly. If it
occurs at not quite so large an output, there may be room for more than one firm.
In situations of duopoly or oligopoly there is nothing in competitive forces to
prevent the competing firms from being in equilibrium on the negative-sloping
sections of their LRAC curves. If the MES is at a very small output, there will
13
Long ago Jo Bain, one of the pioneers of Industrial Organisation, measured scale economies in a
large sample of firms by estimating the proportion of the market supplied by a single plant
operating at its MES and the rise in unit cost at an output equal to half that at its MES. Industries
with ‘very important’ scale economies at that time were automobiles and typewriters and those
with ‘moderately important’ ones were cement, farm machines, rayon, steel and tractors. He
found that product differentiation was a major barrier to entry whose source is complex, and
includes heavy advertising (Bain 1956: 142–3). Interestingly, three of the industries with the
highest man-made barriers were among those with the lowest natural scale barriers to entry.
be room for many firms and there may exist something approaching a situation
in which each firm regards itself as a price taker rather than a price setter,
assuming that they sell an undifferentiated product.
Although empirical observations support these generalisations (see, for
example, Chandler 1990), they might seem at odds with Young’s insistence
(1928: 527) that there is a ‘common error of assuming that wherever increasing
returns operate there is necessarily an effective tendency towards monopoly’.
Someone who assumes that scale effects are sufficient for the emergence of
monopoly is clearly wrong. But much evidence shows that large-scale effects
are often associated with industries containing a small number of firms, occa-
sionally one monopolist but more frequently a few oligopolists.14
Eatwell (2008: 140) takes issue with the use of scale effects as an explanation
of the size of firms when he writes:
Marshall himself recognized the incompatibility of the assumption of com-

petition and presence of increasing returns . . . Piero Sraffa . . . exposed the
entire exercise as ill-founded by demonstrating that neither increasing nor
decreasing returns to scale are compatible with the assumption of perfect
competition in the theory of the firm or of the partial-equilibrium industry
supply curve – a result which, although prominently published and debated,
has apparently escaped the notice of those who still draw that bogus U-
shaped cost curve whilst purporting to analyse the equilibrium of the com-
petitive firm.
Of course, there is nothing incompatible with perfect competition for

firms to have a U-shaped LRATC curves; what is incompatible is for a
firm’s current production to be either in the range of unexploited internal
EoS or in the range where it is encountering DoS. In other words, as
textbooks correctly, not ignorantly, point out, every profit-maximising
firm that is a price taker in a free market must be producing at the minimum
point on its LRAC, whether U-shaped of otherwise (a point that is usually
assumed to be unique for analytical convenience, even if not always for
descriptive realism). However, as long as a firm is a price setter, not a price
taker (as, for example, is any firm selling a differentiated product) there is
nothing inconsistent with its operating on the negatively sloped portion of
its LRAC curve – i.e. having unexploited internal economies of scale.
Chamberlin’s tangency solution is but one example, while price setting
oligopolists in industries with either natural or man-made barriers to
14
Nothing said here contradicts Schumpeter’s point that the resulting high profits will encourage
other firms to find ways to make innovations that provide end runs around the position(s) of the
established firm(s), for example, by inventing close substitutes that will erode the profits of
incumbents.
Unit Cost LRAC
c1
0 q1 2q1 3q1 4q1 Quantity
Figure 2 Variable scale effect with replication
entry, provide many others. In what follows, we assume that firms can be in
equilibrium with such unexploited scale economies.
1.6.2 Replication with Variable Unit Costs
Once the MES of a single plant has been reached and where replication of
plants is possible, a firm’s unit costs are usually assumed to be constant.15 But
this is so only for integer multiples of the MES of that plant. Now consider
desired outputs that are non-integer multiples of q1 as illustrated in Figure 2.
Although it is hard to avoid language that sounds as if a movement over time is
being considered, the analysis is of alternates for a single choice of plants to
produce a given output. To produce any output greater than q1 and less than
2q1 the firm has two choices. It can use one plant whose associated unit cost
exceeds c1 or it can use two plants, one producing at its MES of q1 and the
other smaller plant producing the extra desired amount in excess of q1. We
refer to these respectively as the large-plant and the two-plant solutions.
For small increases in output above q1 the large-plant solution will be the
superior choice. For example, to build the minimum sized plant that would
produce 1 unit when the desired output is q1+1 and underutilize it would result
in a very high unit cost of that output. In contrast, for a large increase above
q1 the two-plant solution will be the superior choice. For example, if the
15
For example, Frevert (1997: 1374): ‘Since any physical production process can be duplicated,
all production should adhere to constant returns to scale.’
desired output is 2q1–1, the unit cost of the second plant will be only
slightly above c1 while the unit cost of the larger plant will be well above
that figure.
One might think that the changeover from the large-plant to the two-plant
solution would be when the unit cost of the large plant equaled the unit cost of
the smaller plant under the two-plant solution. But this must occur at a lesser
output because with the larger plant the cost of all units of output are above c1
while with the two-plant solution only the units produced by the smaller of the
two plants will incur a cost of over c1. To determine the changeover output let
the total cost of output in the one-plant solution be (q1+Δq)(c1+Δc) and for the
two plant solution be c1q1 +csΔq, where cs is the unit cost of production in the
small plant. Equating these and solving for cs gives:
cs ¼ ½Δcðq1 þ ΔqÞ þ Δqc1 Þ=Δq
For example, let the MES occur at 100 units of output with a unit cost of 10 and
the larger desired output be 110 with a unit cost of 11 when produced by a single
plant. The two-plant solution will then be the superior choice if the unit cost
associated with the smaller plant is less than 21.
Assuming that the large-plant solution is the superior choice for some range
of output starting at q1, the LRAC curve takes the shape shown in Figure 2
between q1 and 2q1. The kink occurs at the changeover where the rising cost
associated with the large-plant solution is replaced by the falling cost asso-
ciated with the two-plant solution. More generally, if we consider non-integer
production between nq1 and (n+1)q1, the choice between the two solutions is
the same at the margin for any n, with (i) the two-plant solution applying to the
nth plant producing at q1 units and a smaller plant and (ii) the large-plant
solution with one plant producing an output between nq1 and (n+1)q1. But
since the firm’s total output produced by all the plants operating at their MESs
is larger the larger is n, the average cost in the interval between nq1 and (n+1)q1
is lower the larger is n. Hence the kinked LRAC curve between the each of the
multiples of q1 gets flatter and flatter as n increases, until in the limit it
approaches a straight line joining the two extreme MES points.
If output of the plant that produces q1 at its MES is large relative to the
market demand at the corresponding price, the range over which the firm will
build a plant that has a higher unit cost than c1 can also be large. This challenges
the a priori argument found so often in the literature that firms will never
encounter increasing costs if replication is possible. Here there are no hidden
inputs and although replication is possible, it is not efficient over some ranges
of output.
Almost all of the surveyed authors mention replication as a sources of CoS

and/or CRTS. Varian is among the few authors in our survey who directly deal
with the fact that the replication argument for constant return to scale applies
only to integer multiples of the output at the MES of the most efficient size of
plant. He writes (1992: 15): ‘Another circumstance where constant returns to
scale may be violated is when we want to scale operation up by non-integer
amounts . . . how do we do one and half times what we were doing before?’ He
does not answer this question except to observe that ‘such cases are only
important when the scale of production is small relative to the minimum scale
of output’. This will not be the case for the price taking firms that are the usual
case in the micro textbooks, but it is the common case in manufacturing and
many service industries where firms are price setters and the number of firms
in the industry is small.
1.6.3 Replication with Non-Plant-Specific Fixed Costs
Another interesting case arises with costs that are specific to the firm but do
not vary with either the firm’s total output or its number of plants. The
literature often mentions management functions that may not need to be
increased in proportion to the increases in output when new plants are
added (Griffiths & Wall 2000: 172; McConnell et al. 2012: 154). But these
are usually sources of relatively small savings. We deal with cases where this
effect is large in Section 3. So all we need to observe here is that such costs
confer a falling average total cost even when the output is varied by replicat-
ing existing plants. So, if in the absence of such costs the firm would
encounter constant average cost for outputs that are integer multiples of the
optimum-size plant, increasing output by increasing the number of plants
causes average total cost to be declining. So the firm’s size is not limited by
technical factors but will instead be limited by total demand if it is a monopo-
list, or competition with other similar firms if it has competitors.
2 The Sources of Positive Scale Effects

We now consider the major sources of positive scale effects, both those that
have been alleged in the literature and those that our analysis suggests to be
important. We also discuss a few other minor sources in order to consider all of
the sources listed by at least one of the many authors that we have surveyed.
2.1 Scale Effects in Two-Stage Production

Scale effects, both favourable and unfavourable, are typically divided in the
literature between those that are internal and those that are external to the firm.
Since all but one of the works cited in this Element study a single firm that can be
assumed to be a producer of some final good, scale effects that are external to that
firm but internal to a firm producing the capital goods that it uses are typically
classified as being internal to the final goods producer. Only one work on our
survey considered scale effects in a model that exploits the fact, fundamental to
early capital theorists, that capitalist production is two-stage production with a
capital good being made in stage one and then used to assist in producing some final
good in the second stage. We elaborate on the two-stage analysis of Lipsey, Carlaw
and Bekar (2005: 393) which, although it contains an important flaw, does illustrate
some key points about scale effects in a two-stage production process.
Let there be a firm that is in the business of pasturing other people’s horses.
One square unit of fenced space, the capital good, is required for the accom-
modation of one horse, the final service output being one horse pastured. The
grass is free and the only production cost of final output is the service of the
fenced pasture, which is continuously variable in its production. When the firm
wishes to provide pasture for more horses, it orders an increase in the size of its
fenced field.
The final output of horses pastured depends on the area of the pasture in
a linear homogeneous function where one square unit of pasture, P, is
sufficient to maintain one horse, H.
H¼P (II.1)
The typical treatment found in the literature models only the PF for final goods
as in (II.1) above and in this case there are no apparent scale effects.
Now consider the PF for the capital good, P, which is created by enclosing
land by F feet of fence according to the following PF:
P ¼ ð F=4Þ2 (II.2)
Here we see scale effects:

dP=dF ¼ F=2 (II.3)
which is increasing in F. To check, multiply F in equation (II.2) by λ to get:
P ¼ ðλF=4Þ2 ¼ λ2 ðF=4Þ2 : (II.4)
Assume that the fence is built in period 1 and lasts z years after construction,
covering periods 2 to z +1, and costs $1 per foot to build with no maintenance
costs until it expires. (Thus the returns and economies concepts can be treated
interchangeably in the capital goods industry.) Assume that pasture is rented
out for equal yearly payments over that period. The supplying firm wishes to
make a return of r% on its investment. So it must charge a price, y per unit of

pasture, such that the discounted present value of the stream of payments equals
the production cost plus a desired return on the investment in the fence, F(1 +
r). Assuming the desired rate of profit is the same as the discount rate, the price
that must be charged per unit of pasture is:
ð1 þ rÞF Xzþ1 y
¼ (II.5)
ðF=4Þ 2 t¼2 ð1 þ rÞt
or
16ð1 þ rÞ Xzþ1 1
¼ y t¼2 (II.6)
F ð1 þ r Þt
Letting the constant summation term be C, we have

16ð1 þ rÞ
y¼ (II.7)
CF
which is decreasing in F. So, the price charged per unit of pasture falls as
pasture size, and hence the number of horses pastured, is increased. Note that
the capital goods industry has IRTS while the final goods industry has CRTS.
The two taken together produce EoS for the final goods producer.
Although a special case, this example has some general implications. Scale
effects can occur even when the relevant PF for the final good displays constant
returns to scale. In the present case, there is an IRTS from the point view of the
capital goods industry, while from the point of view of the final goods producer
these real changes become an EoS in the form of a reduction in the cost of its
capital service input. Thus one must look to the first stage, capital goods industry,
in a two-stage production process to understand the scale effects that have their
source in the nature of capital goods. From the economy’s point of view there is a
real resource-saving scale economy, although it shows up as a reduction in the
cost of an input to the final goods producer. It is difficult (impossible?) to
compress this behaviour into a single PF for the final goods producer who
encounters constant returns in terms of its own physical inputs. (Indeed Lipsey,
Carlaw and Bekar (2005) tried to do this but erred by substituting the PF for the
capital good into the PF for the final good, thereby implicitly assuming that the
capital good must be replaced each period.) The difficulties in studying two-stage
scale effect in a treatment that uses only one PF possibly accounts for the large
amount of disagreement that we chronicle below in designating sources of scale
effects as causing either returns or efficiency effects.
2.2 Pervasiveness of Scale Effects

A general proof of the pervasiveness of scale effects with respect to all types of
capital goods was provided some time ago by Eaton and Lipsey (1977). Here is an
intuitive statement of their formal argument.
Definition: The amount of services embodied in a capital good is a
measure of its ‘lumpiness’. Since the amount of embodied services varies
directly with durability, endogenous lumpiness is created when capital goods
are produced to be more durable than is absolutely necessary for them to do
some specific job.
Assumptions: (1) The interest rate is positive. (2) A decision must be made
on the amount of durability to build into any capital good, whether the mini-
mum necessary or some larger amount. (3) The technology of building capital
goods displays constant CoS effects in the sense that the reconfiguration of a
capital good to alter its durability so that it embodies λ more or less capital
services implies that the cost of producing that capital good changes in the same
proportion, λ.
Implication: Interest costs are minimised by minimising the capital good’s
durability.
Empirical observation: It is a matter of simple observation that virtually all
capital goods are made more durable than they could be, a phenomenon that
Eaton and Lipsey call endogenous temporal lumpiness or indivisibility.
Contradiction: The empirical observation of endogenous lumpiness is
inconsistent with the implication drawn from the three basic assumptions:
unit cost of delivering a capital good’s services is minimised by minimising
durability.
Conclusion: Since assumption (1) is known to be true and assumption (2)
may be taken to be to verified by observation of the great majority of capital
goods (we know of no exceptions), the conclusion is that assumption (3) must
be false. This assumption is then altered as follows: There is a universal scale
effect in embodying services in capital goods: as durability of the capital good
is increased, there is some range, starting from the minimum necessary for the
good to do it job, over which the services that it embodies rise faster than the
cost of adding to the good’s durability.
The ubiquity of scale effects related to capital goods is shown by this proof
based on the existence of endogenously determined lumpiness or durability.
More durability usually implies more strength, different handling characteris-
tics, more ability to deal with random shocks and many other sources of scale
effects related to both statistical and natural laws that are considered in a later
section. The one place in which the present treatment is in disagreement with
the Eaton-Lipsey paper is in their treatment of lumpiness and indivisibility as

identical. We return to this issue below where we discuss the various meanings
that are given in the literature to the concept of indivisibility.
2.3 Scale Effects due to Indivisibilities

The concept of indivisibilities plays a large role in the literature of scale effects.
To make the concept more precise we outline several distinct but related types
of divisibility and indivisibility. As we will show later, each one of these has
been used by one or more of our authors to explain scale effects, although we
argue that most of these do not do so. Furthermore, some authors cite more than
one type, and by treating them all as the same indivisibility, they cause confu-
sion. We confine ourselves to capital goods, such as individual machines and
whole plants, although much of what we say applies equally to other types of
goods, such as consumers’ durables.
We distinguish two basic meanings of indivisibility, ex post and ex ante, and
some variations of each version.
2.3.1 Ex post Divisibility and Indivisibility

This refers to altering an individual capital good or plant once it has been
produced. A good or plant is ex post divisible if, once produced, it can be
subdivided such that its parts can do the same type of job as can be done by the
whole. For example, a one-ton bag of wheat, once produced, can be subdi-
vided into two half-ton bags, each one of which can be used to feed half as
many people as the one-ton bag.16 A good is ex post indivisible if the parts
cannot do this. For example, two halves of an airplane cannot carry any
passengers. All goods that have differentiated parts and do some specific
job or set of jobs are ex post completely indivisible in the sense that a part of
them cannot do any amount of the job that can be done by the whole.17
A related ex post indivisibility refers to the integer problem: many consu-
mers’ and producers’ goods are only available in integer amounts. This may be
for physical reasons. For example, a consumer can be the sole owner of one,
two or three cars, but not two and a half; a producer can install one, two or three
16
Morroni (1992: 26) distinguishes between technical indivisibility, which we are discussing here,
and economic indivisibility in which a commodity can only be purchased in some minimum size
or amount, the latter being analytically similar to the integer problem discussed in the next text
paragraph.
17
Although a person is ex post physically indivisible, the existence of part-time hiring makes one
person divisible as far as a firm is concerned. Although the same is true for some types of capital
goods and some firms, for the most part a firm cannot hire part time a factory or single parts of an
integrated production facility.
laths, but not two and a half. It may also be for economic or customary reasons
in that commodities that are physically divisible can only be purchased in
discreet amounts, such as a bag of flour. Handling such issues in consumer
and producer theory, as opposed to the usual assumption that all such items can
be varied continuously, poses some formidable technical problems but does not
affect anything that is at issue here. (For a full treatment and an extensive
bibliography, see Bobzin (1998).)
2.3.2 Ex ante Divisibility and Indivisibility
This refers to altering the size of a capital good or whole plant, making a new
one that is larger or smaller than the original one but that can do the same type
of job. Size is a multi-dimensional concept but the essence of what is at issue
can be seen by defining size in terms of capacity to do some job. For example, a
truck that can carry two tons of bulk cargo is smaller than a truck that can carry
four tons, but larger than one that can carry only one ton. A drill press that can
drill holes in a one-inch-thick piece sheet metal at a rate of one hole per
10 seconds is smaller than one that can do the job in 5 seconds, and larger
than one that takes 20 seconds to do the job. Also a factory that can produce
some product at a rate of 100 items per day is larger than one that can produce
the same product at a rate of 50 per day and smaller than one that can produce at
a rate of 200 per day.
An Indivisible Plant
Two distinct versions of ex ante plant divisibility and plant indivisibility are
found in the literature. In the first, a production process, plant in our terminol-
ogy, is divisible if it can be scaled upward or downward by multiplying all of
the inputs in its production function by some positive constant, λ, and have its
output change in the same proportion. The process is defined to be indivisible in
the downward direction if the result is to alter output by some multiple, γ, where
0 ≤ γ < λ. This, of course, is how constant returns and decreasing returns to scale
are normally defined. Note that all of the plants that give rise to the cost curves
shown in Figure 1 are indivisible in that sense. If they were not, there would be
no need to accept a higher unit cost than c1 by scaling downwards the plant that
has the cost curve shown as SRATC5. Analogous comments apply to each
smaller sized plant down to the one with the cost curve SRATC1. We call these
ncrs-indivisibilities for non-constant returns to scale indivisibilities.
In the second version, a plant is defined as ex ante divisible if a smaller
version can be made to do the same type of job as the larger version. In terms of
Figure 1, this is the case with plants whose short-run cost curves are between
SRATC5 and SRATC2. If the production process that has the short-run cost
curve SRATC1 is the smallest size plant that can produce the product in
question, that production process is ex ante indivisible at that size. We call
this an mps-indivisibility for minimum possible size. Note that all of the plants
shown in the figure are both ex post indivisible and ex ante ncrs-indivisible,
while only the smallest possible version is ex ante mps-indivisible (because no
smaller version can be made) while all larger versions are ex ante divisible
(because smaller versions can be made).18
Two cases in which it is possible to build a plant smaller than the one with
SRATC1 need to be considered. In one case, the plant is technically inefficient
in the sense that it uses more of all inputs for all common production levels than
does the plant with the SRATC1 curve.19 In the other case, although for
common production levels the plant uses fewer of some inputs than used in
SRATC1, it uses so much more of others that the total short-run costs always
exceed those of the larger plant. So, in both cases, the cost curve of the smaller
plant lies wholly above SRATC1, as does the dotted curve SRATC0 in Figure 3
on page 41. So, in both cases the plant that has the short-run cost curve SRATC1
is producing at the lowest technically and/or economically efficient scale. For
subsequent analysis it is convenient to confine ourselves to the case in which
the process that gives rise to SRATC1 is the smallest size that will do the job
technically and economically. However, if either of the other two cases is
possible the analysis is the same because curves such as SRATC0 are irrelevant
and can be ignored.
An Indivisible Capital Good

In the case of a capital good there is no parallel to the ncrs concept of the
indivisibility of a plant. As already observed, reduction of the inputs that go to
18
In the Eaton-Lipsey paper cited earlier the authors describe the services embodied in capital
goods as an endogenous ‘lumpiness’ or ‘indivisibility’. To see the problem it is necessary to
distinguish, as they do not do, between lumpiness and indivisibility. Lumpiness is a variable
measured in the simplest cases by the amount of services embodied in a capital good. As their
argument shows, lumpiness is an endogenously determined variable. In contrast, indivisibility is
a characteristic, not a variable. A thing is either divisible or it is not. The capital goods to which
they refer have various amounts of lumpiness and almost all are ex post indivisible. But they are
not ex ante indivisible since they are being produced with different amounts of capacity. Scale
effects related to the design of capital goods are incentives to embody lumpiness in capital
goods; but they do not cause ex post indivisibility, which is determined by the nature of most
capital goods no matter how many services they do or do not embody. This is not, however, a
serious problem for Eaton and Lipsey’s main argument about endogenous lumpiness since all
references to ‘indivisibilities’ can be replaced by ‘lumpiness’ without affecting their argument.
19
This is doubtless an unusual case but not an inconceivable one because it often takes a lot of
capital to miniaturise various processes and may take no less (or even more) labour to operate
them than their large counterpart.
SRATC0
SRATC1
SRATC2
SRATC3
Unit Cost
SRATC4
SRATC8
SRATC5
SRATC7
SRATC6
c1
0 q1 q2 q3 q4
Quantity
Figure 3 Ex post and ex ante indivisibilities
make capital goods by some constant proportion λ will probably not produce
anything useful. In our three-dimensional world with its physical laws making a
smaller version of some machine never requires scaling all of its inputs down in
proportion. However, the second concept does apply. Note, that almost all capital
goods that do some specific job can be made in different sizes. For example,
Dudley Jackson in his piece cited below notes 288 cases in which machines that do
some given job have been made of indifferent sizes and their relative efficiencies
measured. Undoubtedly, there will be some smallest size of that machine that will
be capable of doing its job. The machine is ex ante indivisible at that size. There
are, however, two further possibilities in this case. First, there may be some other
less specialised machines and labour that will do the same job at a smaller scale but
less efficiently per unit of output. Second, there may be no alternative process to
deliver the needed service and the smallest possible machine must be installed and
underutilised as output falls below what it needed to employ that machine at full
capacity. In this case although the machine itself is subject to an ex ante indivisi-
bility, its services need not be. So the explanation of the higher cost at a lower scale
of output lies not in the service flow being subject to an inequality. In this matter,
there is an ambiguity in the literature in that sometimes the service flow of the
capital good and at other times the good itself is shown in the PF. If it is the service
of the machine that is the variable in the plant’s PF, this input is continually variable
downwards in the short run by operating the machine below its full capacity. Unit
total costs will then rise since the capital cost of the machine does not fall when its
use is reduced and must be spread over fewer units as output falls. So although the
source of the scale effect is on the production side, it is not seen in the plant’s PF
where one of the (continuously variable) inputs is the services of this capital good;
it is only seen in the cost function as some costs stay constant as the machine is
used less intensively. Indeed, it is not easy to think of processes where the services
of some ex ante indivisible capital good cannot be reduced continually either by
leaving the good idle part time or disposing of its services for uses other than in the
production of the good in question. In this case, although the contribution of the
machine’s service to unit variable costs will be constant right down to zero output,
the average total cost will rise as the fixed cost of the indivisible machine is spread
over fewer and fewer units of output.
2.3.3 Which Variable: A Stock or a Flow?
In Section 1 we discussed several problems associated with the specification

of the variables in the production function. The above discussion reveals a
further problem. Most writers when defining a PF formally state that its
variables are the service flows of the various factors of production: for
example, the machine hours used from some specific machine. As already
mentioned, the service flow from some specific machine, or whole plant, can
be reduced continually to zero by under using it, while the capital good itself
cannot be so varied. So if there is an indivisibility it is related to the capital
good not to its service flow. It seems implicit in the writings of those who use
this explanation of scale effects that they are referring to capital goods that
are, in our terminology, ex ante indivisible. Yet although this indivisibility
may explain rising unit cost as output is reduced, it does not explain why the
full services of the capital good must be used, even when production is
reduced (as is implied by the minimum constraint on its use). It would seem
that writers, having specified the capital good’s services as the inputs in the
production function, then slip into using the good itself, which clearly is
subject to a minimum constraint on its size when it reaches the point of ex ante
indivisibility.
2.3.4 Indivisibilities as Sources
If we look at the cost curves for alternative-sized plants shown in Figure 3, we

see economies of scale over the output range from zero to q3 and diseconomies
from q3 at least up to q4. We require an explanation of these different relations,

which implies that we need something that is present over one of these ranges
and absent, or at least behaving differently, over the other range. Can indivisi-
bilities provide such an explanation?
Ex post Indivisibilities
First consider ex post indivisibilities. All capital goods with differentiated
parts, including the plants that give rise to each of the short-run cost curves
SRATC1 to SRATC8 in Figure 3, are ex post indivisible; you cannot cut
them up and expect them to do any job, let alone the job done by the whole.
Thus, this universal characteristic of all capital goods with differentiated
parts cannot explain the different scale effects over different ranges of
output.
Ex ante Indivisibility of a Plant

Now consider the ex ante indivisibility of a whole plant. We first take the case
of ncrs-indivisibilities. Since each of the plants with the cost curves shown in
the figure are ex ante ncrs-indivisible, this universal characteristic of plants that
lie behind the U-shaped LRATC curve cannot explain the varying slope of that
curve. If any of the processes that gave rise to any of the SRATC curves
numbered 1–5 in Figure 1 were divisible in this sense, it would never be
necessary to produce with a process that gave rise to a lower numbered
SRATC curve. The original process could be subdivided to produce a lower
output at an unchanged unit cost. Thus, this type of indivisibility of any
economically relevant processes is a necessary condition for it to show increas-
ing returns to scale as it does between the curves 1–5 in Figure 1. But this type
of indivisibility is not sufficient to explain scale effects that vary over different
ranges of output as do those in Figure 1. Thus, when this type of ubiquitous
process indivisibility is used to explain why there are scale effects in the form
of non-constant returns in some cases but not others, the reasoning is circular: a
process is defined as indivisible if it does not have constant returns to scale and
this indivisibility is used to ‘explain’ the absence of constant returns. Of course,
the argument then begs the question: Why do these non-constant returns occur?
For the explanation, we look to economies of design that are the subject of the
next section.
To consider mps-indivisibilities let SRATC1 in Figure 3 be the short-run cost
curve for the minimum-sized plant that is capable of producing the commodity in
question, the quantity produced at its MES is q2. For simplicity, we assume that
the size of the plant can only be raised above that minimum size in discrete
amounts, as shown by the various SRATC curves in the figure.20 Since there is no
relevant SRATC curve associated with a smaller sized plant, the LRAC curve
follows this short-run SRATC1 curve until it is intersected by the short-run curve
associated with the next largest possible sized plant which occurs at output
quantity q1 in the figure. So, for higher outputs than q1 the LRAC curve follows
the outer segment of each relevant short-run cost curve as larger and larger plants
are utilised to produce larger and larger flows of production.21 The production of
this product is ex ante indivisible at the scale of the plant with SRATC1. So,
outputs from zero to q1 are produced under conditions of falling unit costs
because of this ex ante indivisibility of the plant. This minimum-sized plant
is sufficient to cause economies of scale as it determines the LRAC curve, but
only up to an output of q1, after which this mps-indivisibility is irrelevant as
plants larger than this minimum size are being employed.
Ex ante Indivisibility of an Individual Capital Good

Now consider ex ante mps-indivisibility of an individual capital good. In the
first case that we consider there are other less specialised machines and labour
that can deliver the same service. Although the indivisibility of the machine is a
necessary condition for a switch to a less efficient process to deliver the needed
service, the explanation of the scale effect is in the reasons for the different
efficiencies of the two processes, which lie in economies of design, not in the
indivisibility of the more specialised machine.
In the second case, the machine is indivisible and has no smaller substitutes. For
example, let there be a machine that is used at all scales of production up q4 and let it
be ex ante mps-indivisible at a size whose full employment comes with the output at
the minimum unit cost on the curve SRATC4. Now for lower outputs this machine
will not be fully utilised. So, when output increases from zero to where the machine
is fully employed, there will be downward pressure on unit cost as a result of the
spreading of the machine’s fixed cost. This will be in addition to all of the reasons
for falling unit cost due to economies of design as the larger and larger plants are
built with MESs from q1 to q4.
What if one production process uses several such machines? According to
Morroni (1992: 28), in this case ‘to avoid using individual machines at sub-
optimal levels the overall scale of the production unit must be equal to the lowest
common multiple of the productive capacity of the individual machines’. If there
20
Hirshleifer et al. (2005: 177) mention this case in which the fixed factor can only be varied in
discrete amounts.
21
As noted earlier, if we assume that plant size can be varied by smaller and smaller increments
beyond the plant with cost curve SRATC1, we get, in the limit, the standard smooth LRAC curve
which is the envelop of these tightly packed short run cost curves.
are several such machines, each with its own different capacity, this could be
quite a large number. But Morroni only considers ex post indivisibility and so
does not ask: what if different versions of these machines can be made each with
a different capacity? At the extreme, if all can be made with continually varying
sizes and capacities down to some minimum mps-indivisibility, then this sub-
capacity effect will only be felt below the minimum capacity of the largest
machine. All other machines will be made with sizes that match the desired
capacity of that machine.
Exactly how important this effect of ex ante mps-indivisibility of individual
machines in plants that are larger than the minimum possible plant size compared
to the importance of economies of design is a matter that cannot be decided
without empirical evidence.
2.3.5 Interesting Exceptions
There are some interesting exceptions to the frequent existence of ex ante mps-
indivisibilities of capital goods. These are goods for which there is no relevant finite
minimum size needed to do the type of job in question. Interestingly, this includes
pipe lines that are often quoted as examples of indivisibilities. Of course, as with all
capital goods with differentiated parts, pipe lines are ex post indivisible. But they
are not ex ante indivisible. They can be made as short as needed, depending on the
space over which a liquid is to be transferred, and they can be made with as small a
radius as is needed, depending on the required rate of liquid transfer. Similarly, the
fenced pasture studied in the previous section is not ex ante indivisible. Of course,
like virtually all other capital goods both of these are ex post indivisible, if cut in half
once built, they cannot do their jobs. But they can be made ex ante as small as
required, down to one square inch of fenced pasture and a one foot long narrow
pipe. So, the economies of scale associated with pipe lines, fences, and other similar
capital goods, extend over the whole range of their outputs from zero to an
indefinitely large size and have nothing to do with ex ante indivisibilities, which
are non-existent.
2.3.6 Summary: Indivisibilities as Valid Sources

of Scale Effects
As we have observed, any valid explanation of EoS effects must be able to
explain why these occur only over some range of output, such as is shown in
Figure 1. Ex post indivisibilities cannot do this because they occur at all levels
of output for any capital good with differentiated parts. Neither is the absence
of constant returns to scale an explanation, as explained above in the section ‘ex
ante Indivisibility of a Plant’. It is a necessary but not a sufficient condition for
scale effects. The only valid explanation that we have found is the ex ante mps-
indivisibility in that there is some smallest version of a plant or individual machine
that is necessary to produce some product or capital service. As output varies over
the smallest version from zero to that unit’s capacity, scale effects are encountered.
But as soon as it is economic to use a larger version of that plant or capital good, ex
ante invisibilities are no longer an explanation of what is observed. It follows that
we need to look to economies of design (and a few other minor sources that we
considered later) for an explanation of the majority of the observed scale effects.
2.3.7 Indivisibilities in the Literature
Because of the major difference between the assessments of the importance of

indivisibilities found in the literature and the present treatment, it is important
to survey all of the authors in our list of references who wrote about the
definition of the indivisibilities and then stated that they were sources of
scale effects. Of course, given the various meanings of indivisibility that are
distinguished here, and given that most authors do not make such distinctions,
there is a certain ambiguity in many of these quotations. Nonetheless, over all
they serve to show the variety of different views on how indivisibilities may
cause scale effects – many of which are contradicted by the above analysis.
Ex post Divisibility and Indivisibility

Some authors seem to rely on ex post indivisibility to explain scale effects,
something that we have seen is invalid given that it is a universal characteristic
of all goods with differentiated parts.
I have not found one example of increasing returns to scale in which there is
not some indivisible commodity in the surrounding circumstances. The oft-
quoted case of a pipeline whose diameter is a continuous variable . . .
[requires] one entire pipeline of the requisite length . . . to render the service.
Half the length of line does not carry half the flow of oil. (Koopmans 1957:
152 fn. 3, italics added)
Assume for instance that the only input is some specific capital good (a
machine, plant, ship or pipeline) which is indivisible in the sense that it
becomes useless if physically divided. (Silvestre 1987: 2799)
Frequently, an economy of scale results simply from the fact that a certain
factor of production is indivisible, that is, it cannot be divided into smaller
units. (Ammer & Ammer 1984: 415)
What is being illustrated in Table A.7 is the feature of discontinuous produc-

tion changes arising from indivisibilities. In other words, certain production
processes involving, say, very lumpy capital equipment can only be operated
if the scale is B or C. More to the point it is not possible to halve the
equipment and operate it at a lower scale, because different processes have
different unit levels of operation. This implies that even in the long run
factors are not perfectly divisible. (Shone 1981: 156)
Although somewhat ambiguous, Shone appears to be referring to ex post

indivisibility.
According to Morroni, technical indivisibility ‘refers to the impossibility of
dividing a particular commodity once it is exchanged, into [smaller] amounts
usable for production or consumption.[ex post indivisibility]’ (1992: 26) This is
not to be ‘confused with the possibility of exchanging a given commodity
available in continuously increasing size [ex ante indivisibility]’. (1992: 26)
‘Most fund elements [capital goods] are indivisible.’ (1992: 25) Having expli-
citly ruled out ex ante indivisibility, he gets scale effects by the spreading of
overheads when the capital good is used at less than capacity which as we have
noted is a valid explanation when the good is ex ante mps-indivisible but not
when it is made to any larger size. Thus, his analysis does not explain scale
effects for a firm larger than the one with SRATC1 in Figure 3. It is unclear,
however, that he intends to restrict the influence of indivisibilities to such
minimum-sized plants since he has explicitly defined indivisibilities ex post
and ruled out ex ante mps-indivisibilities.22
Other authors refer to our ex ante indivisibility and then use as an illustration
the ex post indivisibility that applies to all goods with differentiated parts.
Indivisibilities: Some inputs just do not come in small units [ex ante indivi-
sibility?]. We cannot install half a blast furnace or half a locomotive (a small
locomotive is not the same as a fraction of a large locomotive). [ex post
indivisibility] As a result, only if operations are carried out on a sufficiently
large scale will it pay to employ such indivisible units. (Baumol 1977: 274)
‘[I]f a firm doubles its scale, it may be able to use techniques that could not be
used at the smaller scale. Some inputs are not available in small units; for
example, we cannot install half an open hearth furnace [ex post indivisibil-
ity]. Because of indivisibilities of this sort, increasing returns to scale may
occur.’ (Mansfield 1979: 142)
But of course you can have open hearth furnaces of various sizes. Only the
minimum possible size is ex ante indivisible.
22
In another place Morroni seems to define indivisibilities as ex ante ncrs-indivisibilities, ‘[a]
process is indivisible if it is impossible to activate processes that have the same proportions of
inputs and outputs, but on a smaller scale’ (Morroni 1992: 145 note 4).
‘[Indivisibilities are defined as] the minimum physical or technical size

limitations on FACTOR INPUTS. For example, a company wants to pur-
chase a machine that can undertake 5,000 operations a day. Because of
design and technical difficulties, the minimum-size machine available opti-
mally carries out 10,000 operations a day. The machine is indivisible since it
cannot be reduced to two optimal half-machines’ (Pass et al. 2005: 251).
Is not clear what is being referred to here but it seems that the penultimate
sentence refers to ex ante indivisibility and the final one to ex post indivisibility.
Indivisibility, or discontinuity, the technical or physical characteristics of a

factor of production or commodity which prevents its being used except in
minimum quantities [meaning unclear]. Most machinery and capital equip-
ment must be used in ‘lumps’ of minimum size. A workman is also an
‘indivisible’ unit in this respect; the hands used for an assembly line cannot
be separated from the legs that could be simultaneously used for messenger
work [this appears to refer to ex post indivisibilities]. . . . [A]t some scales of
production it may pay a firm to have a market research department or a staff
training school. Where the size of the indivisible unit is large (in relation to
annual output) plants or firms will tend to be large: technical indivisibilities tend
to determine the size of the plant; other, such as financial, the size of the firms
[this appears to be referring to ex ante mps-indivisibilities]. (Seldon &
Pennance 1976: 181)
Yet other authors are unclear if ex ante or ex post indivisibilities are respon-
sible for scale effects.
‘indivisibilities: A production factor that cannot be segregated further into

smaller elements. The input variable is restricted in some way by size’ (Shim
& Siegal 1995: 182).
It is unclear if this refers to ex ante or ex post indivisibility.
‘Indivisibilities: The impossibility of dividing a factor into smaller units’

(Sloman 2006: 132).
Again it is not clear if ex post or ex ante indivisibility is being refereed to here.

Elsewhere Sloman uses R&D as an example of an indivisibility, saying that
‘only a large firm can set up a research laboratory’ He does not, however, deal
with the fact that R&D labs can be of various sizes.
Discreteness
Some of our authors argue that discreteness of an input is a source of scale effects.
The clearest example of this is given by Morroni (1992) who defines economic
indivisibility as occurring when ‘it is impossible to exchange less than a given unit
of some commodity’. This differs from his technical indivisibility because one
can divide the commodity then restore it to its original size as one can divide a 2
kilogram bag of flour into two one kilo bags and then restore the contexts to one 2
kilo bag, which cannot be done by dividing the recombining the typical machine.
If inputs must be purchased in large units, inventories of them must be held as they
are used up over time. This confers an economy of scale as output is increased so
allowing the average inventory of such commodities to be reduced.
The other author in our survey who mentions this effect is Rutherford (2000:
220) who writes: ‘The nature of a factor of production or commodity which is
only supplied in discrete amounts, not increasing or decreasing in quantity
continuously.’ He goes on to say: ‘Indivisibilities are responsible for many
fixed costs in the short run and give rise to production economies of scale at
high levels of output.’
Ex ante Indivisibility and the Absence of Constant

Returns to Scale
A few authors define ex ante indivisibility as the absence of constant returns to
scale as production is decreased. But to define indivisibility as the absence of
constant return of scale for decreases in output and then argue that indivisi-
bilities explain non-constant returns is circular reasoning. The only author in
our survey to note this is Morroni who writes (1992: 144–5): ‘the assertion
often found in the literature that . . . increasing returns to scale depend on
indivisibility, is a tautology’.
Indivisibilities: Certain processes of production . . . can only be operated

efficiently at large output volumes and cannot be operated as efficiently at
lower outputs, even if all factor inputs are scaled down in proportion to one
another. We call this inability to scale down the processes of production
without affecting their efficiency of operation, indivisibility of the produc-
tion process. (Griffiths & Wall 2000: 170)
These authors list indivisibilities as a source of scale effects and so seem to be

involved in the circular reasoning just discussed.
‘Divisibility: Output from any process can be expanded or contracted to any

arbitrary degree by proportional expansion or contraction of all inputs used
in the process’ (Quirk 1987: 145).
We can infer that indivisibility is defined as the absence of the above condition.
The author does not, however, link indivisibilities so defined to scale effects.
The proportionality postulate states: ‘if an activity a = (a1, a2, . . ., an) is
possible, then every activity λa = (λa1, λa2, . . ., λan) of which the net outputs
are proportional to those of a, with a non-negative proportionality factor, λ, is
also possible’ (Koopmans 1957: 76). He goes on to state that a (process)

indivisibility occurs when one or more of the ais cannot be varied continu-
ously, particularly if it is subject of an inequality constraint: ai ≥ aˉ ɩ. This is not
the circular reasoning of the others who define indivisibilities as the absence
of a constant-returns PF, but a statement that indivisibility occurs when one of
the variables in the constant-returns PF cannot be altered in proportion to the
others, or in the limit, not altered at all. Koopmans is the only author in our
survey who appears to believe that all scale effects are due to indivisibilities
as he defines them.23
Having originally agreed with Koopmans that all cases of increasing returns
to scale were due to indivisibilities, Kaldor (1972: 102–3) later recanted and
argued that there were other reasons for scale effects such as geometrical
relations, and also that Koopmans’ concept of indivisible inputs was mislead-
ing. On this he wrote: ‘There is nothing “indivisible” about tubes or pipelines as
such, technically, it may be just as easy to make tubes of a relatively small or a
relatively large dimension there can be a continuous range of sizes in between.’
Although he does not use our terms, Kaldor is clearly chiding Koopmans for
using an ex post rather than an ex ante definition in his pipeline example.
It is not quite clear how to categorise the rather odd treatment by Bannock et
al. (1984: 141, bracketed numbers added):
Indivisibilities: Many types of plant and machinery have, for engineering

reasons, a single most efficient size. Either [1] it will be technically impos-
sible to make the equipment at a different size, or [2] the production costs
associated with other sizes are higher . . . .Certain types of production
processes may only be viable at certain rates of output. The word indivisi-
bilities is used to categorize these sources of scale economies because [3]
they would not arise if the plant and processes were capable of being
increased or decreased in scale by small amounts without any change in
their nature, i.e. if they were perfectly divisible.
The point numbered (1) refers to a (unlikely) form of ex ante indivisibility in

that the machine can only be made to one size while (2) seems to explain scale
efficiencies by themselves in the sense that the variations in unit cost as size
varies is not an indivisibility but a result of some other factor that causes scale
23
Notice that this appears to deduce an empirical proposition – scale effects can only result from
indivisibilities – from a highly abstracted formulation of a production function in which all that
appears is the flow of services from the various inputs, including capital goods, but not the goods
themselves that are subject to the alleged indivisibilities. The correct deduction is not that scale
effects can only result from an indivisibility of inputs, but that this abstract formulation has
removed all possible sources of scale effects other than those that are associated with some
characteristic of the input flows, of which indivisibility is the obvious candidate.
effects. Entry (3) correctly points out that ncrs-indivisibilities are a necessary
condition for EoS to occur.24
Ex ante Divisibility and Indivisibility of a Capital Good

Finally, we come to the use of the term in the literature that seems to apply to the
type of ex ante indivisibility that we have argued is the only type of indivisi-
bility that can explain variable scale effects. Unambiguous statements to that
effect follow.
‘A commodity is indivisible if it has a minimum size below which it is
unavailable, at least without significant qualitative change’ (Baumol 2008:
242). Compare this with Baumol (1977) cited above and note that the last
qualification ‘without. . .’ is ambiguous.
Indivisibilities are defined as ‘the existence of a minimum scale at which any
technique can operate. This applies to all productive techniques’ (Black et al.
2012: 204). I presume they mean ‘operate at full capacity’ as most techniques
can operate below full capacity, which is the range of increasing returns for the
firm with the curve SRATC1 in Figure 1.
‘A commodity is indivisible if it has a minimum size below which it is

unavailable’ (Calhoun 2002: 229).
‘[I]ndivisibilities: the minimum physical or technical size limitations of

factor inputs’. ‘Economies of scale may . . . arise due to: (a) indivisibilities
in machinery and equipment especially where a number of processes are
linked together’ (Pass et al. 2005: 251 and 156).
‘The characteristic of a factor of production or commodity which prevents its

use below a certain minimum level’ (Pearce 1992: 201). The qualification ‘use’
is odd here because, as already mentioned, equipment can be, and frequently is,
used at less than full capacity. Presumably this refers to the machine that
produces the service that is the factor of production. The machine comes in
some minimum size (ex ante indivisibility) but not the service input. This is yet
another example of the ambiguities that enter because of the failure to consider
explicitly two-stage production processes so that the machine that produces a
service input is confused with the input itself.
‘Consider a production process that requires constant input proportions and
yields constant returns to scale. Two persons and two shovels can do twice as
much as one person with one shovel; but we cannot have half a person with half
24
Bannock et al., also quote the division of labour as a kind of indivisibility: ‘Expansion in scale of
activities permits greater specialization and division of labour among workers . . . This, in effect,
is also an “indivisibility”, in that it is the result of the fixed capacity of individual worker and the
fact that it is optimally utilized when devoted exclusively to a specific task’ (1984: 141).
a shovel; consequently, we have an indivisibility in the production process’

(Miller 1978: 458). Presumably one person and one shovel is assumed to be the
minimum ex ante size that can do the digging job (no part time work or hiring of
the services of the shovel for less than full time).
Examples of indivisibilities are the cost of designing and developing a

product, such as an aircraft, car or computer programme, setting up a
production line for a product or for a batch of a product, the capacity of a
machine and a TV advert. A firm that cannot fully utilize the item to which
the indivisible cost relates, or spread its cost over as large an output as its
rivals, is at a disadvantage. However, it may be able to damp down its cost
disadvantage by, for example, acquiring a second-hand machine tool, or by
using a different type of promotion in place of TVadvertising. (Pratten 2004:
283)
Although not altogether clear, this would seem to refer to ex ante indivisibilities
while mixing once-for-all costs of product design with costs associated with the
indivisibilities of production equipment.
2.3.8 Are Indivisibilities a Source of Economies

of Scale or Returns to Scale?
As in most other cases, the authors who cite indivisibilities as important causes
of scale effects differ in how they classify these as sources. Some list them as
sources of IRTS.25 Others say they cause both IRTS and EoS.26 Many others
list them as sources of EoS only.27 Finally, Eaton and Lipsey (1977) argue that
indivisibilities are the key to explaining why production is located at points in
space rather than being spread evenly at points of consumption. If we add ex
ante to their term ‘indivisibilities’ this attribution is satisfactory.
2.4 Scale Effects Due of Geometry and Physical Laws

Several key characteristics of our world are important for understanding the
pervasiveness of scale effects related to the design effects of reconfiguring capital
goods. One source of these is the world’s three dimensionality. For example, the
volume of any regular container increases more than in proportion to the materials
needed to make up its sides. Another source is our world’s physical laws where non-
25
These are Kamerschen and Valentine (1977: 179) and Koopmans (1957: 76).
26
These are Griffiths and Wall (2000: 170), Bannock et al. (1984: 141), Setterfield (2001: 489),
Carlaw and Lipsey (2008a: 222), Seldon and Pennance (1976: 120, 295) and Pass et al. (2005:
156, 475).
27
These are Ammer and Ammer (1984: 415), Bain (1968: 492), Baumol (1977: 274), Black et al.
(2012: 123, 204), Brush (1994: 339), Calhoun (2002: 229), Pearce (1992: 201), Silvestre (1987:
2797), Rutherford (2000: 220), Sloman (2006: 132) and Graaff (1987: 7599).
linear effects also abound. For example, the observed intensity of a light is inversely
proportional to the square of the distance from the light source. Yet another source is
the stochastic nature of many aspects of our world’s behaviour. For example, if the
probability of one event happening is 1/r (1 < r), then the probability of n such
events happening at the same time, but independently of each other, is 1/rn, which is
diminishing non-linearly in n.
So, when we change the scale of almost anything, from one machine to a
whole plant or skyscraper, we should, as have already observed in the intro-
duction, expect to encounter scale effects of the sort that we call design effects
and for these pervasive reasons alone – and there are other causes as well.
These effects should be no surprise; the only surprises should occur if we
encounter no scale effects when the scale of any operation is changed.
2.4.1 Empirical Evidence of Design Effects

with Capital Goods
Somewhat less general than the Eaton-Lipsey treatment discussed above, but
equally interesting is the treatment provided by Dudley Jackson (1996) who is
one of the very few authors in our survey who introduced detailed empirical
content into their discussions of scale effects. (All quotations are from pagers
229–30.) He deals with these effects at the level of individual pieces of capital
equipment. Importantly, he defines scaling up in the way found in the engineer-
ing rather than the economics literature. Instead of using the returns definition
of multiplying all inputs by some constant λ, one that has few if any real-world
counterparts, he defines scaling up as occurring when ‘each piece of equipment
of smaller capacity is replaced by a piece of equipment (of the same type) but
with a larger capacity’. As he puts it,
[E]conomies of scale refers to the situation in which, at a given and unchan-

ging set of input prices, the unit cost of production is lower in a plant of larger
scale of (annual) output than in a plant of smaller scale. This happens
because the physical requirement for labour and capital inputs per unit of
output each tends to be lower the larger is the scale of annual output which
the plant is designed to produce.
Clearly he is referring to somethings close to, but not identical with, what the
literature calls IRTS, which of course, given constant input prices, leads to EoS.
He then defines the power rule as:
Let K denote the acquisition cost of a piece of equipment and V . . . [its]

capacity (however measured); the subscripts i and j denote, respectively,
pieces of equipment of the same type but different capacity, i being the larger
capacity and . . . let s denote a parameter of the power rule . . . Ki/Kj = (Vi/Vj)s.
He goes on to argue that for economists ‘the interest of such data lies in the
value of s for each type of equipment; nearly always less than one and centering
around approximately 2/3. For example, in a sample of 288 types of equipment
the average value of s was 0.6559 (with a standard deviation of 0.2656)’. Thus,
a doubling of capacity of some generic capital good increases its cost on
average by about two-thirds. He concludes that ‘economies of scale is a wide-
spread and inherent feature of the behaviour of the acquisition cost of equip-
ment or of a plant as capacity is “scaled up” ’.
This is strong evidence of the pervasiveness of scale effects related to
individual pieces of capital equipment. This even though he uses a different
definition of IRTS than is found in the economics literature, one that applies
where it is possible to build different sized versions of what is basically the
same type of capital equipment. Note, however, that when the scale of opera-
tions is increased, building a larger set of capital goods to do the same range of
jobs is not always optimal behaviour compared with reorganising the whole
production process using quite different equipment.
2.4.2 Sources of Design Effects: Geometry and Physics
As already observed, we live in a three-dimensional world that entails many

scale effects, both increasing and diminishing. For example, the geometrical
relation governing any container typically makes the amount of material used,
and hence its cost, proportional to one dimension less than the service output,
giving increasing returns to scale with respect to the inputs of materials over the
whole range of output. For example, the capacity of a closed cubic container of
sides s is s3. The amount of material required for construction is 6s2. So,
material required per unit of capacity is 6/s, which is diminishing in s, indicat-
ing IRTD. Also, the total amount of welding required to seal the sides of the
container is proportional to the total length of the seams, which is 12s, or 12/s2
per unit of capacity. So not only does both the amount of materials and the cost
of sealing the seams fall per unit of capacity as the capital good’s capacity is
increased, they fall at different rates. This relation holds for more than just
storage containers. Blast furnaces, ships, gasoline engines, office and residen-
tial buildings are a few examples of the myriad technologies that show such
geometrical scale effects.
The geometrical reasoning given above cannot produce a final conclusion
about scale effects related to containers; one needs to know some physics as
well. It is imaginable, for example, that as the capacity of a container is
increased, the walls would need to be thickened proportionally, making the
volume of material increase linearly with the container’s capacity. Physical
relations dictate that in most cases this is not so. Although some thickening is
often required, in many cases, the thickening is less than in proportion to the
increase in the surface area. Then the volume of material used increases less
than in proportion to the increase in capacity (although more than in proportion
to the increase in surface area). In either case, the returns depend on the
technical relations that govern each case in question.
To make an old fashioned light bulb last longer, what is required is to alter the
strength of the filament without a proportionate change in most of its other
components. Similarly, to make a light bulb deliver a larger wattage of light per
unit of time, what is needed is to change the resistance of the filament with no
change in the other components. This gives IRTD that occur when either the
duration or rate of flow of the services of the light bulb are varied over a wide
range of duration and wattage. This example generalises to the large number of
technologies in which the flow or intensity of its service is a function of only
part of the device.28 In all such cases there is in effect a spreading of overheads
as the parts that do not require alteration have falling cost per unit of output as
the technology is changed.
There are many scale effects associated with ships, some depending on
geometrical relations and others on physical laws. First, the maximum speed
that a displacement hull can be driven through the water is proportional to the
square root of the length of the hull on the water line (planing hulls obey
different laws) (Hiscock 1965: 138). No amount of a priori reasoning could
reveal this rather mysterious relation. Second, while a ship’s carrying capacity
is roughly proportional to the cube of its length on the water line, geometrical
relations plus the physics governing structural strength of a hollow body dictate
that the ship’s cost is related approximately linearly to its water line length
(Rosenberg & Birdzell 1986: 83). Third, altering the ship’s size also alters its
handling and safety characteristics in complex ways. Fourth, as the size and
other characteristics of a ship are changed, there is an alteration in the materials
best used for its construction. Finally, the amount of horsepower per ton of
cargo required to move the ship through the water, changes as the size of the
ship changes, falling over a wide range as size increases. Thus, building larger
ships alters carrying capacity, construction costs, operating costs, speed and
other handling characteristics, each in a different proportion.
The cost per passenger mile for airplanes falls steadily as the size of the
aircraft is increased over a wide range of sizes. Also, there is an improvement in
its stability in the face of turbulent conditions and in several other handling
28
Indeed, all that is needed is that the output be differently related to the various parts of the
device, not that the relation be zero for variations over a minimum necessary amount for some
parts and a given positive number for the others.
characteristics. Analogous considerations cause many scale effects in land

transport – again in response to both geometrical properties and physical laws.
According to the physics of heat, the heat loss from a blast furnace is
proportional to the area of its surface, while the amount of ore that can be
smelted is proportional to the cube of the surface sides. This relation will be
considered in more detail in Section 5 and is a source of increasing returns in
the relation between fuel used and output capacity of such furnaces. Once
again, the relation arises from a combination of the physics and geometry of our
world. (Since the time of Isaac Newton, practical engineers have known that a
small body loses heat faster than a large body. See, for example, Cardwell
(1995: 158).)
As with blast furnaces, the heat loss from a steam engine’s cylinder is
proportional to the cylinder’s surface area while the power it generates is
proportional to the volume of the cylinder. This is one of the several reasons
why the thermal efficiency of a steam engine is an increasing function of its size
over a wide range starting from zero. This in turn is why steam powered
factories were built much larger than the water powered factories that they
displaced. There are no similar scale effects with electric motors (beyond a very
small size), which is one reason why small-scale parts manufacturers (feeding
into large-scale assemblers) became efficient when electricity replaced steam
as the major power source for manufacturing. Also, when the absence of
significant scale effects allowed unit-drive machines (one motor per machine)
to replace machines driven by belts emanating from a central drive shift driven
by a single large steam engine, a more rational arrangement of the machines in
the factory became possible. This in combination with new machine tools that
could cut pre-hardened steel, eventually led to the introduction of the assembly
line with its massive scale economies – a development that would not have
been possible in steam-powered factories. Although these changes took place
over time and so were another example of what Lipsey, Carlaw and Bekar
(2005) call historical returns to scale (see Section 5), they illustrate the influ-
ence at any one time of scale effects on the general economy. Each of these
technologies was associated with a given and different set of scale effects, so
each influenced in different ways the current organisation of firms and indus-
tries and also had important effects on productivity.
The Literature
Although none of our surveyed authors refer to physical laws, many mention
geometrical properties as a source of scale effects. We have already mentioned
the important material on the magnitude and scope of scale effects in machinery
cited by Dudley Jackson (1996), effects that he attributes to the geometrical

properties of machinery. Of the rest, Sloman (2006: 132) says more about
geometrical effects than most of the others in our survey. He refers to the
‘container principle’, which he says is a source of IRTS and EoS at the plant
level: ‘Any capital equipment that contains things (blast furnaces, oil tankers,
pipes, vats, etc.) tends to cost less per unit of output the larger its size. The
reason has to do with the relationship between a container’s volume and its
surface area.’ His comments might be seen as implying geometrical sources
more generally, although all of the examples he cites are applications of his
‘container principle’, while as we have seen above, there are causes that rely on
geometrical properties other than the container principle. Carlaw and Lipsey
(2008a: 222) list spatial facts about the world (e.g. surface area-volume rela-
tionships) as causes of IRTS in capital goods which lead to EoS for the final
goods producers that use them. Jackson (1996: 230) argues that spatial relations
such as that between surface area and volume are the most important source of
EoS in the process of ‘scaling up’ production facilities. Pass et al. (2005: 156)
list under the heading EoS ‘economies of increased dimensions for many types
of capital equipment (e.g. tankers, boilers) both set-up and operating costs
increase less rapidly than capacity’. Bannock et al. (1984: 141) refer to the
volume effect in general and shipping in particular as sources of both IRTS and
EoS. Setterfield (2001: 488) and Miller (1978: 208) also mention geometry as
causes of both. Baumol (1977: 274) lists them under a returns section but in
places refers to them as economies, while Griffiths and Wall (2000: 171),
Kamerschen and Valentine (1977: 185), Quirk (1987: 149) and Varian (1992:
15) call them sources of IRTS.
In contrast to these writers, Eatwell (2008: 140) argues as follows:
There are some examples in which outputs are an increasing function of

inputs for purely technical reasons. . .. Such technical examples are not,
however, the examples which typically come to mind in the discussion of
increasing returns to scale. More typical are examples of mass production, of
production lines, or, today, of production integrated by means of sophisti-
cated information systems.
One wonders if many of the examples just mentioned ‘came to Eatwell’s

mind’ only to be dismissed, or if he was unaware of them. It is symptomatic of
many economists’ lack of interest in the technology of the real world that
Eatwell dismisses these nearly ubiquitous cases as ‘purely technical’, as if
economics was not much concerned with the technical matters that underlie so
many economic relations. One also wonders, if the scale effects associated with
sophisticated information systems are not due to ‘technical reasons’, what

mysterious non-technical reasons do account for them.29
2.4.3 Scale Effects Due to Random Behaviour

As already observed, we live in a world in which outcomes are often probabil-
istic rather than being deterministic (to say nothing of those that are genuinely
uncertain). As a result, statistical relations, such as the law of large numbers,
confer scale effects on many of our activities, effects that depend on design
changes that exploit neither geometrical relations nor physical laws.
In insurance, the greater the number of independent events that are insured
against, the smaller is the probability that the loss in any one year will exceed
the premiums paid in that year, yielding EoS in insurance. Similar considera-
tions apply to many other activities whose outcomes are subject to chance, such
as inventories held by producers and precautionary money balances held by
agents.
These examples illustrate that when the scale of output is altered in the long
run, and capital is altered by reconfiguration with known technology rather than
replication, the nature of the world in which we live will almost always produce
a complex set of scale-related changes in the real resource cost of capital goods.
Other changes will alter the capital good’s performance characteristics in ways
that are only indirectly reflected in the relevant service flow.
An interesting set of cases where the design of capital goods in response to
probabilistic behaviour creates scale effects has been studied by Kenneth
Carlaw (2004). As we have already mentioned, capital goods are commonly
comprised of several distinct components that are complementary to each
other. These technological complementarities, plus uncertainty about how
long each component will last, create problems for designers of capital
goods. To study this issue Carlaw assumes, in one of his many illustrative
cases, that the probability that any one component of the capital good will be hit
by a damaging stress in any one period is 0.5 and the designer of the capital
good wants it to yield services for three periods at a rate of one unit of service
per period. When a stress arrives, it does so at the beginning of the period, and
the component delivers its services in that period only if the number of stresses
it has been designed to withstand has not then been exceeded. For simplicity, he
assumes a zero rate of time discount. First let a given capital good have only
one component. The expected total value of services over three periods
29
Lipsey (2009: 853–4) discusses other cases in which the reluctance of many economists to be
concerned with the details of relevant technologies hinders their attempts to explain important
events.
increases linearly as the component is built to withstand up to two stresses but

then by a smaller amount when resilience to a third stress is added.30 Now
suppose that the good is made of two components, each with the same but
independent probability (0.5) of a damaging stress occurring at the beginning
of each period and that the loss of any one component renders the good useless.
In this case (again assuming a zero rate of time discount) the expected value of
total services increases at an increasing rate over some range as the number of
stresses each component can withstand is increased.31 So the cost-minimising
designer will create durability to exploit the scale effects inherent in the nature
of multiple component systems and risk of failure.
Obviously, the conditions of this example can be altered in many ways,
including more periods to be served, more components liable to failure,
different costs of protection against failure for each component or each addi-
tional unit of protection in any one component. However, there is no a priori
reason to expect constant returns to design when any one of these conditions is
altered.
Four of our surveyed authors mention probability issues: Bannock et al.
(1984: 141) who refers to the law of large numbers so that ‘the size of stocks
will vary less than proportionately with the scale of output, expenditure or
receipts’; Baumol (1977: 274) who refers to inventories and uses both the terms
returns and economies in discussing them; Carlaw and Lipsey (2008a: 222)
who cite this as a source of IRTS leading to EoS; and Kamerschen and
Valentine (1977: 185) who refer to the statistical law of large numbers.
2.4.4 Specialisation of Capital and Labour
One particular source of design effects, the increased specialisation of labour

and capital as output rises, is quoted by many writers as a source of increasing
returns to scale. Specialisation of labour and capital equipment usually go
30
If the component is built to withstand no stresses, the probability of surviving to be used for
three periods is 0.125 and the expected total services are 3(0.125) = 0.375. With the ability to
withstand one stress, the probability of surviving three periods is 0.5 and the expected total
services are 3(0.5) = 1.5. With the ability to withstand two stresses, the probability of surviving
three periods is 0.875 and the expected total services are 3(0.875) = 2.625. With the ability to
withstand three stresses, the probability of surviving three periods is 1 and the expected total
services are 3.
31
If each component is built to withstand no stresses, the probability of surviving three periods is
0.0156. With the ability of each to withstand one stress it is 0.25. With the ability to withstand
two stresses it is 0.766. With the ability to withstand three stresses the probability of surviving
three periods is again obviously 1. Expected total services in each case are calculated as in the
previous footnote. We see in this case that there is a range of increasing returns from zero to two
hits. But building the components to each withstand a third hit yields diminishing returns for this
last unit of durability. Assuming a constant cost of increasing resilience to hits, the average cost
curve is U-shaped.
together, one requiring the other. So, we will discuss mainly the use of
specialised capital as output expands. In this case labour and simple tools are
replaced by specialised machines, not just larger versions of one generic
machine. While in most of our surveyed literature this is merely asserted,
Vassilakas presents an extended treatment calling it the major source of
increasing returns to scale. He quotes Adam Smith’s three reasons for this:
‘First, . . . the increase of dexterity in every particular workman, secondly . . .
the saving of the time which is commonly lost in passing from one species of
work to another, and lastly . . . the invention of a great number of machines
which facilitate and abridge labour, and enable one man to do the work of
many’ (as quoted by Vassilakas 1987: 4626). He then covers extensively the
history of the concept since Smith’s time. However, he offers no further reasons
why increasing the division of labour is a source of ubiquitous scale effects;
instead he confines his discussion to the circumstances under which these
increasing returns are or are not exploited by firms. This is an interesting
question, but not our present concern, which is the source of scale effects. If
we accept Smith’s reasons, we adapt his third point to our static analysis by
saying: ‘allowing the use of existing machines which . . .’.
Why does increased specialisation of capital goods (and the accompanying
labour) typically lower unit costs? There is no fundamental overarching
principle involved here. It is merely a common observation that replacing
labour by specialised machines (and their specialised operators or super-
visors, if there are any) often lowers unit costs because the machine can do
the job faster, more precisely, with fewer errors and less variations of perfor-
mance due to fatigue than can a labourer. This is often so much more efficient
that the accumulated labour, materials and other costs that go to make the
machine, plus the wages of its operator (if there is one), are lower per unit of
output than the cost of the less specialised labour and machines that it
replaces. This was the case for example when machine-operated X-rays and
lasers replaced the lumber mill’s experienced sawyer in determining where to
make the first cut in a log. More generally it was the case when the shift of
output from artisan-style to mass-production-style production when many of
the things that experienced artisans did were replaced by much faster moving
machines.
All of this raises the question of why the less specialised version of the
production process is ever used if the more specialised version is more pro-
ductive. The answer lies in the ex ante indivisibility of the more specialised
version.
The Literature
Labour and capital specialisations are usually listed separately in the literature.
Several authors state that labour specialisation causes both a returns and an
economies effect unconditionally.32 Others, argue that it is a source of both
types of effect under the assumption that input prices are constant.33 Yet others
in our survey state that it is a source of EoS only.34 Pass et al. (2005: 156, 475)
appear to argue incorrectly that the presence of an EoS implies that of an IRTS
(while, as we have seen, the latter implies the former, the former does not imply
the later).35
Several writers also cite managerial specialisation as a separate source of
scale effects, although there would seem to be no obvious analytical difference
between white and blue collar specialisation. Some list it as a source of both
IRTS and EoS, sometimes with and sometimes without the qualification of
fixed input prices, while yet others list it as a source of only EoS.36 On the
specialisation of capital, some writers make it a source of both IRTS and EoS
either unconditionally or with the qualification of fixed input prices.37 Others
list EoS only.38
Closely related to the ability to use more specialised capital as the scale of
output expands is the ability to use different and better organisations of
production, indeed both often occur at as one integrated set of changes. Two
32
These are Krugman and Wells (2009: 323), Pass et al. (2005: 156, 475), Setterfield (2001: 489),
Seldon and Pennance (1976: 120, 295), Pindyck and Rubinfeld (2009: 215, 245) Griffiths and
Wall (2000: 171) Mansfield (1979: 185) and Bannock et al. (1984: 141). Kamerschen and
Valentine (1977: 185) discuss this issue under the heading of ‘returns to scale’ but make
statements that might can be interpreted as meaning economies as well.
33
These are Perloff (2012: 172), McConnell et al. (2012: 154), Sloman (2006: 132) and Parkin et
al. (2005: 195, 214).
34
These are Ammer and Ammer (1984: 415), Bain (1968: 492), Mankiw (2006: 274) Abraham-
Frois (2008: 232), Bannock et al. (2003: 114) Black et al. (2012: 123), Farrell (1997: 432) and
Shim and Siegal (1995: 118).
35
They state, ‘Where economies of scale are present, a doubling of factor inputs results in a more
than proportionate increase in output’ (Pass et al. 2005: 475). They may of course be using the
term EoS to mean IRTS.
36
Pass et al. (2005: 156, 475), McConnell et al. (2012: 154) and Seldon and Penance (1976: 295)
say IRTS and EoS, in McConnell’s case given fixed input prices. Bain (1968: 492), Calhoun
(2002: 137) and Pearce (1992: 122) say EoS only.
37
These are Setterfield (2001: 489) Seldon and Pennance (1976: 120, 295) and Griffiths and Wall
(2000: 171) unconditionally and Case et al. (2012: 196), Parkin et al. (2005: 214) and Perloff
(2012: 172) with constant input prices.
38
These are Abraham-Frois (2008: 232), Bain (1968: 492), Bannock et al. (2003: 114), Farrell
(1997: 432), McAffee (2006: 4–101), Shim and Siegal (1995: 118) and Ammer and Ammer
(1984: 415). Sloman (2006: 132) cites plant specialisation as a source of EoS without mention-
ing specialisation of capital equipment, although possibly it is meant to be included in plant
specialisation.
of our surveyed authors mention this, one as a source of EoS (Sloman 2006:
132) and the other as a source of both (Pass et al. 2005: 156, 475).
2.5 Once-for-all Costs

All of the sources cited so far affect the position and slope of the LRAC curve
and so influence the firm’s long-run decision on the size of plant: major scale
effects in plant and equipment make the SRATC curve steep while small effects
make it flattish. We now come to a set of sources that do not affect the firm’s long-
run decision but do confer something similar to scale effects but only over the
short run – a distinction made by none of those in our survey who discussed once-
for-all costs.
One characteristic generic source in this group is a once-for-all fixed cost
attached to the firm’s entry into the relevant market for its product, other than
those for plant and equipment. More detailed sources of this sort include the
cost of acquiring a patent and other rights, legal costs, training costs and a host
of other once-for-all startup costs.
These once-for-all fixed costs affect the size of the firm’s profits but, being
independent of the scale of output, they do not affect the size at which the firm will
choose to enter.39 Thus unlike the scale effects considered so far, these do not
influence the selection of the firm’s initial scale of operations and if conditions are
static, will not influence any subsequent behaviour. However, if demand rises
unexpectedly after the firm’s size has been determined, there will be a short-term
effect when these costs are spread over more units of output as the firm moves
down its short-run cost average total curve. The firm is now inside its LRAC and
in the long run will want to move to a new point on that curve as result of the
increased demand – a point that once again is unrelated to the bygone cost of entry.
2.6 Mixed Cases

Many of the examples of these once-for-all costs that are cited in the literature
are actually mixed cases that include both influences that affect the firm’s long-
39
Since many readers have doubted this result and its interesting implications, I give a simple
proof for the case of a quadratic LRAC curve and a linear demand curve. Let the equation of the
Viner envelope be AVC = F/q – aq + bq2. Note that this not a production function but the
equation of the U-shaped envelope, which tells the firm what combinations of output and unit
costs it can chose when it enters. Let the demand be: p = c – dq. Totals are TC = (AVC)q = F – aq2
+ bq3 and TR = pq = cq –dq2.Total profit is: TR – TC = π = cq –dq2 –F + aq2 – bq3. To choose the
maximising location, differentiate π with respect to q and set the result equal to zero: dπ/dq = c –
2dq + 2aq – 3bq2 = c – 2q(d–a) – 3bq2 = 0, which is a quadratic indicating the qs that maximises
and minimises profits. Since it is independent of F, the entry value of q does not change,
although the firm’s profit does fall, as F increases. Note that this does not tell us anything about
the firm’s short run costs except that the equilibrium long run choice of q must also be on the
relevant SRATC curve.
run decision and those that do not do so, but do confer falling unit costs in the
short run as just discussed.
A typical example of such a source is advertising costs. To be able to classify
this alleged source of scale effects, we need some empirical knowledge – none
of which was cited by the authors in our survey. At one extreme, a new firm
entering with a new differentiated product may require some minimum amount
of advertising to establish its market. This is similar to the cases just considered
where once-for-all fixed costs do not affect the specifics of the entry decision
(as long as they are not so high as to preclude profitable entry). Whether, after
this minimum necessary amount of advertising the marginal returns to further
advertising first rise then fall in a U-shaped fashion, or fall from the outset, is a
matter that may well vary with time, place and type of commodity. On the
assumption that sales vary directly with a given continuing flow of advertising
expenditure, the additional advertising cost can be added to the marginal cost of
production. The effect of these costs depends on the strength of the relation
between advertising and sales. Depending on this elasticity, its contribution to
marginal costs could be rising, falling or constant per unit of extra output.
Unlike the case of the once-for-all fixed costs, these costs shift the firm’s cost
curves in various ways depending on the relations just discussed and so will
shift the LRAC curve, but also in ways that cannot be determined until the
relevant elasticities are known. But one way or the other, they do contribute
scale effects that influence the firm’s long-run decision concerning its scale of
operations.
A similar case is provided by the R&D costs of developing a product that a
new firm requires to enter the market. In the simplest case the detailed nature
of the product, and hence sales, are independent of the cost of development –
the product is either acceptable or it is not. In a more realistic case, the more
that is spent on development, the better is the product and hence the greater
the sales. In the first case, we have a simple fixed cost situation as already
discussed. In the second case, everything depends, as with advertising, on the
strength of the relation between R&D and sales. Depending on this elasticity
there could be EoS, CoS or DoS to the expenditure on R&D. Of course, this is
only a ceteris paribus scale effect. What happens to the firm’s overall unit cost
as output varies will depend on the sum of everything that affects this cost at
the margin.
A related effect concerns the modern situation in which many firms must
innovate a continual stream of new products in order to stay competitive. Such
cases are taken up in section 3.
The Literature
Several authors in our survey refer to up-front costs as sources of scale effects,
although none of them give a hint of any of the complications of which the
above are just examples. These costs give rise to EoS according to some.40 One
author says they are a source of both EoS and IRTS without reservation,41 while
yet others say they are a source of both under the assumption of fixed input
prices.42 Advertising is cited by various authors as a source of either EoS or
IRTS of both under the assumption of given input prices and others without
such a qualification.43
2.7 Purely Pecuniary Sources

Although not always clear, the following cited sources may be pecuniary but in
some circumstances may be based on real resource savings.
Ammer and Ammer (1984: 415) refer to discounts on inputs in general and,
separately, to ‘greater bargaining power with labour unions banks and other
lenders’. Access to, and favourable terms of obtaining, financial capital are
cited by several who say they are sources of EoS, and two who say EoS and
IRTS.44 In so far as transportation costs vary in direct proportion to the amount
shipped, these are not scale effects. In so far as quantity discounts become
available as quantity shipped rises, they are pecuniary EoS to the firm whose
goods are being shipped but may be real economies to the shipper if larger
quantities require few resources per unit shipped than do smaller quantities.45
2.8 External Scale Effects

So far, we have dealt with cases where the scale effects are internal to the firm,
either the producer of a final good or the producer of a capital good. We now
40
These are Frank and Bernake (2009: 239), Bannock et al. (1984: 141, 2003: 114) and Barbosa-
Filho (2008: 606).
41
Krugman and Wells (2009: 323).
42
These are Rutherford (2000: 142) and McConnell et al. (2012: 156).
43
These are Brush (1994: 340) who refers more generally to marketing costs, Calhoun (2002: 137)
who lists these as sources of DoS, and McConnell et al. (2012: 154–6), Pass et al. (2005: 156)
and Seldon and Pennance (1976: 295) all of whom list them as a source of both EoS and IRTS –
McConnell et al. with the qualification of fixed input prices, and Pass et al. and Seldon and
Pennance unqualified.
44
EoS: Ammer and Ammer (1984: 415), Bain (1968: 492), Calhoun (2002: 137), Farrell (1997:
433), Pearce (1992: 122), Rutherford (2000: 142), Sloman (2006: 133), Abraham-Frois (2008:
232), and Bannock et al. (2003: 114). EoS and IRTS: Pass et al. (2005: 156) and Seldon and
Penance (1976: 295).
45
These are cited by Bain (1968: 492), Bannock et al. (2003: 114) and Case et al. (2012: 196) as
sources of EoS.
consider some cases of sources that are mainly or wholly beyond the control the
individual firm.
2.8.1 Network Effects

These can be defined in different ways. For one example, consider the output of a
monopoly firm that is in the business of providing phone services to customers.
Assume that each customer can be connected to the exchange at a constant cost
and that the value of the phone service to the nth customer added is an increasing
function of the number of other subscribers to whom she can connect. This firm
can produce its connections at CRTS but can sell them at a price that is an
increasing function of its size measured by the number of the customers that it
serves. This is certainly a scale effect but not one that is covered by the conven-
tional definitions of IRTS or EoS as it raises the value of output and hence the
price at which it can be sold. Also as this example shows, it is not necessarily an
external economy. It is certainly internal to the industry and, if the number of
firms is small (one in the limit), it will be partially (or wholly in the limit) internal
to the firm. The source of these scale effects is in the combinatorial nature of
interconnections while the ability to exploit such latent effects depends on the
size of the market. (For example, if a given percentage of the residents in some
area want a service subject to a network effect, its per-customer value will
increase as the total population increases.)
Network effects stemming from the demand rather than the cost side are
listed by Frank and Bernake (2009: 238) as sources of EoS, while Krugman and
Wells (2009: 323) say that they are sources of both IRTS and EoS.
2.8.2 Specialised Suppliers/Services

If the expansion of an assembly firm causes its suppliers to increase their
outputs and thereby reach a lower point on their declining LRAC curves,
there is an IRTD for the supplier firms and an EoS for the assembly firm.46 In
a rare case of complete agreement on whether we have a case of IRTS or EoS,
these are listed as a source of EoS by all of the authors in our survey who
mention them.47
46
If one uses the type-2 meta-production function described in Section 1.4.5, this is a case of IRTS
to the parts supplier. But viewed as either an IRTS or an IRTD, it is a real resource saving change
from the view point of the pas supplier and an EoS from the assembler’s.
47
These are Ammer and Ammer (1984: 415), Farrell (1997: 434), Stockfisch (1968: 272), Sloman
(2006: 133), Bannock et al. (2003: 114) Black et al. (2012: 123), Calhoun (2002: 137), Pearce
(1992: 122) and Graaff (1987: 7599).
2.8.3 Shared Infrastructure

There is little that can be concluded about any scale effects of shared infra-
structure in the absence of much more information – except that in various
circumstances it could generate no scale effects, an EOS or a DoS. To illustrate
the possibilities, assume that the services of the infrastructure are an input to the
firm in question and are provided at a constant price per unit of use, a price that
might be zero (as with many roads) or some positive constant. Initially there is
excess capacity on that facility. As all the firms in the industry expand, they get
one of their ‘inputs’ at a constant price both before and after the expansion.
Thus, there is no scale effect from their point of view. But the firm that is
supplying the infrastructure encounters EoS as the fixed cost of creating the
infrastructure is spread over more and more customers. If the infrastructure
reaches capacity there will be an external DoS as the time and convenience
costs of use to each firm rise as industry use rises. Now assume that the price
that is charged for the use of the infrastructure’s services is positively related to
the unit cost of providing the service – a cost that is dominated by the high
construction cost and hence a fixed cost to be spread over the users. Now each
user will get an external EoS as industry use rises up to the facility’s capacity.
After that, as more and more users exacerbate the congestion, the internal time
and convenience costs to the user will rise and eventually overcome any
external gain from falling price due to more industry use. In another case, the
provider of the infrastructure may impose a congestion charge that will impose
an external DoS in addition to the internal costs imposed on each firm by the
congestion. But as the congestion price rises, the congestion cost to users will
fall if the price-disincentive is effective. On balance, the total cost to users may
rise or fall depending on the elasticity of use with respect to the congestion
charge. Hence there may be a range of EoS, of DoS and of CoS as overall use
and eventually the congestion charge rise. Shared infrastructure without any
further analysis is cited as a source of EoS by several of our authors.48
2.8.4 A Local Talent Pool?

Several writers list the development of a local talent pool as conferring an
EoS.49 It is difficult to determine that this is a source of EoS until one knows
whose actions developed it and at what cost.
48
These are Sloman (2006: 133), Abraham-Frois (2008: 232) and Graaff (1987: 7599).
49
These are Farrell (1997: 434), Graaff (1987: 7599), Griffiths and Wall (2000: 173), Rutherford
(2000: 142), Sloman (2006: 133) and Stockfisch (1968: 272).
2.9 Other Miscellaneous Causes

For completeness we give the few remaining causes that are mentioned by at
least one of the authors that we have surveyed.
2.9.1 By-Products
Sloman (2006: 132) and Ammer and Ammer (1984: 415) cite the production of
by-products as a source of scale economies. Although they do not specifically
link this to indivisibilities, which they list separately, if there is a minimum size
of operation for the production of any by-product, ex ante indivisibility, then as
the firm’s main activity is increased, it will at some point become profitable to
produce by-products rather than treating their materials as wastes. The authors
list this as a source of EoS for the firm.
2.9.2 Tax Relief?
It is difficult in the absence of much more information to see how tax relief
confers an external scale effect. It is, however, cited by Abraham-Frois (2008:
232) as a source of EoS.
2.9.3 Learning by Doing?

A few authors cite learning by doing as a source of both EoS and IRTS,
given fixed input prices.50 It is hard to see this as a genuine scale
economy. Rather it seems to be best regarded as a lag in the adjustment
of costs to changes in output. If the scale of output is increased and new
divisions of labour, new more specialised capital or new production
processes are introduced, there will be a lag before labour has learned
how to use the new processes with their full potential efficiency, a lag
documented in many cases by empirical learning curves. But if costs per
unit of output fall at this higher scale of production, it is not due to
learning by doing but by the increased productivity inherent in the new
processes, while the time taken to learn how to master them is a lag but
not a scale effect.
50
Specifically, McConnell, et al. (2012: 156), Barbosa-Filho (2008: 606) and Setterfield (2001:
489). The last is the only one to provide a reason, which he states as follows: ‘Experience can
be accumulated simply by repeating a task at the same level of throughput. However, an
increase in the number of “doers” . . . would also increase the stock of experience within a firm
and may thus be associated with a more than proportional expansion of output’ (489). There is
no apparent reason for this assertion. Most learning-by-doing is associated with learning how
to work efficiently with new technologies.
3 Constant and Variable Returns with Replication

We saw in Section 1.6.3 that where replication is possible, a firm’s costs can be
constant but only for integer multiples of the MES of the plant that defines the
minimum point of the long-run envelope cost curve. Between those integer
multiples the firm typically first encounters a range of DoS followed by a range
of EoS.
3.1 Replication with Non-Plant Specific Costs

We also saw in that section that where the firm has significant costs that do not
vary with the number of plants, it will encounter EoS over what can be an
indefinitely large number of plants and hence firm size. (Here we confine
ourselves to integer multiples of output of the optimum-sized plant.)
In many cases closely related to growth, the R&D costs of developing new
products are large relative to the cost of manufacturing them. Consider a new
product, say some new drug. Its high R&D costs were X and the MES for
production of the new drug in each of its plants is at output q1 and unit
production cost c1. Let the demand be such that n plants are fully employed
in manufacturing the product. Now the unit cost for each plant, including its
share of R&D costs, will be c1 + X/nq1, which is declining in n. For many firms
these are not just once-for-all entry costs. Instead, a steady flow of new
products is needed to maintain profits. In some, such as aircraft, if new products
are not developed, the firm will find its sales falling, eventually to zero. In
others, such as pharmaceuticals, the sale of new products during the lifetime of
their patents is key to maintaining high profits. Since once patents expire,
competition from generic forms of the product typically reduce profits greatly,
a continual flow of new products is needed to maintain high profits. In these
cases where R&D costs are large relative to production costs, average total
costs fall dramatically as output and sales rise.
This type of effect also covers many retail industries. Modern fast food,
garment retailers and motels are examples of the many industries that are
replete with chains in which more or less identical ‘plants’ are operated,
sometimes by the main firm and sometimes by independent franchisers, but
all of whom gain EoS effects from a variety of centralised operations. For
example, until the mid-1950s, most fast-food outlets were independently
owned and operated. Then came the chains, first McDonald’s and then others
such as Burger King, Wendy’s and A&W. They offered standardised menus and
quality, both of which were highly variable in the earlier days. This increased
the customer appeal of their products. Also several costs such as brand name
advertising, centralised purchases and menu selection guided by consumer
research, offered EoS that extended to an indefinitely large number of branches.

Analogous comments apply to retailers such as Urban Outfitters and American
Eagle, where centralised decisions regarding purchases, plant design and lay-
out, sometimes in-house design of their product, and other related activities,
confer EoS effects. Similar comments also apply to such hotel and motel chains
as Holiday Inn, Marriot, Travelodge and Best Western.
In all of these cases, where replication of production facilities is possible,
production at integer multiples of the MES of one plant will encounter an
indefinite range of EoS. The size of all such firms is not limited by each
reaching their MESs, but either by market demand if it is a monopolist or,
more typically, by its ability to compete with other large firms operating similar
but differentiated competing enterprises and all with unexploited EoS because
of costs that can be spread over any number of plants.
Although several authors state that the US market is large enough for scale
effects to be exhausted, this may be true of some industries, but is clearly not true for
all the industries, including those mentioned above, in which non-plant specific
costs are significant enough to confer an EoS on the firm as it increases the number
of its ‘plants’ indefinitely. As noted, the size of such firms is not limited from the
cost side but by competition from other firms with similar unlimited EoS.
3.2 Replication and the Constant-Returns Production Function

Many authors use the possibility of replicating plants, each producing at its MES,
to imply the existence of a constant-returns PF for outputs larger than that
produced by a single plant with costs such as SRATC5 in Figure 1.51 But this is
not a valid implication. Let such a plant be producing at its cost-minimising point
of y1 and using input quantities x1 and x2. All that replication implies is that, where
λ is an integer, the points (λx1, λx2) in input space, all of which lie the ray through
the origin of slope x1/x2, will be associated with outputs λy1. To assume the
existence of isoquants that pass through each of these points and are asymptotic
to the axes, as is implied by a constant-returns PF, is to make the wholly unjustified
assumption that the output y1 that each duplicated plant would produce at its
minimum-cost point can also be produced with technical efficiency by an array of
different techniques that use the factors in all possible proportions. In other words,
constant returns to one particular factor combination does not imply constant
returns to other combinations, although it does imply a ‘scalloped’ long-run
average cost curve with the same minimum point at the output that is an integer
multiple of the outputs of the most efficient plant, as shown in Figure 2.
51
For example, Black et al. (2012: 111), Case et al. (2012: 199), Eatwell (2008: 140) and Varian
(1992: 15).
For example, it is unlikely that any fast-food outlet could operate effectively,
if at all, at either extreme of almost all labour and little capital or (at least today)
of almost all capital and almost no labour.
4 Sources of DoS
Unit costs can rise with the scale of output for reasons located both within the
firm and at a higher level of aggregation. Causes within the firm are related to
individual pieces of capital equipment designed to do some job, the whole plant
and the firm’s organisation. Causes external to the firm occur at a more
aggregated level and include such things as the costs of congestion and pollu-
tion. All of these can be sources of DoS and DRTD.
4.1 Internal Effects

4.1.1 Capital Equipment
There are many cases of diminishing returns to size in an individual piece of
capital equipment due to physical laws. For example, if all the dimensions of a
bridge are altered in the proportion λ, its structural strength is altered by 1/λ and
its weight is altered by λ3 (under the simplifying assumption that it is optimal to
use the same types of materials in bridges of all sizes) (Adams 1991: 81). In
other words, bridges and other similar structures, exhibit diminishing returns to
design in the sense that as their size and the amount of materials used in their
construction is increased, their strength decreases per unit of input. More
generally, structural strength of any three-dimensional body tends to diminish
as its dimensions are increased, ceteris paribus. Thus, a small-scale model can
carry a heavier load expressed as a proportion of its size and weight, than can
the scaled-up version. Also, a small model airplane can tolerate a hard landing
that would destroy a larger version that had been scaled up proportionately. In
many cases, resistance to turbulence arising from the motion of a body through
a gas or a liquid increases more than in proportion to the increase in the
dimensions of the body. As already discussed, many characteristics of ships
exhibit increasing returns to size, but the fact that handling characteristics
become increasingly sluggish as size is increased beyond some critical value
is one of the factors that limit the economies of size that can be exploited.
As discussed earlier in this Element, many kinds of capital equipment are
made of components that work together. Although one component may encoun-
ter increasing returns, if another cooperating component has decreasing returns,
the whole typically has a range of IRTD followed by a range of DRTD. The
smallest workable size is seldom the most efficient size. As size increases, most
characteristics encounter favourable scale effects over some range. However,
sooner or later many characteristics encounter decreasing returns which even-

tually dominate so that further increases in capacity now result in rising costs per
unit of capacity delivered. The optimal size of a specific capital good is then the
one at which the economies of scale in some aspects of the technology just
balance the diseconomies in other aspects. The case of providing a flow of air to
smelting ore discussed in more detail in Section 5 is a case in point. If we
consider the whole history of this relation, we see unit costs falling over time
as better and better technologies for injecting air into furnaces are developed. But
the example also illustrates how increasing returns and decreasing returns to
various parts of one composite technology can combine to produce U-shaped
LRAC curves at any one time as the existing technology of providing air to the
smelting ore limits the economic size of the smelter. Also as shown earlier in
Carlaw (2004) and Carlaw and Lipsey (2008b: 223), the variable durability of the
individual components of a specific capital good, combined with uncertainty, can
lead first to a range of what we call IRTD followed by a range of DRTD, but
which these writers call IRTS and DRTS.
These relations apply to many situations and are some of the most impor-
tant sources of diminishing returns to design that limit the extent to which
other sources of increasing returns can be exploited by building larger ver-
sions of some generic capital good – once again both geometrical properties
and physical laws contribute to these relations.
4.1.2 The Plant
Similarly, to individual capital goods, individual production facilities typically

have an efficient size beyond which cost per unit of output begins to increase.
This is partly determined by the scale effects of the individual pieces of capital
equipment that are incorporated into the plant as well as by more macro
characteristics of the plant itself. For example, there is an upper limit to the
speed at which any production facility can be operated efficiently. There are
many factors that determine the most efficient speed, such as the frictions
created as machinery, including robots, are speeded up. But that there is an
upper limit can be easily seen by the reductio ad absurdum argument that if an
assembly line operated at the speed of sound an intolerable sonic boom would
be created.
4.1.3 The Firm
In the literature on scale effects, it is commonly argued that because of the

possibility of replication, one should never expect to see diminishing returns to
scale at the firm level. We have already considered increasing costs in the
interval between integer multiples of the most efficient plant. Here we consider
variations in unit costs at integer multiples of the output of the most efficient
sized plant.
Writing in the New Palgrave, Eatwell argues without recourse to empirical
evidence that ‘barring indivisibilities, there can be no barrier to replication . . .
In other words, there can be no such things as decreasing returns to scale’
(2008: 166, Italics in the original). In a similar vein Silvestre (1987: 2897)
states ‘an exact clone of the production process that exhaustively lists all
factors of production should give exactly the same output. The failure to double
the output suggests the presence of an extra input, not listed among the
arguments of the PF that cannot be duplicated.’ This replication argument
against decreasing returns to a firm is also used by Black et al. (2012: 111)
and Frank (2008: 280–1) who repeat the argument that DRTS are a result of
unspecified inputs being held constant as production is increased so that there
are no genuine DRTS. Similarly, Varian (1992: 16) holds that DRTS due a fixed
input is ‘the most natural case’ of DRTS – although he does not go on to
consider any ‘less natural cases’. Finally, Cowell (2006: 126) rules out DRTS
after defining them away with an axiom of additivity that applies unless ‘certain
essential features of the firm are non-expandable’. This last qualification might
cover some of the considerations given below, although it is not precise enough
to allow us to be sure what the author had in mind.
These arguments against decreasing returns to scale assume that replication
is always possible in any relevant production process, as long as there are
neither input indivisibilities nor hidden non-variable inputs. The qualifications
just listed raise the empirical question: ‘Under what specific circumstances is
replication possible?’ But this is an issue that cannot be settled by a priori
reasoning alone. Of course, if the list of possible inputs is defined as anything
that might cause the PF to display decreasing returns without detailed specifi-
cation of these exceptions, the proposition becomes tautological and hence
uninteresting empirically. As is well known, propositions about real behaviour
cannot be deduced from definitional identities.
To go further requires an appeal to empirical evidence. Here we can only
illustrate some of the empirical possibilities. On the one hand, to produce more
razor blades, a new plant identical to existing ones can be set up in a green field
and managed independently. This should yield CRTS and CoS for integer
multiples of the output of the most efficient plant’s MES. On the other hand,
if more output is required at a point in space, or less stringently, if spatial
location matters in any significant way, it may be impossible to replicate
exactly. For example, Eaton and Lipsey have argued in number of publications
(see e.g. Eaton &Lipsey 1977, 1997) that when space is introduced into the
neoclassical model, many new considerations become important and many

common generalisations no longer hold. Lipsey, Carlaw and Bekar (2005:
400) give another example, this one concerning the draining of early coal
mines where it was necessary to deliver power at the pit head to operate
pumps. Until the steam engine was invented, pumps were operated by horses
turning capstans and although there is no limit to the amount of energy that can
be obtained from horses if there is room for them to operate, there were
physical limits to the number that could be applied to any one pithead. Also,
long before that absolute limit was reached, costs per horsepower increased due
to such problems as non-linear increases in difficulties of coordinating the
operation of horses, both when at work and when changing shifts.
As is true in so much of the economics literature concerning the firm, the
replication argument applies most commonly to manufacturing firms. However,
these constitute a relatively small and declining portion of GDP in most advanced
economies. In contrast, firms in the primary producing sectors cannot easily
replicate. Those in mining and forestry find each production location unique in
many relevant characteristics and, although possibly less so, this is also important
with farmers. Similar considerations apply to retailers. Every location in a city is
unique and although some are more or less similar to some others, it is rarely
possible to duplicate exactly all of the relevant characteristics of two different
locations. Also, many high tech firms find that the externalities associated with
clustering of related firms are important in determining costs so that costs rise
with distance from the centre of the cluster.
Thus, replication is possible in some circumstances, where no worse than
constant returns is predicted for integer multiples of the smallest plant that has
the lowest minimum-cost point, and not in others where the DRTS may apply.
Where replication is possible is an empirical matter that cannot be settled by a
priori reasoning.
Carlaw and Lipsey (2008b: 223) are alone among the authors surveyed in
referring to inputs that are not reproducible due to spatial relations as a source of
DoS. Mansfield (1979) mentions increasing geographical size of a market as
scale increases, presumably leading to increased costs of both delivered inputs
and outputs.
Two other alleged sources of DoS for firms are frequently mentioned in the
literature but might work in either direction of increasing or diminishing unit
costs as scale rises. The first is organisational complexity (managerial complex-
ity, bureaucratic complexity). In the absence of compelling empirical evidence,
there seems no reason in today’s world of automated and robotised production
to assume a systematic positive relation between scale and the complexity of
organisation, nor a negative one between the latter and unit costs. Neither, of
course, is there any strong reason to deny that they might be causes of non-
negative scale effects under circumstances that remain to be specified. In the
literature surveyed, these complexities were listed as sources of DoS by many
authors.52 Others list them as causes of DRTS.53 Some list them as both DoS
and DRTS without qualification,54 while list them as both with the qualification
that input prices be constant.55
The second commonly alleged source is labour conflict (worker motivation,
etc.). Once again, in the absence of compelling evidence there seems no reason
in today’s world of automated and robotised production to assume a systematic
positive relation between scale and labour conflict. Small firms can have labour
problems just as can medium and large firms. Nonetheless, various authors list
them either as sources of DoS or DRTS.56
Finally, one author, Rutherford, mentions materials fatigue as a source of
both DoS and DRTS, providing that input prices are fixed, although there is no
apparent reason for assuming that materials are more prone to fatigue the larger
the scale of production
4.2 External Effects

The first two sources listed here are genuine scale effects not already discussed.
4.2.1 Increasing Factor Prices

The classic case in economic theory is the Hecksher-Ohlin two-industry GE
model where both industries have constant-returns PFs that use the two factors
in different proportions. Factor prices change as one firm expands and the other
contracts in a general equilibrium context for the whole economy, causing the
cost of output to rise in one industry and fall in the other. Of course there will be
less general cases as when the expansion of one industry increases its demand
for an industry-specific input. These are obviously not DRTS as the source is
not in the PF of the final goods producer. Instead they are pecuniary DoS
52
These are Ammer and Ammer (1984: 415), Bain (1968: 492), Brush (1994: 339), Farrell (1997:
433), Jackson (1996: 230), Mankiw (2006: 274), Sloman (2006: 133), Bannock et al. (2003:
100), Abraham-Frois (2008: 232), Black et al. (2012: 111, 123), Rutherford (2000: 123),
McAffee (2006: 4–101) and Shim and Siegal (1995: 106).
53
These are Case et al. (2012: 200), Perloff (2012: 172), Pindyck and Rubinfeld (2009: 215) and
Mansfield (1979: 143).
54
These are Krugman and Wells (2009: 323), Seldon and Penance (1976: 120, 295) and Pass et al.
(2005: 134).
55
These are McConnell et al. (2012: 156) and Parkin et al. (2005: 215).
56
Black et al. (2012: 111, 123), Brush (1994: 339) and McAffee (2006: 4–101) list these as sources
of DoS, while Case et al. (2012: 200) and Perloff (2012: 172) say they are sources of DRTS,
while McConnell et al. (2012: 156) says they are sources of both with the qualification that input
prices be constant.
effects. Such increasing prices may also result from a real diseconomy if firms
supplying the inputs face rising real unit costs and hence must raise the price of
the input they supply to a final goods producer.57
4.2.2 Congestion and Pollution

These are typically external to the firm and often to the industry. But they can be
internal to either as when a single firm in a company town increases its output
causing increased pollution that has a non-linear effect on its real costs. When
pollution alters the firm’s PF by causing the same inputs to produce less output, we
have a case that is not strictly any of DRTS, DoS or DRTD. If congestion raises a
firm’s real delivery costs, and if we define output to include delivery to markets,
we get DRTS even with replication as the one cost rises more than in proportion to
all the others. Or this may be a case of DoS as when congestion raises the prices of
the firm’s inputs. The majority of our authors who cited one or both of these as
sources classify them as causing DoS, while some others select either DRTS or
both.58
4.2.3 Other Listed Sources

A few other minor alleged sources are listed by the authors in our sample.
Transportation distances are mentioned by Bannock et al. (2003: 114) who
makes them a source of DoS. However, these are an example of the importance
of geographical factors previously emphasised in our discussion of internal
DRTS. The marginal cost of attracting new customers is listed by Rutherford
(2000: 123) as a source of both DoS and DRTS, assuming that input prices are
constant. This has already been discussed above in the section on advertising
where we argue that it is internal to the firm and may be the cause of EoS or
DoS. Finally, infrastructure deterioration is listed by Abraham-Frois (2008:
232) as a source of DoS. If infrastructure is maintained there is no issue here; if
it is not, then it is hard to see this neglect as being positively associated with
scale. This is typical of the ad hoc suggestions that are so often provided in
many of the cases cited above that might go either way, or more likely, not be
associated systematically with increases in output.
57
These price effects are listed by Kamerschen and Valentine (1977: 236), Bannock et al. (1984:
142), Sloman (2006: 134), McAffee (2006: 4–102) and Shim and Siegal (1995: 106) as sources
of DoS.
58
Abraham-Frois (2008: 232), Bannock et al. (2003: 100, 1984: 127), Stockfisch (1968: 269) and
Bohm (2008: 189) list both of these as sources of DoS, while congestion alone is listed by Pass
et al. (2005: 188) as a source of both DRTS and DoS, by Frevert (1997: 1376) as a source of
DRTS and Farrell (1997: 434) as a source of DoS.
5 Conditions Leading to the Exploitation

of Scale Effects
The main topic of this Element is the sources of scale effects. However, the
discussion of this issue suggests some obvious points on the conditions favour-
ing the exploitation of such effects, which we now mention with no pretence of
giving an exhaustive survey of the issue.
5.1 Market Size

Ever since Adam Smith; discussion of specialisation in the pin factory, it has
been accepted that size of market is an important determinant of how many of
the latent scale economies are actually exploited. This was also the thrust of
Allyn Young’s essay already mentioned. Although he did not use this distinc-
tion, both indivisibilities and design effects are important in this context. As
long as capital goods and specialised processes have an ex ante indivisibility,
scale effects associated with them will not be exploited if market demand is not
high enough to employ these production facilities at their minimum indivisible
size. Further scale economies associated with design economies (covering the
range from SRATC1 to SRATC5 in Figure 1) will only be exploited when
demand is sufficient to purchase the resulting output. Such expansions of
demand may be associated with the rise of some particular industry or with
general economic growth that increases the demand for all products with
positive income elasticities.
5.2 Historical Increasing Returns

Lipsey, Carlaw and Bekar (2005) stress the ubiquity of latent scale effects in
our world and observe that the ability to exploit these at any one point in time is
limited by existing technologies. Then as new technologies are developed, such
scale effects can be further exploited leading to temporary, but often prolonged,
periods of increasing returns while this is being done. Since these are disequi-
librium effects, they will not be captured by static models in which all transitory
effects have been eliminated.
One of the examples they cite concerns the smelting of ore where the heat
loss is proportional to the surface area of the ore’s receptacle while the service
rendered is proportional to its cubic capacity, thus creating a scale economy in
size. But to work efficiently, air must be infused into the whole body of ore
being smelted. The ability to exploit the physical economies arising from the
relation between surface area and volume was limited through historical time
by the ability to deliver the needed air as the size of the receptacle was
increased. For any given technology of delivering this air, it becomes harder
and harder to deliver an even flow of air to all parts of the receptacle as its size
was increased, yielding eventually decreasing returns to the size of any smelt-
ing process using a given air-injection method. Early ore was smelted in open
fires and such smelting was efficient at only a very small scale. Later it was
placed in a kiln and hand operated bellows delivered the needed air. When blast
furnaces replaced kilns, water-wheel-operated, and later yet, steam-engine-
operated, bellows greatly increased the efficient size of smelters. Then pressur-
ising, and also preheating the air to be injected, increased the efficient size even
further so that more of the ubiquitous scale economies related to surface area
and volume could be exploited.
There are many other important cases of such historical increasing returns.
They depend critically on the exploitation of scale effects that already exist in
nature but cannot currently be exploited because of limitations in complemen-
tary technologies. For another example, larger passenger aircraft tend to have
lower material inputs in construction and lower costs in operation per passenger
mile than smaller aircraft. But the ability to build larger aircraft has evolved
over time as the technologies of both design and materials have been steadily
improved.
5.3 Policy Changes

In the above cases new innovations were required to exploit the scale econo-
mies. Cases where growth of markets provided the opportunity for the exploi-
tation of scale effects without the need for the invention of new technologies
are clearest where barriers to trade have been reduced through public policy.
Before the European Common Market was formed economists using the then
standard trade theory in which each country produced an array of well-defined
homogeneous products had predicted that various countries would specialise in
different products – e.g., one in cars, one in furniture, one in TVs, etc. Because
the full significance of product differentiation has not been appreciated, the
outcome was quite different than expected. Different countries specialised in
different versions of each generic commodity. Scale economies were exploited
as countries reduced the number of versions of each generic commodity that it
produced, increasing the volume of output in the versions in which it specia-
lised and reaping associated scale economies. Similar effects were observed
when Canada and the United States formed a free trade area. These gains from
scale economies were of course additional to the gains from the reallocation of
resources according to local comparative advantage.
The sources of these scale effects can be studied in static treatments of
economies that exist either exploited or unexploited at any one time. There is
nothing in this static approach that is inconsistent with Lipsey, Carlaw and
Bekar’s (2005) studies of evolving economies. In most of this Element we have
followed the majority or writers in concentrating on the sources of scale effects
while accepting that two of the forces that allow these effects to be exploited
over time are the growth of the economy and the development of new technol-
ogies. All we are doing is distinguishing between the sources of scale effects
that exist independently of time, as studied by the literature on scale effects, and
the conditions that allow their exploitation to alter over time, as studied in
economic history and economic theory.
5.4 Which Came First: New Technologies or Growth

in Market Growth?
Growth in market size and new scale-exploiting technologies interact in a
system of positive feedback, although it is possible to argue sometimes that
one was at least the proximate cause of the other. We first consider two
important historical cases where a general increase in market size was an
important cause of growth in the now-advanced countries. But each of these
expansions in market size were themselves caused by a technological advance
that was not itself primarily a response to general economic growth. The first
was the great expansion of European trade that followed the development of the
three-masted sailing ship and the subsequent spread of European trade over
much of the world (as discussed, for example, by Mokyr (1990)). The three-
masted sailing ship, with square sails on the two forward masts, a lateen sail on
the after mast and triangular sails on an extended bowsprit, evolved as the
Portuguese slowly felt their way down the coast of Africa looking for a sea
route to India. It was not previous economic growth that motivated these
explorations but, among other things, the difficulties associated with
European access to the land route to the East. Once fully developed, this new
type of ship allowed merchants to sail the world over in search of valuable
cargoes with a reasonable chance of returning home. This initiated a bout of
what Mokyr calls Smithian growth – economic growth that results from a rapid
growth in trade. As it developed through the sixteenth century, it exploited
many scale economies in transportation, distribution and administration, going
on to become an early example of truly globalised trade.
The second case concerns the development of American transcontinental
railways, which fused many local markets into one large national market. The
railroad was the natural application of the steam engine. The high pressure
engine developed at the beginning of the nineteenth century had applications
for non-stationary power delivery which were apparent to those concerned with
both land and sea transport. The railway lowered transport costs while raising
the reliability and speed of overland transportation. These developments were
particularly important in the vast expanses of North America compared with
the relatively small size of the United Kingdom where the engine was first
developed. By welding the US market into a single whole for many products, it
lead to the exploitation of an array of scale economies and regional specialisa-
tions according to comparative advantage, both of which raised productivity
and output, leading to economic growth (see Chandler 1990: 53–8).
Some more general evidence suggests that developments in technology not
directly induced by economic growth were also the proximate cause in many
other cases. This comes from the relative growth rates of technologies that have
driven growth and those that are being more or less dragged along by it.
Although overall economic growth is responsible for the steady rise in market
size measured in real purchasing power units, most of the increases in the
markets associated with many of the new technologies that at one time or
another were leaders in growth, such as the various electronic devices that have
been developed in Silicon Valley and other similar agglomerations of innova-
tive talent, have been high relative to the general growth in market size of the
economies in which they have occurred. In such cases, overall economic
growth seems to be more a consequence of the arrival of these new technolo-
gies and the new markets that they often created, rather than these new
technologies being enabled by a growth in the overall size of the national
market.
A case where growth was arguably the proximate cause of technological
development that then contributed to further growth is the steam engine. As
Britain’s economic growth proceeded in the seventeenth and eighteenth centu-
ries, the demand for coal increased and coal mines went deeper underground
where flooding became increasingly important. The demand for removing water
led to a series of inventions that culminated in Newcomen’s atmospheric engine
early in the eighteenth century. This was a clear case of general growth providing
the incentive for technological advance. But the incentives for the subsequent
turning of the atmospheric engine into a genuine steam-driven engine undertaken
by James Watt is not so clear. Once developed the steam engine’s use spread
through the economy in the late eighteenth and early nineteenth centuries,
contributing to economic growth particularly when combined with new innova-
tions in textile production.
Although much can be said about these historical events and many others
that are similar to them, this is not an Element on economic history. There are
two conceptually distinct sequences. In one, economic growth influences the
growth of markets, which in turn influences the technological change that
underlies shifts in firms’ long-run cost curves when firms exploit scale effects.
In the other technological changes allow the exploitation of scale economies
that contribute to bursts of rapid growth. There is little doubt that both of these
interact in positive feedback loops and that the new technologies, whether they
were the cause or the consequence of economic growth, often succeed in
exploiting untapped scale economies.
6 Consequences of Exploiting Scale Effects

There is an extensive literature on the consequences of scale effects, which is
beyond the scope of this Element. Here we only mention one or two key points,
again with no pretence of giving an exhaustive survey of the issue.
Many of the early writers on development economics emphasised the impor-
tance of scale effects that could be exploited when economic development got
underway. In his masterly essay ‘The Rise and Fall of Development Economics’,
Paul Krugman notes the importance given to scale effects by writers in the tradition
of Albert Hirschman (1958). He argues that because they could not, or would not,
model economies subject to these scale effects, their writings fell into obscurity,
ignored by economists whose models almost exclusively assumed constant returns.
He goes on to show how Murphy et al. (1989) restored interest in this approach by
modelling in ways relevant to development economics, economies subject to scale
effects.
Another writer whose important works on the effects of scale effects were
ignored for years while constant returns was the order of the day’s modelling
was Brian Arthur, many of whose writings on this issue are reprinted in a
collection of essays (Arthur 1994).
In his seminal article incorporating endogenous technological change into a
macro growth model, Paul Romer (1990) assumed universal scale economies
associated with all such changes, a phenomenon that subsequent researchers
were unable to establish conclusively using empirical evidence. Later, Carlaw
and Lipsey (2011) separated scale effects and endogenous technological change
by developing a growth model in which endogenously generated general-pur-
pose technologies produced sustained growth without any non-constant scale
effects. What has not been done as yet is to build a model of endogenous growth
in which scale effects are present but constrained in the amount by which they
can be exploited by the forces discussed in this Element.
7 Conclusion and Summary

In conclusion, we summarise some of our most important results.
The world in which we live, with its three dimensions, physical laws and
random components of much of its behaviour, is replete with non-linear rela-
tions that give rise to scale effects. But exploitation of these scale effects is
limited both by the extent of the market and by decreasing returns to some of
the components of capital goods and production facilities such as the reduction
in structural strength of many bodies as dimensions are increased proportion-
ally. Thus, if we consider alternative sizes of a machine designed to do a
particular job, or of a whole plant designed to produce a particular product,
the unit costs of whatever is being produced will typically fall over a range of
outputs starting from zero up at to some critical value, after which it will begin
to rise – in other words the long-run average total cost curve for the production
of a product by a plant (or machine) will be U-shaped. For small changes in the
scale of output, firms may use a larger version of what is basically the same
production process, but for large changes wholly new production techniques
will be adopted, requiring different kinds of equipment and a differently trained
labour force, things that are difficult, if not impossible, to capture within the
confines of a single production function. The falling unit costs associated with
these effects are called economies of design in this Element.
Also, the operation of multiplying all physical inputs by some constant has few
real-world counterparts. For this reason, we replace the textbook definition of
returns to scale with the concept of returns to design: what happens to costs when
output is varied over the long run using the lowest available cost of producing each
output.
Although indivisibilities are commonly cited in the literature as a major
source of scale effects, the various meanings that can be, and commonly are,
given to the term are not typically distinguished. All capital goods with
differentiated parts are ex post indivisible in the sense that parts of them will
typically produce nothing. This universal characteristic of such capital goods
cannot, therefore, be used to explain why firms encounter different scale effects
at different scales of output, although some writers have assumed that they
could.
Most, but not all capital goods, plants and individual pieces of equipment,
have minimum sizes at which they are ex ante indivisible, no smaller size will
do the job. Ex ante indivisibilities of plants cause unit costs to be negatively
associated with output only when it is efficient to use the smallest plant that will
produce the product. Ex ante indivisibilities of pieces of capital equipment
contribute to a negative relation of unit costs and output only when the capital
good cannot be made smaller even though the current level of production will
not allow the smallest feasible version to be employed up to its full capacity. If
individual pieces of capital only came in one size, the plant that uses them
would have to operate the output that was the lowest common multiple of the
capacities of its individual capital goods so as to achieve the lowest possible
unit cost. But if capital goods can be made in various sizes down to some
minimum possible size that will do its job, the plants cost’s will fall as output
rises for this reason until it reaches the output of the smallest possible version of
the machine with the largest minimum capacity and other machines (whose ex
ante indivisibility is at a lower capacity) will be designed to have the same
capacity.
Some capital goods, such as pipelines, have no ex ante indivisibilities, being
producible at any size needed to do any specific job. Thus, the design econo-
mies that they confer have nothing to do with any (non-existent) indivisibilities.
Each firm has the option of reconfiguring its production processes or dupli-
cating the most efficient one for integer multiples of the output of its MES.
Where exact duplication is possible, the firm should not have to accept dis-
economies of scale for integer multiples of the outputs at their MES. But
although microtheory takes place on the head of a pin, actual production
takes place in space, and where spatial conditions matter, exact duplication is
often not possible so that diseconomies may have to be accepted.
The firm that is duplicating production facilities may still encounter econo-
mies of scale, particularly when there are large costs that are specific to the firm
but not to the individual plant and do not depend on the scale of output. The size
of such firms is then not typically limited by cost considerations but by their
ability to compete with similar firms each having a large number of more or less
identical ‘plants’ (such as fast-food outlets, clothing stores and hotels).
The above sources of economies and diseconomies are real in the sense that
they depend on variations in the amount of inputs needed per unit of output.
However, it is virtually impossible to analyse them using a single PF defined as
it usually is to relate physical inputs to output. If we wish to have inputs that are
used at all scales of output, it is necessary to use indexes of a few broad classes
of inputs. Even then, the PF will not fully span the resulting input space since
many of its points will be technically inefficient – although they give the
maximum output that can be produced by the indicated inputs, that output
could be produced with a different input ratio that uses less of all inputs. For
example, for very small outputs, craft methods are often technically efficient
while it is technically inefficient to use complex machines whose ex ante
indivisibilities occur at high levels of output, while for large scales of output
craft methods that use much labour and simple forms of capital equipment are
often technically inefficient.
The typical firm also encounters pecuniary economies of scale as the prices
of its inputs change with its scale of operations. Some of these are the result of
real upstream changes as design economies are reaped by firms operating at

these earlier stages of production, as more efficient machines can be hired or as
quantity discounts result from lower real costs of supplying inputs in bulk.
Others are purely pecuniary as when the firm is able to force down the price of
its inputs as a result of growing market power in input markets.
The exploitation of the above scale economies is both a cause and a con-
sequence of growth in evolving economies. Three of the major changes that
allow the vast range of latent scale economies to be exploited are (1) general
economic growth that increases demand for products whose production
encounters favourable scale effects; (2) the development of new technologies
that remove limits that prevent existing technologies from further exploiting
their latent scale effects; and (3) policy changes that enlarge market size.
Sometimes the problems and opportunities associated with economic growth
help to induce technological changes that then exploit favourable scale effects,
while at other times the new technologies are developed for reasons unrelated
to growth but then became causes of growth as they exploit latent scale effects.
It would appear that all of these things can be adequately studied using the
traditional concepts of scale effects, once returns to scale are replaced by
returns to design (thus avoiding the artificial exercise of multiplying all physi-
cal inputs by the same constant). Although these effects contribute to dynamic
changes in the economy, comparative statics exercises analyzing how a firm’s
and an industry’s physical inputs and its costs vary as its output undergoes a
single once-for-all change can provide all that as needed for understanding how
scale effects influence the ongoing growth of an economy. No novel new
definition of scale effects involving continual dynamic changes seems to be
needed.
Appendix A
Table of Definitions
Scale effects: Anything that affects the firm’s real or money unit costs as a
result of changing its scale of operations.
Real resource effect: There is a change in the amount of inputs per unit of
output.
Pecuniary effect: There is a change in the price of one or more of the firm’s
inputs with no related reduction in real resources used per unit of output.
Efficiency effects: Variations in unit costs as production is varied over the
long run. Also called efficiencies of scale.
EoS: Unit costs fall as output is expanded over the long run.
DoS: Unit costs rise as output is expanded over the long run.
CoS: Unit costs remain constant as output is expanded over the long
run.
Efficiencies of design: Variations in unit costs as production is varied over
the long run due to the replacement of one PF-1 by another. Similar to
‘efficiency effects’ except that it refers explicitly to the replacement of
one PF by another.
EoD: The change in PF needed to accomplish an increase in the scale
of output leads to a reduction in unit costs.
DoD: The change in PF needed to accomplish an increase in the scale
of output leads to an increase in unit costs.
CoD: The change in PF needed to accomplish an increase in the scale
of output leaves unit costs unchanged.
Returns effects: Variations in physical output resulting from variations in
inputs over the long run. Also called returns to scale.
IRTS: An equi-proportionate change in all inputs results in a more
than proportionate change in the output.
CRTS: An equi-proportionate change in all inputs results in the same
proportionate change in the output.
DRTS: An equi-proportionate change in all inputs results in a less than
proportionate change in the output.
Replication of a capital good: Creating more units identical to those already in
use.
Reconfiguration of a capital good: Using differently designed capital goods.
Type-1 production function (PF-1): Gives the maximum output that can be
produced by each given bundle of inputs on the assumption that the firm is
using a specific technology of production.
Type-2 production function (PF-2): Gives the maximum output that can be
produced by each given bundle of inputs on the assumption that the firm can
utilise any technology of production that is currently available.
Ex post divisibility and indivisibility: Refers to altering an individual capital
good or plant once it has been produced.
Ex ante divisibility and indivisibility: Refers to altering the size of a capital
good or whole plant, making a new one that is larger or smaller than the
original one but that can do the same type of job.
ncrs-indivisibilities (for non-constant returns to scale indivisibilities): A

production process is divisible in this sense if it can be scaled upward or
downward by multiplying all of the inputs in its production function by
some positive constant, λ, and have its output change in the same propor-
tion. The process is defined to be indivisible in the downward direction if
the result is to alter output by some multiple, γ, where 0 < γ < λ.
mps-indivisibility (for minimum possible size): A plant is defined as ex ante
mps-divisible if a smaller version can be made to do the same type of job
as the larger version. The smallest size plant that can produce the product
in question is ex ante mps-indivisible at that size.
Appendix B
Problems with Production Functions
There are several problems associated with either or both type-1 and type-2
production functions. To start we note some typical definitions of the PF drawn
from the literature.
B1. Definitions
If all available inputs are used, at any time, then there will be a maximum
attainable amount of product (output) that can be produced by each and every
combination of inputs – that is, the maximum amount of product that can be
attained from any specific combination of inputs that existing technology
(know-how) permits – is called the production function.
Feiwel and Feiwel (1997: 1229)
The firm’s production function f(z1, z2) shows the maximum output which can be
produced from the input combination (z1, z2)
Gravelle and Rees (1992: 180)
Given values for all the inputs and values for all but one output, the production
function specifies the maximum attainable value for the remaining output.
Henderson (1994: 811)
The production function for a firm shows the maximum output which can be
produced with specific levels of inputs, given the available technology.
McAuliffe (1999: 165)
The production function shows only the maximum amount of output that can be
produced from given levels of labor and capital, because the production function
includes only efficient production processes.
Perloff (2012: 155, italics added)
In constructing this [production] function, all methods, techniques, or processes

that require more of one input and no less of any other input are rejected.
Smith (1968: 512)
B2. Alternative Factor Combinations at a Given Output

Consider the typical textbook shape of a two-input PF’s isoquant such as shown
by I1 in Figure 4. The highly capital-intensive techniques used at the left-hand
end of this isoquant must be very different from the highly labour-intensive
techniques used at the right-hand end of I1. Students are not typically invited to
think about such things, and, if they do, they probably conceive of a putty-style
I’ I2
I1
E1
E2
K
E
E3
k2
E4
a3 a4
a2
a1 I2
k1 I’1
I1
0 l2 l1
L
Figure 4 Factor inputs with four different production technologies
capital that can be spread evenly and more or less thickly over the labour force.
To get closer to reality while still stylising, assume that there are four produc-
tion techniques whose most efficient K/L ratios at some given relative price of
the two inputs are given by the slopes of the lines E1 for robotised, E2 for mass
produced, E3 for artisan-style with a high degree of specialisation of labour
using specialised tools and E4 for artisan-style with a low degree of specialisa-
tion of labour using simple tools. If the level of output given by this isoquant is
very large, it is highly unlikely that techniques 3 or 4 could produce efficiently
at that scale, and possibly not even technique 2. Assume that this is so. Now the
amounts of K and L needed to produce the output indicated by a1 using each of
the three non-robotised techniques are, for example, respectively, a2, a3 and a4,
all of which are technically inefficient for that output. So technique 1 will be
used for that level of output whatever is the relative price of K and L.
Some readers have found it counter-intuitive that the best combination for
techniques that use the ratios E2 to E4 could be technically inefficient at some
output levels. To see the conditions for this let a1 in Figure 4 represent an output
of y3 produced with inputs k3 and l3. Let the most efficient sized plant that uses
the inputs in the ratio indicated by E4 have the much smaller output output y4
produced by inputs k4 and l4. Processes using the ratio indicated by E4 will be
technically inefficient to the process using the ratio E1 whenever k4/k3, l4/l3 >
y4/y3. For example, if the output of the most efficient E4-type plant is 10 per
cent that of the output at a1 while the capital and labour inputs are respectively
12 per cent and 20 per cent of those at a1, the 10 duplicated E4-type plants
needed to produce the same output as a1 will use respectively 120 per cent and
200 per cent of the capital and labour used at a1.
We now see a problem with the common definitions of the PF that were
illustrated above. If we insist on saying that it gives the maximum that can be
produced from every possible combination of the specified inputs, as do most
definitions of the PF, we must include points a2, a3 and a4 and others like them.
This makes the defined isoquant in the case illustrated positively sloped over
most of its range. Or if we follow Smith, or the Italicised portion of Perloff’s
definition, the isoquant for the production associated with input combination a1
spans only a small part of the input space, covering just the input combinations
that can be produced efficiently by a robotised plant. Some substitution
between labour and capital may be possible within that basic technology. But
the scope for this will be small in any robotised plant as shown by the isoquant
labelled I′1. That line becomes horizontal not far to the right of a1 (the point l1,
k1) indicating that no further efficient substitutions of labour for capital are
possible in robotised production, and vertical not far to the left of a1 (the point
l2, k2), indicating a minimum amount of labour that is needed for any existing
robotised production technique. So without too much violence to the facts, we
can assume that the ratio is fixed, giving a Leontief-style isoquant as shown by
I2. So the isoquant I1 has degenerated to the point a1, or if some substitution
between K and L is possible in the robotised plant, the segment of curve I′1 that
lies between the points k1, l1 and k2, l2.
So, if we define the PF as indicating only technically efficient factor combi-
nations, most of the factor space in this example will be empty: there is no
technically efficient way in which to use the factor amounts indicated by these
points. For high levels of outputs, most of the technically efficient points will
lie close to E1 because there is no technically efficient way to produce large
outputs of many goods and services using a large amount of labour and a very
few simple tools. Indeed, in many cases there will be no way to do it, even
inefficiently, if the good’s production requires some complex machines that
cannot be dispensed with. For example, silicon chips could not be produced by
simple craft forms of production using much labour and only simple tools. In
contrast for low levels of output, the efficient points will lie close to E3 and E4
because there is no technically efficient way to produce a small output of most
manufactured goods using only a small amount of labour and a large amount of
complex capital goods, most of which would have to lie idle most of the time
because they had reached the minimum size at which they became mps-
indivisible. In the absence of detailed technical information, there is nothing to

rule out the possibility of such blank spaces in most production functions.
Some theoretical treatments make formal assumptions that ensure that the
isoquants are asymptotic to the axes. This implicitly assumes that the array of
existing technologies is such that every possible ratio of inputs can be make
technically efficient at every possible level of output. Of course, students are
only asked to follow the formal treatment of such assumptions and seldom if
ever asked to consider what impossible things these imply about behaviour in
the real world.
B3. Expansion Paths at Given Factor Prices

Now consider Figure 5 where we assume that the K/L ratio is constant at all
levels of output for each production process at the relative price shown by the
slope of the equal outlay SS curves. Assume that for any location in factor space
between the lines S1 and S2 the outputs associated with the actual or near-
Leontief-style isoquants for the four processes that are ‘tangent’ to the price
line stand in the order x(E1) > x(E2) > x(E3) > x(E4). As total output is
decreased, lowering the S line, it will reach a position where the ‘tangent’
indicates a higher isoquant for process 2 than 1, then for 3 over than 2 and
finally 4 than for 3. As this occurs, each more capital-intensive production
process will be abandoned for a less capital intensive one. Each time this
happens, the K/L ratio will change abruptly to that indicated by the slope of
E1
K
S1
E2
S2 x7
E3
x6
S3
x5
S4 E4
x4
x3
x2
y2
x1
y1 S4 S3 S2 S1
0
L
Figure 5 Expansion paths with four different production technologies
E2, then E3 and finally E4. The expansion path for this relative price will thus
follow the heavy broken line starting from 0 and going through all the xs from
x1 to x7.
B4. PFs with Replication and Resulting

Constant Returns to Scale
Many treatments argue that a constant-returns PF, such as a Cobb-Douglas PF,
can be used when firms are large enough to have more than one replicated
production facility. Assume for illustration that the optimum-sized plant has an
output associated with the inputs shown by x6. All that replication assures is
that the outputs associated with points that are integer multiples of the coordi-
nates of x6, and hence all lie on E1, will increase by the same multiples. Nothing
about replication allows us to assume that there exist other technologies that
will allow those higher and higher outputs to be produced with technical
efficiency (if at all) by radically different factor combinations, such as much
labour and a little capital.
B5. Type-1 and Type-2 Production Functions

If we wish to study what is happening within the firm at each scale of its output,
there is no disagreement that we need separate short-run PFs for each produc-
tion process:
y1 ¼ f 1 ðK 1 ; L1 Þ; y2 ¼ f 2 ðK 2 ; L2 Þ; y3 ¼ f 3 ðK 3 ; L3 Þ;
and y4 ¼ f 4 ðK 4 ; L4 :Þ (A-1)
where each fi refers to the production facility that it is optimal to use when the
process of that number is being employed, the yis refer to the outputs of y
produced by the corresponding production facility, the Ks are the vector of
capital good that is fixed for any short-run PF while the Ls are the vector of
inputs that are variable. The physical bundles of fixed inputs, K , and variable
inputs, L, may be, and often are, different for each different production process
used at each different scale of production. There will be a range of outputs at
which it will be efficient to use each production technique as illustrated in
Figures 4 and 5.
The negative slope of the left-hand section of any U-shaped LRAC curve
depends on the different designs of the production facilities as the scale of
operations changes, i.e. to the different short-run PFs and different elements in
K and L.
Given the information in (A-1), if we wish to define a single type-2 PF that
covers all scales of output, we might use the form
y ¼ max½f 1 ðK 1 ; L1 Þ; f 2 ðK 2 ; L2 Þ; . . . f n ðK n ; Ln Þ: (A-2)
This will define the most output that can be produced with every possible
technically efficient combination of inputs. Once again there will be many
input combinations that are technically inefficient and so would never be
employed. For some levels of output more than one technique may be
technically efficient leaving the choice between them to depend on relative
factor prices, while at other levels there may be only one technically
efficient technique, leaving the input ratio to be determined by relative
prices only if there is room for factor substitution in that production
process. It is not obvious, however, how the information given by (A-2)
could be compressed into a single PF equation of the sort found in the
textbooks in which each input only occurs once and a single function
defines all of the technically efficient possibilities that use very different
production technologies.
B6. Conclusion
We conclude that the typical textbook representation of the PF has some serious
problems. When it contains isoquants that span all, or most, of the input space,
it is implicitly assuming that a given level of production can be obtained with
technical efficiency using a wide range of input ratios and hence a wide variety
of production techniques. This cannot be shown by a single type-1 PF. With a
type-2 PF, the requirement is that different production techniques, such as mass
and artisan production are both technically efficient for each level of output.
Simple observation suggests that there are many levels of output for which both
technologies cannot be technically efficient. For example, above a relatively
low level of output, capital-intensive mass production techniques technically
dominate labour-intensive craft production techniques at all relative input
prices. Furthermore, since many products, such as computers and electron
microscopes, require complex capital goods for their production, there is no
way that they can be produced by much labour and only a few simple tools. At
the opposite extreme, until robots become much more efficient than they are
today, there are many service products such as haircuts and high-end restaurant
meals that cannot be produced by a little labour and an array of complex capital
goods. When these impossibilities are taken into account, there is no reason to
expect any single isoquant to span anything close to the whole possible range of
the input ratios, nor to expect the expansion path for given relative prices to be
continuous.
This is not the place to debate how much harm is done by the fiction that the
PF typically has the form given it in the textbooks. But surely it is not a bad
thing to suggest that students (and their teachers) should be aware of the reality
that they are abstracting from when they use the textbook definition of a type-2
PF and that, in some applications, the resulting fictions may be misleading.
Classified References
A. Dictionaries, Encyclopaedias and Introductory Textbooks1
Abraham-Frois, G. 2008. ‘Average and Marginal Cost’. In William A. Darity
Jr., ed., International Encyclopedia of the Social Sciences (2nd Edition).
Detroit: Macmillan Reference USA.
Ammer, C. and D.S. Ammer. 1984. Dictionary of Business and Economics
(Revised and Expanded Edition). London: The Free Press.
Bain, J.S. 1968. ‘Economies of Scale’. In D.L. Stills, ed., International
Encyclopedia of the Social Sciences. New York, NY: Gale.
Bannock, G., R. Baxter and R. Davies. 1984. Penguin Dictionary of Economics
(3rd Edition). Harmondsworth, Middlesex; New York, NY: Penguin
Books.
2003. Penguin Dictionary of Economics (7th Edition). London; New York,
NY: Penguin Books.
Barbosa-Filho, N.H. 2008. ‘Verdoorn’s Law’. In William A. Darity Jr., ed.,
International Encyclopedia of the Social Sciences (2nd Edition). Detroit:
Macmillan Reference USA.
Baumol, W.J. 2008. ‘Indivisibilities’. In S.N. Durlaf and L. Blume, eds., The
New Palgrave Dictionary of Economics (2nd Edition). New York, NY:
Palgrave Macmillan.
Becattinni, G. 2008. ‘Internal Economies’. In S.N. Durlaf and L. Blume, eds.,
The New Palgrave Dictionary of Economics (2nd Edition). New York,
NY: Palgrave Macmillan.
Becker, G. 1999. ‘Pecuniary Economies’. In R.E. McAuliffe, ed., The
Blackwell Encyclopedic Dictionary of Managerial Economics. Oxford:
John Wiley and Sons, Inc.
Black, J., N. Hashimzade and G. Myles. 2012. A Dictionary of Economics (4th
Edition). Oxford: Oxford University Press.
Bohm, P. 2008. ‘External Economies’. In S.N. Durlaf and L. Blume, eds., The
Palgrave Macmillan.
Brush, B.C. 1994. ‘Economies of Scale’. In D. Greenwald, ed., The Mcgraw-
Hill Encyclopedia of Economics (2nd Edition). New York, NY: McGraw-
Hill, Inc.
1
I have grouped introductory texts under A rather than B with the rest of the textbooks because
their converge is less formal than the micro texts, and hence more like those in the other
categories under A.
82 Classified References
Calhoun, C. 2002. Dictionary of the Social Sciences. Oxford: Oxford

University Press.
Carlaw, K. and Richard G. Lipsey. 2008a. ‘Returns to Scale’. In William A.
Darity Jr., ed., International Encyclopedia of the Social Sciences (2nd
Edition). Detroit: Macmillan Reference USA.
2008b. ‘Returns to Scale, Asymmetric’. In William A Darity Jr., ed.,
International Encyclopedia of the Social Sciences (2nd Edition).
Detroit: Macmillan Reference USA.
Case, K.E., R.C. Fair and S.M. Oster. 2012. Principles of Economics (10th
Edition). New York, NY: Prentice-Hall.
Cowell, F.A. 2006. Microeconomics: Principles and Analysis. Oxford: Oxford
University Press.
Eatwell, J. 2008. ‘Returns to Scale’. In S.N. Durlaf and L. Blume, eds., The
Palgrave Macmillan.
Feiwel, G.R. and I. Feiwel. 1997. ‘Production and Cost Functions’. in F.N.
Magill, ed., International Encyclopedia of Economics. Chicago: Salem
Press Inc.
Frank, R.H. 2008. Microeconomics and Behavior (7th Edition). New York, NY:
McGraw-Hill/Irwin.
Frank, R.H. and B.S. Bernake. 2009. Principles of Microeconomics (4th
Edition). New York, NY: McGraw-Hill/Irwin.
Farrell, M.J. 1997. ‘Economies and Diseconomies of Scale’. In F.N. Magill, ed.,
International Encyclopedia of Economics. Chicago: Salem Press Inc.
Frevert, P. 1997. ‘Returns to Scale’. In F.N. Magill, ed., International
Encyclopedia of Economics. Chicago: Salem Press Inc.
Graaff, J.d.V. 1987. ‘Pecuniary and Non-pecuniary Economies’. In J. Eatwell,
M. Milgate, and P. Newman, eds., The New Palgrave Dictionary of
Economics (1st Edition). New York, NY: Palgrave Macmillan.
Gravelle, H. and R. Rees. 1992. Microeconomics (2nd Edition). London, UK:
Longman.
Henderson, J.M. 1994. ‘Production Function’. In D. Greenwald, ed., The
Mcgraw-Hill Encyclopedia of Economics (2nd Edition). New York, NY:
McGraw-Hill, Inc.
Jackson, D. 1996. ‘Economies of Scale’. In A. Kuper and J. Kuper, eds., The
Social Science Encyclopedia (2nd Edition). New York, NY: Routledge.
Jorgenson, D. 2008. ‘Production Functions’. In S.N. Durlauf and L.E. Blume
eds., The New Palgrave Dictionary of Economics (2nd Edition). New
York, NY: Palgrave Macmillan.
Krugman, P. and R. Wells. 2009. Microeconomics (2nd Edition). New York,

NY: Worth Publishers.
Mankiw, N.G. 2006. Principles of Microeconomics (4th Edition). Mason, OH:
Thomson South-Western.
McAffee, R.P. 2006. Introduction to Economic Analysis. Open-source,
retrieved from www.mcafee.cc/Introecon/IEA2007.pdf.
McAuliffe, R.E. 1999. ‘Production Functions’. In R.E. McAuliffe, ed., The
Blackwell Encyclopedic Dictionary of Managerial Economics. Oxford:
John Wiley and Sons, Inc.
McConnell, C.R., S.L Brue and S.M. Flynn. 2012. Microeconomics:
Principles, Problems, and Policies (19th Edition). New York, NY:
McGraw-Hill Irwin.
Parkin, M., M. Powell and K. Mathews. 2005. Economics (6th Edition). Essex:
Pearson Education Ltd.
Pass, C., B. Lowes and L. Davies. 2005. Collins Dictionary of Economics.
London: Collins.
Pearce, D.W. 1992. The MIT Dictionary of Modern Economics (4th Edition).
Cambridge, MA: MIT Press.
Perloff, J.M. 2012. Microeconomics (6th Edition). Boston, MA: Pearson
Education Inc.
Pindyck, R.S. and D.L. Rubinfeld. 2009. Microeconomics (7th Edition). Upper
Saddle River, NJ: Pearson Prentice Hall.
Pratten, C. 2004. ‘Economies of Scale’. In Adam Kuper, ed., Social Science
Encyclopedia (3rd Edition). New York, NY: Routledge.
Rutherford, D. 2000. Routledge Dictionary of Economics. Cornwall: TJ
International Ltd.
Seldon, A. and F.G. Pennance. 1976. Everyman’s Dictionary of Economics.
London: J.M. Dent & Sons Ltd.
Setterfield, M. 2001. ‘Increasing Returns to Scale’. In P. O’Hara, ed.,
Encyclopedia of Political Economy. London; New York, NY: Routledge.
Shim, J.K. and J.G. Siegal. 1995. Wiley Dictionary of Economics. Hoboken,
NJ: John Wiley & Sons.
Silvestre, J. 1987. ‘Economies and Diseconomies of Scale’. In J. Eatwell, M.
Milgate and P. Newman, eds., The New Palgrave Dictionary of Economics
(1st Edition). New York, NY: Palgrave Macmillan.
Sloman, J. 2006. Economics (6th Edition). Essex: Pearson Education Ltd.
Smith, V.L. 1968. ‘Production’. In D.L. Sills, ed.., International Encyclopedia
of the Social Sciences. New York, NY: Gale.
Stockfisch, J.A. 1968. ‘External Economies and Diseconomies’. In D.L. Sills, ed.,
International Encyclopedia of the Social Sciences. New York, NY: Gale.
Varian, H.R. 1992. Microeconomic Analysis (3rd Edition). London; New York,
NY: W.W. Norton and Company Inc.
Vassilakas, S. 1987. ‘Increasing Returns to Scale’. In J. Eatwell, M. Milgate
and P. Newman, eds., The New Palgrave Dictionary of Economics (1st
Edition). New York, NY: Palgrave Macmillan.B.
B. Microeconomics Textbooks
Baumol, W.J., 1977. Economic Theory and Operations Analysis. Englewood
Cliffs, NJ: Prentice Hall.
Binger, B.R. and E. Hoffman. 1988. Microeconomics with Calculus. Glenview,
IL: Scott, Foresman and Company.
DeSerpa, A.C. 1985. Microeconomic Theory: Issues and Applications. Boston:
Allyn and Bacon, Inc.
Eaton, B.C., Eaton, D.F. and D.W. Allen. 2012. Microeconomics: Theory with
Applications (3rd Edition). Toronto: Pearson Canada.
Griffiths, A. and S. Wall. 2000. Intermediate Microeconomics: Theory and
Applications. Essex: Pearson Education Limited.
Hirshleifer, J., A. Glazer and D. Hirshleifer. 2005. Price Theory and
Applications: Decisions, Markets, and Information (7th Edition).
Cambridge; New York, NY: Cambridge University Press.
Jehle, G.A. and P.J. Reny. 2001. Advanced Microeconomic Theory (2nd
Edition). New York, NY: Addison Wesley.
Kamerschen, D.R. and L.M. Valentine. 1977. Intermediate Microeconomic
Theory. Cincinnati, OH: South-Western Publishing Co.
Mahanty, A.K. 1980. Intermediate Micro-Economics with Applications. New
York, NY: Academic Press.
Mansfield, E. 1979. Microeconomics: Theory and Applications (3rd Edition).
New York, NY: Norton.
Miller, R.L. 1978. Intermediate Microeconomics: Theory, Issues, and
Applications. McGraw-Hill Inc.
Nicholson, W. 1979. Intermediate Microeconomics and Its Applications.
Hinsdale, IL: The Dryden Press.
Quirk, J.P. 1987. Intermediate Microeconomics. Chicago: Science Research
Associated Inc.
Sher, W. and R. Pinola. 1986. Modern Microeconomic Theory. New York, NY:
Elsevier Science Publishing Co.
Shone, R. 1981. Applications in Intermediate Microeconomics. Oxford: Wiley.C.
C. Other References
Adams, James. 1991. Flying Buttresses, Entropy and O-Rings: The World of an
Engineer. Cambridge, MA: Harvard University Press.
Arthur, W. Brian. 1994. Increasing Returns and Path Dependence in the
Economy. Ann Arbor, MI: University of Michigan Press.
Bain, J.S. 1954. ‘Economies of Scale, Concentration, and the Condition of
Entry in Twenty Manufacturing Industries’. The American Economic
Review 44(1): 15–39.
Bain, J.S. 1956. ‘Barriers to Competition: Their Character and Consequences
in Manufacturing Industries’. Cambridge, MA: Harvard University Press.
Blaug, M. 1978. Economic Theory in Retrospect (3rd Edition). Cambridge;
New York, NY: Cambridge University Press.
Bobzin, Hagen. 1998. Indivisibilities: Microeconomic Theory with Respect to
Indivisible Goods and Factors. New York, NY: Physica-Verlag
Cardwell, D.S.L. 1995. The Norton History of Technology. New York, NY:
Norton.
1971. From Watt to Clausius: The Rise of Thermodynamics in the Early
Industrial Age. London: Heinemann.
Chandra, Ramesh and Roger Sandilands. 2006. ‘The Role of Pecuniary
External Economies and Economies of Scale in the Theory of
Increasing Returns’, Review of Political Economy 18 (2): 193–208.
Chandra, Ramesh and Roger Sandilands. 2009. ‘Reply ot Roy Grieve on
Increasing Returns’. Review of Political Economy 21(4): 655–64.
Carlaw, Kennith I. 2004. ‘Uncertainty and Complementarity Lead to Increasing
Returns to Durability’. Journal or Economic Behaviour and Organization
53(2): 261–82.
Carlaw, Kennith I. and Richard G. Lipsey. 2003. ‘Productivity, Technology and
Economic Growth: What Is the Relationship?’ Journal of Economic
Surveys 17(3): 457–95.
2004. ‘Total Factor Productivity and the Measurement of Technological
Change’. The Canadian Journal of Economics 31(4): 1118–50.
2011. ‘Sustained Endogenous Growth Driven by Structured and Evolving
General Purpose Technologies’. Journal of Evolutionary Economics 21
(4): 563–93.
Chandler Alfred, D. 1990. Scale and Scope: The Dynamics of Industrial
Capitalism. Cambridge, MA: Belknap Press.
Eaton, B. Curtis and Richard G. Lipsey. 1977. ‘The Introduction of Space into
the Neo-Classical Model of Value Theory’. In M.J. Artis and A.R. Nobay,
eds., Studies in Modern Economics. Oxford: Basil Blackwell.
1997. On the Foundations of Monopolistic Competition and Economic

Geography: The Selected Essays of B. Curtis Eaton and Richard G.
Lipsey. Cheltenham: Edward Elgar Publishing.
Grieve, Roy H. 2010. ‘Pecuniary External Economies, Economies of Scale and
Increasing Returns: A Note of Dissent’. Review of Political Economy
22(1): 127–40.
Hirschman, A. 1958. The Strategy of Economic Development. New Haven, CT:
Yale University Press.
Hiscock, Eric C. 1965. Cruising Under Sail. Oxford: Oxford University Press.
Kaldor, N. 1972. ‘On the Irrelevance of Equilibrium Economics’. The
Economic Journal 82(328): 1237–55.
Koopmans, T.C. 1957. Three Essays on the State of Economic Science. New
York, NY: McGraw Hill.
Krugman, Paul. ‘The Fall and Rise of Development Economics’. http://web
.mit.edu/krugman/www/dishpan.html.
Lipsey, Richard G. 2009. ‘Some Legacies of Robbins’ An Essay on the Nature
and Significance of Economic Science’. Economica (Special issue
Robbins’s Essay at 75) 76(1): 845–56.
Lipsey, Richard G., Kenneth Carlaw and Clifford Bekar, 2005. Economic
Transformations: General Purpose Technologies and Long-Term
Economic Growth. Oxford: Oxford University Press.
Morroni, M. (1992). Production Process and Technical Change. Cambridge:
Cambridge University Press.
Mokyr, Joel 1990. The Lever of Riches: Technological Creativity and
Economic Progress. New York, NY: Oxford University Press.
Murphy, R.A., A. Shleifer and R. Vishny. 1989. ‘Industrialisation and the Big
Push’. Journal of Political Economy 97(5): 1003–26.
Romer, Paul. 1990. ‘Endogenous Technological Change’. Journal of Political
Economy 98(5): S71–102.
Rosenberg, Nathan. 1982. Inside the Black Box: Technology and Economics.
Cambridge: Cambridge University Press.
Rosenberg, Nathan and L.K. Birdzell. 1986. How the West Grew Rich. New
York, NY: Basic Books.
Viner, Jacob. 1931. ‘Cost Curves and Supply Curves’. Reprinted in George
Stigler and Kenneth Boulding, eds., AEA Readings in Price Theory.
London: George Allen and Unwin.
Young, A. Allyn. 1928. ‘Increasing Returns and Economic Progress’. The
Economic Journal 38(152): 527–42.
Evolutionary Economics
John Foster
University of Queensland
John Foster is Emeritus Professor of Economics the University of Queensland, Brisbane. He
is Fellow of the Academy of Social Science in Australia; Life member of Clare Hall College,
Cambridge; and Past President of the International J.A. Schumpeter Society. He is also
Director of the Energy Economics and Management Group at UQ and Focal Leader for
Renewable Energy at the Global Change Institute.
Jason Potts
RMIT Univeristy
Jason Potts is Professor of Economics at RMIT University, Melbourne. He is also an Adjunct
Fellow at the Institute of Public Affairs. His research interests include technological change,
economics of innovation, and economics of cities. He was the winner of the 2000
International Joseph A. Schumpeter Prize and has published over 60 articles and six books.
About the series

The Cambridge Elements of Evolutionary Economics provides authoritative and
up-to-date reviews of core topics and recent developments in the field. It includes
state-of-the-art contributions on all areas in the field. The series is broadly concerned
with questions of dynamics and change, with a particular focus on processes of
entrepreneurship and innovation, industrial and institutional dynamics, and on
patterns of economic growth and development.
Evolutionary Economics
Elements in the series

A Reconsideration of the Theory of Non-Linear Scale Effects: The Sources of
Varying Returns to, and Economics of, Scale
Richard G. Lipsey
9781108555029
A full series listing is available at: www.cambridge.org/core/series/

elements-in-evolutionary-economics

A Reconsideration of The Theory of Non-Linear Scale Effects

Uploaded by

Copyright:

Available Formats

A Reconsideration of The Theory of Non-Linear Scale Effects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Reconsideration of The Theory of Non-Linear Scale Effects

Uploaded by

Copyright:

Available Formats

Downloaded from https://www.cambridge.org/core.

University of Delhi, on 16 Jan 2019 at 07:37:44, subject to the Cambridge Core

The Sources of Varying Returns to,

Cambridge University Press is part of the University of Cambridge.

The Sources of Varying Returns to, and Economies of, Scale

Elements in Evolutionary Economics

Abstract: The main thrust of this Element is a critical assessment of the

Keywords: economies of scale, production function, returns to scale,

© Richard G. Lipsey 2018

1 Background Conceptual Issues 2

2 The Sources of Positive Scale Eﬀects 22

3 Constant and Variable Returns with Replication 56

5 Conditions Leading to the Exploitation of Scale Eﬀects 64

6 Consequences of Exploiting Scale Eﬀects 68

7 Conclusion and Summary 68

We live in a three-dimensional world subject to both physical laws and uncer-

1 Background Conceptual Issues

1.2 Issues of Deﬁnition

An example of a purely pecuniary effect is when the growth of an assembly

1.3 Approaches to Studying Scale Eﬀects

Figure 1 Short- and long-run costs

Second, in the encyclopaedias, in most introductory textbooks and in some

1.4 Displaying Scale Eﬀects

1.4.1 Through Cost Curves

1.4.2 Through a Production Function

The analysis of scale effects in microtheory textbooks typically uses the

f ðλs1 ; . . . ; λsm Þ ¼ αλ f ðs1 . . . ; sm Þ; (1.2)

1.4.3 Through Growth

1.4.4 Replication and Reconﬁguration

1.4.5 Two Types of Production Functions

Others do so implicitly without saying it in so many words.8 Yet others just

different from those used in Fordist mass-production factory, which in turn

This critique of the returns to scale deﬁnition reveals what should be

Returns and Eﬃciency Eﬀect’s in the PF

Further Problems with the PF

1.5 Eﬃciencies of Design

to design changes such as changes in input prices as result of changes in market

1.6 Scale Eﬀects and Firm Size

1.6.1 Implications of MES Relative to Market Size

Marshall himself recognized the incompatibility of the assumption of com-

Of course, there is nothing incompatible with perfect competition for

Unit Cost LRAC

0 q1 2q1 3q1 4q1 Quantity

Figure 2 Variable scale effect with replication

1.6.2 Replication with Variable Unit Costs

Almost all of the surveyed authors mention replication as a sources of CoS

1.6.3 Replication with Non-Plant-Speciﬁc Fixed Costs

2 The Sources of Positive Scale Eﬀects

2.1 Scale Eﬀects in Two-Stage Production

Here we see scale effects:

which is increasing in F. To check, multiply F in equation (II.2) by λ to get:

P ¼ ðλF=4Þ2 ¼ λ2 ðF=4Þ2 : (II.4)

make a return of r% on its investment. So it must charge a price, y per unit of

Letting the constant summation term be C, we have

2.2 Pervasiveness of Scale Eﬀects

the Eaton-Lipsey paper is in their treatment of lumpiness and indivisibility as

2.3 Scale Eﬀects due to Indivisibilities

2.3.1 Ex post Divisibility and Indivisibility

2.3.2 Ex ante Divisibility and Indivisibility