|

Received: 11 August 2018    Accepted: 22 August 2018

DOI: 10.1111/nup.12225

ORIGINAL ARTICLE

Middle‐range theories as models: New criteria for analysis and

Mark Risjord1,2

1
Institute for the Liberal Arts, Emory
University, Atlanta, Georgia Abstract
2
Nell Hodgson Woodruff School of Nursing, This essay argues for a new perspective on the analysis and evaluation of middle‐
Druid Hills, Georgia
range theory. The commonly used criteria for theory evaluation in nursing are not as
Correspondence useful as they should be, and the root of the problem is an inappropriate understand‐
Mark Risjord, Institute for the Liberal Arts,
ing of middle‐range theory. In spite of their name, middle‐range theories should not
Emory University, Atlanta, GA.
Email: [email protected] be analysed and evaluated as concrete or limited versions of more general theories.
Rather, they are best understood as models. The latter sections of this essay present
a new method for analysing middle‐range theories and new criteria for their evalua‐
tion. Analysing and evaluating theories relevant to nursing as models has the capacity
to increase their scholarly and clinical usefulness.

1 |  TH E N E E D FO R N E W C R ITE R I A O F For example, consider empirical adequacy. It seems like an ob‐
TH EO RY E VA LUATI O N vious criterion for theory choice, since empirical support is the
hallmark of good science. It is sound advice, then, to direct scholars
Theories are central to scientific understanding because they to look for theories that are well supported by evidence. Fawcett’s
articulate an understanding that gets behind the observed pat‐ criterion of empirical adequacy has us ask “Are theoretical asser‐
terns and explains them. By doing so, theories permit us to see tions congruent with empirical evidence?” (Fawcett, 2005b, 132).
relationships among phenomena that might otherwise seem dis‐ “Congruent” is a vague word in this context. It might mean that the
connected. In addition, because they say something about the evidence agrees with the theory. So far so, so good: when evidence
underlying constitution or causes of the phenomena we are con‐ disagrees, it is a mark against a theory. However, “congruence” might
cerned to prevent or promote, theories help us develop effective also mean mere consistency or compatibility, and this is too weak
interventions. Hence, understanding what a given theory tells us to be a valuable criterion for theory choice. The theory that animal
about the way things work and how it might be clinically useful is species came into existence through an act of divine creation is con‐
crucial for nurse scholars, as well as for clinical nurses and other sistent with the evidence of the fossil record, as creationists often
healthcare providers. point out. The theory of evolution, on the other hand, is more than
When looking into a new phenomenon, then, scholars begin by consistent with the evidence; it is strongly supported by it. Hence,
casting about for an appropriate theory. Scholars ask: what makes while it is in the neighbourhood of an important criterion for theory
a theory good for my project? What is the best theory available? choice, Fawcett’s criterion provides very little purchase on the ac‐
To help emerging scholars and nurses in training develop tools for ceptability of a theory.
answering these questions, textbooks and essays outline criteria Parse’s version is worse than unhelpful, since her criterion
for the analysis and evaluation of theory. Indeed, the nursing lit‐ only asks “How is the theory used as a guide for research?” (Parse,
erature abounds in advice about how to evaluate theory (Fawcett, 2005, 136). A theory may guide research in important ways, yet
2005a, 2005b; Masters, 2015; Meleis, 2007; Parse, 2005; Peterson be completely wrong. In the nineteenth century, for example, evi‐
& Bredow, 2008; Tomey & Alligood, 1998). Unfortunately, the cur‐ dence mounted that light (and other forms of electromagnetic en‐
rently published criteria in the nursing literature run the gamut from ergy) travelled in waves. This led to the natural idea that there must
unhelpfully vague to downright dubious. They provide poor guidance be a medium through which the waves propagated, just as waves
for identifying useful, informative, accurate, or reliable theories. travel in water. This was the “aether theory” of light propagation.

Nursing Philosophy. 2018;e12225. wileyonlinelibrary.com/journal/nup © 2018 John Wiley & Sons Ltd  |  1 of 10
https://doi.org/10.1111/nup.12225
 |
2 of 10       RISJORD

The Michelson–Morley experiment was designed to measure the propositions have the form of laws, and the concepts take the form of
velocity with which the earth moved through this subtle medium. variables. Discussions of theory often distinguish between the axioms
Instead, it provided crucial evidence that the aether does not exist. or fundamental laws of a theory and the derived laws. Using these
The aether theory thus guided research, which is a point in its favour laws, the theory explains a range of observable phenomena and en‐
according to Parse’s criteria. The research it guided showed that the tails testable hypotheses. Scientific theories provide insight, on this
aether theory is false, which surely must be a point against it. The ca‐ view, because they identify the laws that underlie the patterns we
pacity to guide research says almost nothing about whether a theory observe in nature. They forge unexpected connections by showing
is well supported. Neither Fawcett’s nor Parse’s criteria for empirical that apparently different phenomena arise from the same laws. And
adequacy point the scholar towards the relevant question: whether by manipulating the variables of the laws, we can design effective
the theory is strongly or weakly supported by the existing evidence. interventions.
Turning to a different sort of example, currently published crite‐ Newton’s mechanics (theory of motion) is a paradigmatic scien‐
ria for theory evaluation typically include a large number of criteria tific theory, and the conception of theory espoused in nursing fits it
focusing the character of the theory’s propositions. Fawcett asks: “Is perfectly. Newton proposed three laws of motion that related the
the theory content stated clearly and concisely?”, “Do the concepts concepts of inertia, momentum, force, velocity, acceleration, and
reflect semantic clarity and semantic consistency?”, and “Do the prop‐ mass. From these laws, he deduced a number of important princi‐
ositions reflect structural consistency?” (Fawcett, 2005b, 132). Parse ples (e.g., the conservation of energy) and explained known regular‐
adds: “Are the descriptions of the assumptions, concepts, and princi‐ ities (the elliptical orbit of the planets around the sun). The theory
ples clearly written at an abstract level?”, “Is the word usage economi‐ showed how phenomena as different as projectile motion and the
cal?”, and “Is there a logical flow from the philosophical assumptions to tides arose from the same underlying laws. And on the practical side,
major concepts to principles?” (Parse, 2005, 136). Theories that satisfy it provided the scientific understanding that set the stage for the
these criteria might be easier for a novice to understand, but it does industrial revolution.
not follow that the theories will support research or practice. Many Seen against this conception of scientific theory, the some of the
scientists are, frankly, terrible writers. Their failure to compose elegant standard criteria for theory evaluation begin to make sense. The em‐
prose does not show that their theories are scientifically inadequate. phasis on parsimony, consistency, and logical derivability arise from
Moreover, the deepest and most important theories in science, such thinking of theories as axiom systems, where the fundamental laws
as Einstein’s theory of relativity or the theory of evolution, have often must entail lower‐level generalizations and testable hypotheses. The
taken generations to clarify. Economical word usage is simply irrele‐ insistence that good nursing theories be logically connected to the
vant to theory evaluation. concepts of the metaparadigm reflects the idea that the different
Examples could be multiplied, but these suffice to support the theories of a discipline should fit together into a single logical struc‐
conclusion that the currently published criteria are not as useful as ture deriving from the highest level laws.
they could or should be. Scholars searching the literature looking for a The understanding of scientific theory that underpins nurs‐
theory to use in research, especially scholars at the beginning of their ing’s criteria of theory evaluation is no longer popular among phi‐
career, must have some way of sorting wheat from chaff. Criteria of losophers of science. The problem is not so much that the account
theory evaluation must identify theories that provide deep insight, of theory is flat wrong; rather, the account is partial and thereby
make unexpected connections, and guide helpful interventions, and distorts our picture of scientific knowledge and research practice.
the criteria must distinguish theories with potential from those with‐ Philosophers have pointed out that the biological and social sci‐
out it. The standard criteria in the nursing literature do not do so. ences—two domains of inquiry that share important boundaries
with nursing research—have few, if any, theories that conform to
the ideal. Inquiry in these domains typically takes the form of very
2 |  TH EO R I E S A N D M O D E L S : W H AT I S A local theorizing. Models of specific phenomena are constructed and
TH EO RY A N Y WAY ? tested. Microeconomics and population ecology, for example, do not
exhibit the systematicity of Newton’s mechanics. Neither has a de‐
Useful criteria for theory evaluation can only arise from appropri‐ limitable set of laws to unify the field. They are better understood as
ate understanding of how theories work and why they provide scien‐ a collection of models that share certain idealizations and construc‐
tific understanding. The currently published criteria are based on a tion techniques.
conception of scientific theory that was articulated by prominent Indeed, even our understanding of Newton’s mechanics is dis‐
nurse scholars in the 1970s and is reiterated in nursing theory text‐ torted by the conception of theory common in nursing. Consider
books to this day.1 On this view, a theory is a set of propositions the Newtonian explanation of a billiard ball collision. 2 The analysis
relating two or more concepts. In some exemplary cases, the of the collision begins by deriving some principles from the theory,

1 2
In Nursing Knowledge (Risjord 2010), I explored the history and consequences of this con‐ Such explanations can be found in any elementary physics text. A nicely detailed, but
ception of theory—the so‐called “received view of theory”—in the nursing literature. accessible, version of the problem discussed here is found on the Real World Physics
Bender (2018) develops the argument that nursing theory is better understood as model Problems web page: https://www.real-world-physics-problems.com/physics-of-billiards.
building. html
RISJORD |
      3 of 10

crucially, the conservation of linear momentum. However, the the modelling relies on the conception of a utility function. While
analysis must proceed by specifying some details of the system, called a “theory,” expected utility theory is a very different animal than
e.g., that one ball travels with an initial velocity and direction, Newton’s theory of motion. The so‐called axioms of expected utility
striking another ball with an equal mass. That is, the analyst pro‐ theory are a set of assumptions that must be satisfied if the agents’
ceeds to build a model of the phenomenon to be explained. The utility functions are to be mathematically tractable. While Newtons
model makes a number of assumptions that are in no way required “axioms” are plausible laws of nature, the axioms of expected utility
by the theory: that the masses are exactly equal, that the collision theory are mathematical definitions. Most of the empirical, scientific
is perfectly elastic, that friction with the surface and air is negligi‐ work of microeconomics proceeds by drawing on the specifics of the
ble, that the balls do not spin, and so on. Given the model, the an‐ phenomenon to create models. And, again unlike the dynamical mod‐
alyst can show how the motion of the balls after the collision is els of Newton’s mechanics, microeconomic models draw on a wide
determined by the initial velocity and angle of impact. variety of sources to articulate the interactions of firms and consum‐
There is no well‐defined distinction between “theories” and ers. Many fields exhibit the same pattern: models are constructed by
“models,” and both scientific and philosophical usage is inconsistent. drawing on whatever generalizations are relevant to the phenomenon.
As used here, a (scientific) “theory” is a coherent set of empirically Ambitions of the Grand Theorists notwithstanding, nursing
grounded generalizations. The generalizations are, as the standard has no unifying theory analogous to Newtonian dynamics. Rather,
view holds, propositions relating two or more concepts, and the like the biological and social sciences, it has excelled where it has
concepts may take the form of variables. Theories are made coher‐ been able to devise localized understandings of specific phenom‐
ent by the way in which the propositions are related. This might be ena (Donaldson, 2000). That is to say, theories in nursing research
an axiomatic structure, like Newton’s mechanics, or the looser rela‐ have been much more like microeconomics or population ecology:
tionships of mutual support and explanation found in contemporary the construction of models guided by generalizations drawn from a
evolutionary theory. The generalizations of theory are empirically variety of sources. Using criteria for the evaluation of theory that are
grounded in the sense that they are supported (or supportable) by designed for theories like Newton’s mechanics is, therefore, bound
systematic observation and testing. to be misleading. Indeed, it may be the source of the chronic vague‐
Models, on the other hand, are representations of how and why ness in the nursing criteria for evaluation. Recognizing that the prop‐
a particular phenomenon arises. Where a theory is analogous to a ositions and concepts of theories useful to nursing did not function
narrative description, a model is analogous to a map. A map iden‐ in the same way as Newton’s laws, perhaps these writers softened
tifies number of salient features and shows how they are related. the edges. Nonetheless, the result was a square peg jammed into a
A model identifies a number of parts (with specific properties) and round hole.
shows how they interact to produce an effect. The phenomenon to
be explained by the Newtonian model, above, was the regular rela‐
tionship between the initial velocity and path of a billiard ball and the 3 | M I D D LE‐ R A N G E TH EO R I E S A S
post‐collision velocities and paths. To understand this phenomenon, MODELS
the model decomposed it into parts (the two balls) with their prop‐
erties (their velocities, masses, and elasticity) and it specified the in‐ To create useful criteria for theory evaluation in nursing, we need
teraction among these parts (the collision and its angle of incidence). to change our perspective on nursing theory. To understand something
The resulting model then shows how these initial conditions give rise as a theory (as opposed to a model) is to emphasize its representation
to those subsequent conditions. in propositions: typically laws and definitions. To understand some‐
Newtonian mechanics thus involves both a theory (the three thing as a model is to emphasize the way it represents different ele‐
laws of motion) and set of models constructed with its guidance. ments of a system working together to produce a larger phenomenon.
In the example of the billiard balls, the theory’s primary contribu‐ I suggest that everything treated as a middle‐range theory in nursing
tion was to generate the formula for the conservation of linear mo‐ can be better understood when analysed as a model (cf. Bender 2018).
mentum. This principle governs the interactions of the billiard balls This does not mean that every intellectual product useful to nursing
within the confines of the model’s assumptions. The model analyses scholarship or practice is a model. Identifying themes in interviews is
the phenomenon to be understood into parts, and it specifies the in‐ not an example of modelling, insofar as such research typically makes
teractions among them. The model shows why the phenomenon oc‐ no attempt to articulate how the themes relate or show how thinking
curs—why the billiard balls respond to a collision in such predictable of their world in terms of these themes leads the research subjects to
ways. The theory provides the resources for building the models; the act in particular ways. Neither is grand theory well represented in
models provide the scientific understanding. terms of modelling.3 Most of these theories make no attempt to ex‐
Newtonian mechanics exhibits just one of the forms that theory‐ plain how some specific phenomenon works. Middle‐range theories,
model relationships take in the sciences. In the biological and social
3
sciences modelling often proceeds with little or guidance from laws. This depends a bit on what theories are counted as “grand.” Benner’s theory of expertise
is sometimes regarded as a grand theory, but it is much better understood as a middle‐
Consider, for example, the field of microeconomics. There are no laws range theory, hence as a model of expertise. The Neuman Systems Model (Neuman 1982)
of microeconomics analogous to Newton’s laws of motion. Much of might also be represented and evaluated as a model.
 |
4 of 10       RISJORD

on the other hand, are intended to be “abstract enough to extend be‐ elements and relationships present in a phenomenon. It therefore
yond a given place, time, or population but specific enough and suffi‐ highlights some features of the phenomenon modelled at the ex‐
ciently close to empirical data to permit testing and generate distinctive pense of others. Thinking in terms of models thus highlights what
questions for study or specific interventions for practice” (Lenz, Suppe, is ignored by the model, and it lets us ask whether what is getting
Gift, Pugh, & Milligan, 1995, 2). In the social and biological sciences, ignored is important. Importance depends on our purposes. If you
this is achieved by modelling; the same is true for nursing. want to get across New York City with public transportation, a
To think of middle‐range theorizing as model building highlights subway map is better than a street map; if you want to find a new
the following six characteristics of middle‐range theories, and it restaurant, the street map is better. Thinking of middle‐range theo‐
shows why middle‐range theories can be useful for understanding ries as models, then, forges a tighter link between the adequacy of
nursing phenomena and devising nursing interventions. These char‐ the model and the purposes for which we need to use it.
acteristics are obscured by the standard advice concerning the anal‐ Fourth, a phenomenon may be modelled in multiple ways. Again,
ysis and evaluation of middle‐range theory. this is hard to see when we focus on the laws of a theory. Since laws
First, analysing middle‐range theories as models shifts the focus are universal generalizations, theories of the same phenomenon that
from the structure of the theory (the character of the propositions deploy different laws, such as Newton’s mechanics and Einstein’s
or concepts, their logical relations, etc.) to its content. By represent‐ relativistic mechanics, are inconsistent. For example, in Newtonian
ing the activities of entities within a system, a model shows how mechanics, mass is conserved in all interactions, while in relativity,
something works. Once we understand how something works, we mass can be converted to energy. It is natural then to think of one
can design interventions that fix it when broken. One might object as true and the other false. Models are more like maps, and there is
that this perspective on middle‐range theory is overly mechanistic. nothing odd about using both a street map and a subway map to find
However, while it is easy to imagine models of clocks or electrical one’s way. The reason that there can be multiple models of one phe‐
circuits, modelling is not limited to physical systems. There can be nomenon follows from the third point, above: all maps misrepresent
models of psychological systems and social relationships, as well as the territory by ignoring some features. Different maps thus repre‐
models that mix biological, psychological, and social elements. The sent different aspects of the same thing. Multiple models, based on
theory of planned behaviour (Ajzen, 1985, 1991, 2005 ), analysed in different assumptions and background theories, are often necessary
Section 4, is easily understood as a model. to make sense of complicated phenomena—just the sort of thing that
Second, while models are not all mechanistic (in the sense of concerns nurses.
having physical parts with mechanical relationships), models do em‐ Fifth, thinking of middle‐range theories as models highlights
phasize causality. A model shows how some phenomenon is brought the relevant empirical tests useful for demonstrating empirical ade‐
about through processes that involve the component entities and quacy. Relevant tests determine whether the elements actually work
their effects on each other. The possible interventions are shown together as the model demands. The billiard ball model described
immediately, since they would be some change in either the inter‐ above claims that varying the velocity and angle of impact will change
actions or the elements. While models are typically causal, they the subsequent velocities and directions of travel. We can thus di‐
need not be linear, simple, or mechanistic. Many systems modelled rectly test the model by changing the initial velocity and looking to
in the biological and social sciences involve feedback loops, self‐ see whether the subsequent velocities change in the way that the
replication, and emergent properties. One might object here that model predicts. Because models specify interactions, we can always
emphasizing causality is limiting: are not we after a holistic science ask whether changing one entity’s activity changes the others as de‐
of nursing? Is not causality reductionist? I would argue that, to the scribed by the model. If not, then the model is inaccurate. Notice that
contrary, the goal of nursing theory is to improve nursing practice. when the model fails such direct testing, it shows the model to be
“Improve” is a causal notion. While not all understanding is strictly inaccurate in a way we care about. If we have developed the model to
causal, in the absence of any ability to intervene, questions of how capture what is important about a phenomenon, we have identified
to improve must be moot. By emphasizing the causal relationships of the entities and activities relevant to our purposes. If empirical test‐
a system, thinking of middle‐range theories as models helps identify ing shows the model to be inaccurate, the model cannot be used to
those with the potential to improve the quality of nursing care. intervene in the way we want to. Thinking of middle‐range theories
Third, all models ignore some aspects of the system, and a focus as models, then, lets us evaluate their empirical support in ways that
on modelling helps illuminate what a middle‐range theory leaves are directly relevant to their suitability for nursing.
out. Any description or explanation must leave something out, but Finally, values are embedded in both models and theories.
it is harder to see when thinking in terms of theories. Consider again Analysing middle‐range theories as models makes the implicit
the example of Newtonian mechanics, above. If we focus on the law values visible. Because a model ignores some aspects of the real
F = ma, the question of what is left out is hard to understand. After world and emphasizes others, a model implicitly takes some things
all, law is universal, applying to all objects with mass. However, when to be more important (more valuable) than others. To adopt a
the dynamical model has been constructed, the question has a clear model is to commit to the importance and value of the aspects of
answer: the model described above ignores (abstracts away from) the world it highlights. Of course, precisely because all models are
friction and spin. Like a map, a model picks out only some of the partial, to recognize that a model abstracts from some aspects of
RISJORD |
      5 of 10

a phenomenon is not to judge that the ignored features have no things. Some may be attitudes, while others refer to actions; some
value. We can adopt a physiological model of pain, for instance, may be organ systems while others are social structures. Note that
without at the same time saying that the existential and emotional some of the parts may not be ordinary objects at all, but symbols,
dimensions of pain are unimportant or do not exist. Nonetheless, social roles, or institutions. Knowing what kinds of things are pos‐
by modelling only some aspects of a phenomenon, we do obscure ited by a model is crucial for understanding how the model works,
other aspects. Difficulties arise when those obscured aspects are because different kinds of thing have different capacities. Social
crucial to understanding and intervening. The question for the anal‐ roles, all by themselves, do not make pathogens more virulent, but
ysis and evaluation of models, then, is not whether there are val‐ social roles do influence behaviour, which in turn can have an in‐
ues encoded in the model; all models embed values of some kind. fluence on infection.
The question is whether those elements and relationships are the Finally, the model must postulate some sort of relationships
right ones to prioritize in the context of the research and practice in among the elements. The phenomenon is brought about and/or
which the model will be used. maintained through some interaction of the parts. These interactive
processes can take a number of forms, but the most common types
found in nursing and the health sciences are causal relationships,
4 | A N A LYS I S O F M I D D LE‐ R A N G E TH EO RY constitutive relationships, and stages.
In causal4 relationships, change in one element (or group of ele‐
To think of middle‐range theories as models is to turn attention ments) is either necessary or sufficient to bring about change in an‐
away from the definition of their concepts and the character of their other element. The model is telling us that if one thing changes, then
propositions. We are interested not in words, but in things; not in something else will change. Models with causal relationships often
sentences, but in interventions. This means that both the analysis have inputs and outputs, and the output is often the phenomenon of
and the evaluation of middle‐range theory will be different from the interest. In the model of the billiard ball collision, described above, all
standard view. of the relationships are causal, and the model shows how the later
Considered as a model, a middle‐range theory shows how the directions and velocities (outputs) are the product of the initial direc‐
phenomenon is brought about by analysing the phenomenon into tions and velocities (inputs). Many biological models, such as models
parts (or elements) and identifying their relationships (or interac‐ of disease transmission, are quite similar.
tions). This means that to understand a middle‐range theory—what it In constitutive relationships, the parts are arranged in a kind of hi‐
tells us, how it might be tested, what interventions it might support— erarchy. Something at a higher level is made up of things from a lower
we must identify: (a) the phenomenon modelled, (b) the elements of level. The higher level may be the phenomenon itself, but it may also
the model and their properties, and (c) the relationships it postulates. be another element of the model. Constitutive relationships are not a
The first step of analysis, then, is to identify the phenomenon matter of one event happening after another, as the post‐collision ve‐
modelled. We must ask: What is this theory about? Of what is it a locity of a billiard ball is caused by the pre‐collision velocity. Rather, it
model? Alternatively, we can ask this way: What is being explained is more like how the flour, eggs, butter, sugar, and baking powder com‐
by this model? Fortunately for the student, most middle‐range the‐ bine to constitute a cake. Making a cake is more than simply putting
ories are named after the large‐scale phenomena they explain: we the ingredients into the same bowl. The cake is not created until ingre‐
have middle‐range theories of Caring, Comfort, Unpleasant Symptoms, dients interact in the right way; gluten is created by the flour and water,
and so on. One must be careful, of course, since these short pithy the eggs emulsify the water and butter, the baking soda releases CO2,
phrases can be misleading. Benner’s theory (1984) is about exper‐ and so on. Therefore, to postulate a constitutive relationship among
tise, but is about expertise in a specific sense. It is not about the elements of the model is to say that the lower‐level elements interact
theoretical expertise of a witness called to testify in court about the in some way to bring about or constitute the higher level elements. Of
soundness of an engineering design. Rather, Benner gives us a model course, a model may not mention the details of the interaction, just as
for the formation of practical expertise: the confident, fluid, mastery a cake recipe abstracts from the chemistry of baking.
of a practice. To illustrate causal and constitutive relationships, and to help
The next step in the analysis of a middle‐range theory is to show how this sort of analysis can lead to insight about a theory, let
determine what the main parts (elements) of the model are. us analyse Ajzen’s well‐known theory of planned behaviour (Ajzen,
According to the model, of what cells, organ systems, persons, 1985, 1991, 2005 ). The phenomenon analysed by this model is, as
attitudes, appraisals, developmental stages, social statuses, in‐ the name suggests, planned behaviour. That is, it provides a model
stitutional roles, etc., is the phenomenon composed? These will for behaviour that is deliberate and intentional, as opposed to that
often correspond to the main concepts that the author presents, which is reflexive, spontaneous, or habitual. An augmented version
but again, we are not interested in definitions per se. What do the 4
The word “cause” has a notoriously broad use. There is an important sense in which all of
definitions identify? To understand what a model is telling us, it is the relationships mentioned in this essay are “causal.” To call this first sort of relationship
important to notice how the elements differ. In a field like nurs‐ “causal,” then, is not to suggest that the others are not causal. What are here identified as
“causal relations” are something like what Aristotle would have called “efficient causes.”
ing, there is no reason to suppose that the parts are all the same What are called “constitutive relations” are in the neighborhood of Aristotle’s material or
kind of thing. Elements may be biological, psychological, or social formal causes. Sometimes, stages are steps in a causal process.
 |
6 of 10       RISJORD

F I G U R E 1   Theory of planned behavior

of the standard diagram for the theory of planned behaviour has “Generally speaking, people intend to perform a behaviour when
been reproduced in Figure 1.5 On the diagram, the primary elements they evaluate it positively, when they experience social pressure to
are identified in the circles: Behaviour, Intention, (Behavioural) perform it, and when they believe that they have the means and op‐
Attitude, Subjective Norm, and (Perceived) Behavioural Control. portunities to do so” (Ajzen, 2005, 118). Because the Intention is a
Ajzen also associates a distinctive kind of belief with each of the ele‐ behavioural disposition, it makes sense for it to be causally brought
ments of Attitude, Subjective Norm, and Behavioural Control. These about by the evaluations of Attitude and Subjective Norm, and the
are indicated in the boxes with dashed lines. It is important to recog‐ beliefs of Behavioural Control.
nize that these elements are not all the same kind of things. Behaviour The relationship between the various kinds of belief (repre‐
is a bodily motion. The Intention is an enduring behavioural disposi‐ sented in the boxes with dashed lines) that support the elements
tion; it is the fact that with proper opportunities (and sufficient per‐ of Attitude, Subjective Norm, and Behavioural Control is somewhat
ceived control), the person will act (Ajzen, 2005, 99). Both Attitude less clear. While Ajzen sometimes describes this relationship in
and Subjective Norm are evaluations of the action, while Perceived terms of “determination,” many of his depictions are at odds with the
Behavioural Control is a belief (or set of beliefs). first‐this‐then‐that pattern of causality. He writes that the attitude
Turning to the relationships, Ajzen uses a variety of arrow styles is formed “automatically and simultaneously” when the relevant be‐
in his original diagram (which are faithfully reproduced in Figure 1). liefs are formed (Ajzen, 1991, 191; Ajzen, 2005, 29). Causal relation‐
As is sadly common in such diagrams, he does not provide a key. To ships are not simultaneous; causes precede effects. Additionally, he
understand the diagram—and this is true of all presentations of a treats each of these elements as proportional to a sum of beliefs
theory or model—we have to carefully consult the text. Doing so, it is and evaluations (Ajzen, 1991, 191, 195, 197; Ajzen, 2005, 124–125).
apparent that Ajzen intended for all of the relationships represented These remarks suggest that the relationship between the various
by arrows to be causal relationships. The arrows numbered (7) and kinds of beliefs and the Behavioural Attitude, Subjective Norm,
(8) are clearly causal: “According to the theory of planned behaviour, and Perceived Behavioural Control is constitutive. Each of the con‐
performance of a behaviour is the joint function of intentions and structs is at least partly constituted by the relevant sort of belief.
perceived behavioural control” (Ajzen, 1991, 185). The relationships Understanding them this way also helps make clear how the three
indicated by arrows (1) through (3) express the idea that the content elements can causally influence one another. If the Subjective Norm,
of one of the elements will influence the others, e.g., that one’s eval‐ for example, is partly constituted by normative beliefs, when these
uation of smoking a cigarette (Attitude) will be influenced by one’s beliefs change, the Subjective Norm must change. Changing be‐
perception of how others regard smoking (Subjective Norm). Finally, liefs in one area are likely to change beliefs in another. Hence, if the
the relationships indicated by arrows (4) through (6) are also causal: Subjective Norm changes, the Attitude is likely to change.
Ajzen’s model does not include stages, but many models in nurs‐
5
Figure 1 is based on the figures Ajzen often uses to present his theory (Ajzen 1991, 182; ing and the health sciences do. In an earlier era of medicine, much
2005, 118). It has been modified by the author.
RISJORD |
      7 of 10

of our knowledge about diseases consisted of nothing more than of interactions in which they might participate. Answers to the
the stages through which the disease would progress. Today, even three questions should form a coherent whole. Upon analysis,
where a more detailed understanding of the underlying mecha‐ it should be clear how the phenomenon modelled works (how it
nisms is available, identifying where a patient stands in the stages comes about, why it changes, etc.) in virtue of the elements and
of a disease can be valuable. Many biological, psychological, and their relationships.
social processes important to nursing can usefully be divided into
stages. Benner’s account of expertise (1984), as well as the Dreyfus
and Dreyfus (1980) model of skill acquisition on which it was based, 5 | E VA LUATI O N O F M I D D LE‐ R A N G E
are stage models. The elements of the models are stages, and the TH EO RY
relationship among the elements is both a temporal progression and
a development. Attaining a given level of expertise depends on at‐ Analysing a middle‐range theory as a model helps us focus on the
taining the prior level. particularly relevant dimensions of evaluation. The criteria can be
Stage models need to be analysed carefully, since the relation‐ grouped into pragmatic criteria and epistemic criteria. Pragmatic cri‐
ships among stages can take several forms, and the language of teria relate the model to the research project, the clinical context, and
“stages” can be misleading. The Kübler–Ross model (Kübler‐Ross, the researcher him or herself. Epistemic criteria evaluate the model
1969) is commonly understood to say that a person passes through as a piece of scientific knowledge. Recalling the analogy between
five stages of grief in sequence. In spite of her temporal language, models and maps, the pragmatic evaluation of a map would deter‐
however, “later” stages do not depend on the “prior” stages in any mine whether the map is good for me, given what I need to do. The
way, and a person can manifest more than one stage at the same epistemic evaluation would determine whether it was an accurate
time. Benner’s stages, by contrast, are temporal: one is a novice map. The analogy shows that the two kinds of criteria are related.
before being an advanced beginner. However, Benner’s model A map might be inaccurate in certain ways, but it might still be good
does not explain why a person moves from one stage to another, enough for one’s purposes; for other purposes, it might be useless.
or what changes underlie the development. The Dreyfus and A crucial error of much advice about theory evaluation is that
Dreyfus model is more articulate about how the stages are related. it supposes that theories can be evaluated individually, as if each
Dreyfus and Dreyfus postulate the existence of four “mental func‐ theory could be pronounced good or bad. The weight of argument
tions” that change as the learner develops expertise. Each stage in the history and philosophy of science contends that theory
thus clearly establishes the conditions for the next. evaluation is always a comparative matter. A group of scientists
It is very common for a theorist to encode relationships among may know that the theory they use is inadequate in important
elements in a diagram like Figure 1. Diagrams are a succinct way to ways—Kuhn called these “anomalies” (Kuhn, 1970)—but if there
communicate the elements and relationships of the model. However, is no better alternative, the theory endures. All of the criteria
the analyst of a middle‐range theory must approach diagrams with below should be understood as ultimately deployed in the service
care. It is common to relate different elements of the model with ar‐ of a comparative judgment. We want to know whether the the‐
rows, but the same arrows may represent causal relationships in one ory under discussion is the best of the available alternatives, and
part of the diagram and temporal stages in another. Relationships thereby whether it fares better or worse on these criteria than the
represented by overlap or inclusion (such as the ubiquitous con‐ competitors.
centric circle diagrams) are particularly difficult to parse. Some re‐
lationships may be captured by up/down, left/right or other spatial
5.1 | Pragmatic criteria of evaluation
relationships rather than by explicit elements of the diagram like ar‐
rows. Relationships might be implicit in the diagram and need to be The primary pragmatic question about a model is whether it is good
teased out of the text. for the research programme. This is the criterion of usefulness. All
From the perspective of modelling, then, the analysis of a mid‐ research should be considered as an attempt to answer a question.
dle‐range theory requires answering three questions: Before a model can be evaluated, then, one has to get clear about
what question one is trying to answer with it. While framing an inter‐
• What is the phenomenon modelled? esting, important, and answerable question is often the most diffi‐
• What are the elements of the model? cult part of research, we will assume here that the research question
• What are the relationships among the elements? has already been specified. The criterion of usefulness, then, asks
whether the model helps answer the research question. Now, the
Answers to these questions are interdependent, and there‐ answer to this is unlikely to be a flat “no.” You probably would not
fore a scholar who is analysing a model will not answer these in be analysing and evaluating the model unless there was some way
the above order. One’s initial impression about what the model that it related to your research. The more subtle question is how it
is explaining (the phenomenon) might have to be revised in light helps answer your question. It is likely that any given model will help
of what the model can explain, given the elements and processes in some ways and not in others. What role will it play in explaining
identified. Understanding what the parts are influences the sort phenomena of concern to you?
 |
8 of 10       RISJORD

Thinking about how a model helps answer your research ques‐ and reliable the model is: whether the postulated elements exist
tion gives rise to two further pragmatic criteria of evaluation. First, and whether they interact in the way proposed by the model. These
there are questions about abstraction. As argued above, all models are primarily issues of how testable the model is, and whether it
idealize, ignore, or abstract away from some features of the real has been successful when tested. The epistemic criteria also speak
world. The question of abstraction, then, is whether the model has to the usefulness of the model for devising interventions, since an
ignored something that is important, given the question you are accurate model will support more robust interventions than an in‐
trying to answer. If it does abstract away from an important aspect accurate one. And again, these criteria are always in service of a
of the phenomenon, then it is not so good for your research, at comparative judgment.
least not by itself. It will need to be supplemented by other mod‐ No model (or theory) can be tested for accuracy or reliability
els that capture the missing dimensions. As an example, consider unless it is possible to make observations that would show that the
again the theory of planned behaviour. According to it, the be‐ model is inaccurate. This means that the model must be operation‐
haviour occurs when the person has the intention and believes s/ alizable to some degree and sufficiently precise to allow for mean‐
he has the ability or control. However, there may be barriers that ingful tests. Operationalizing a model means finding something
the person does not recognize or acknowledge. The model explic‐ observable that correlates with (often unobservable) elements and
itly ignores any objective barriers to implementing the intention. their properties. The theory of planned behaviour, for example,
This means that the model is more useful for explaining behaviour has been operationalized with questionnaires designed to identify
when the control beliefs are accurate, and the subjects have con‐ the Behavioural Attitude, Subjective Norm, and Perceived Control.
trol over the relevant circumstances of action. Because Benner provides detailed criteria for the identification of
Second, because all models ignore something, they implicitly dis‐ each stage of expertise, her model is also well operationalized. As
value some things and value others. The final pragmatic criterion of the example of Benner’s theory shows, operationalization is not
evaluation, then, concerns the values that are embedded in a model. always a “quantitative” matter; qualitative criteria can be just as
Imagine, for example, research into the barriers to healthcare access useful.
for a marginalized and oppressed population. The theory of planned The question of precision is whether predictions be made about
behaviour, as just remarked, abstracts away from objective con‐ how change in one element will correlate with change in another. A
straints and focuses on the individual’s beliefs, attitudes, and evalu‐ test of the model will look to see whether the phenomenon mod‐
ations. Barriers to access for a marginalized population are likely to elled behaves as the model says it does. This means that the model
be social or structural, such as a lack of transportation or linguistic must support predictions about how changes in the entities con‐
competence. To use only the theory of planned behaviour to under‐ stitute or cause changes in the larger whole. If the model is impre‐
stand the problems of access is to implicitly place the responsibility cise, then it will be more difficult to test and judge its accuracy.
on the individuals, rather than the environment. But when the pop‐ Moreover, an imprecise model will be less useful for designing inter‐
ulation is marginalized or oppressed, shifting the burden in this way ventions. A troublesome feature of many middle‐range theories is
is morally unacceptable. that they are vague. The theory of planned behaviour, for example,
Including consideration of the moral or political values among says very little about how and when changes in the Behavioural
the criteria of evaluation does not presuppose that the resolution Attitude, Subjective Norm, and Perceived Control will change the
of value questions is straightforward. The question of how moral intention. How strong must the Subjective Norm be to support an
or political judgments are to be grounded is difficult, and it is well Intention? The Kübler–Ross model has been criticized because the
outside the scope of this essay. The inclusion of value consider‐ relationships among the stages are so unclear. Knowing that a per‐
ations here serves to remind us that science has a social context, son is in the denial stage, for example, tells us nothing about what
and that nursing science in particular is part of the social mandate stage might be next. Vague models are less useful for nursing prac‐
of nursing. We do not get to choose our values in a vacuum, iso‐ tice than precise ones, and they do little to advance our theoretical
lated from the possibility of criticism. Therefore, scientists need to understanding.
ask themselves whether the values embedded in a model are ra‐ A significant advantage of representing a middle‐range theory
tionally defensible. Can reasons be given in favour of those values? as a model is that it highlights the ways that the model may be
Do they follow from other commitments shared by researchers in significantly tested. If the model is operationalized and reasonably
the field, healthcare providers, or our larger community? To adopt precise, the relevant test would measure or otherwise observe
a model is to commit to the values implicit in it, and if one cannot whether the elements change in the way that the model predicts.
find good grounds for such commitment, then it is a model to be Such tests are direct tests of the model. Whether a theory has
avoided. been subject to direct tests is therefore a third epistemic criterion:
whether the model has empirical support. The theory of planned
behaviour has been extensively tested in this way by observing
5.2 | Epistemic criteria of evaluation
whether the elements are correlated. This is a point in its fa‐
The epistemic criteria of evaluation are not dependent on the re‐ vour. The Kübler–Ross model, by contrast, has not fared so well.
searcher’s particular interests. Rather, they speak to how accurate Maciejewski, Zhang, Block, and Prigerson (2007) tested whether
RISJORD |
      9 of 10

bereaved persons moved through the stages of grief in the order one or more of these criteria is not a reason to reject it outright. As
postulated by the Kübler–Ross model. They do not, which shows already mentioned, even our best scientific theories have known
that if the Kübler–Ross model is understood as postulating a tem‐ anomalies. While a model may be empirically weak, it may still be
poral progression, it is not accurate. the best one we have. It is crucial, nonetheless, to identify the
Theorists often point with pride to the use of their model to weaknesses of a model. The capacity for identifying and correct‐
develop a successful intervention. However, an intervention which ing errors is perhaps the distinctive characteristic of a science. The
is inspired by the model is typically not a direct test and offers only proper attitude for nurse scholars, then, is to get a clear‐eyed view
weak, indirect support. A given intervention can often be justi‐ of the flaws of a model and then revise the model to make it bet‐
fied by different, even inconsistent, models. Therefore, the fact ter: more clearly operationalized, more precise, more successful
that an intervention is successful does not, ordinarily, show that upon direct testing, and more integrated with relevant theories.
the model on which it is based was accurate or reliable. An inter‐
vention would be a direct test only if it attempted to directly ma‐
nipulate the elements postulated by the model and to determine 6 | CO N C LU S I O N : M I D D LE‐ R A N G E
whether the larger phenomenon changed in the way predicted. TH EO R I E S A S M O D E L S
Models can also garner epistemic support from other scien‐
tific knowledge. We expect that our scientific knowledge will fit Thinking of middle‐range theories as models is a fundamentally
together into a whole, and it is evidence of accuracy when there different perspective from the standard advice available to nursing
is evidence from other domains that the elements and relation‐ scholars. The modelling perspective has salutary effects on both
ships exist. This is theoretical support for the model. Sometimes, the way theories are understood and the way they are evaluated.
when the model is the direct product of a specific theory, this re‐ It shifts attention away from the definitions and propositions of
lationship is very tight. In the Newtonian model of the billiard ball a theory and towards its content. By emphasizing the parts of a
collision, discussed above, the relationships were derived from the system and the relationships in virtue of which the phenomenon is
theory’s laws. Insofar as these laws have found empirical support, brought about, a modelling perspective facilitates a deeper under‐
then, we can have confidence in the model. But as noted above, standing of the theory. It thereby presents a clearer picture of how
models and theories in the health sciences do not have the form effective interventions might be designed. The consequences of a
of Newton’s mechanics. The relationship between the model and modelling perspective for theory evaluation are, perhaps, the most
other theories may be rather loose. The Dreyfus and Dreyfus dramatic. The widely disseminated criteria for theory evaluation
model of skill acquisition drew heavily on cognitive psychology in are very little help to students and scholars who need to choose
its characterization of the relationship between the stages. This is among the plethora of theories currently on offer. A modelling per‐
a strong point in its favour. The theory of planned behaviour, by spective permits a much more detailed and robust evaluation of,
contrast, does not draw on known mechanisms in psychology or and comparison among, theories than the standard criteria.
neuroscience. Given that its entities are all psychological, that has The astute reader will have noticed another feature of the mod‐
not been integrated with contemporary cognitive psychology is a elling perspective: there is no mention of “grand theory” and no
reason to be suspicious of its accuracy. criterion requires a relationship between middle‐range theory and
Figure 2 summarizes the criteria for model evaluation dis‐ grand theory. In the view of this author, the lack of reference to
cussed in this Section. The fact that a model may be criticizable on grand theory is a feature, not a bug. Grand theories in nursing should
not be seen as functioning in the way that fundamental theories like
quantum mechanics or evolution do (Risjord, 2010). Indeed, from
the modelling perspective, there is nothing “middle‐range” about
middle‐range theories. We could just call them “theories,” but that
would obscure the point of this essay. What’s in a name, anyway?

AC K N OW L E D G E M E N T S

This essay took many years to see publication; too many. It was
developed and refined through many presentations. I would like
to thank my nursing colleagues at the International Philosophy of
Nursing Society, Emory University, the University of Lisbon, the
University of Ostrava, the University of Oslo, Laval University,
and the Sahlgrenska Academy at the University of Gothenburg
for their many insightful comments and tough criticisms. I would
also like to thank the many classes of nursing PhD students at
F I G U R E 2   Model Evaluation Summary Emory who patiently endured my attempts to make these ideas
 |
10 of 10       RISJORD

clear and gave me many useful suggestions. Finally, thank you to Kübler‐Ross, E. (1969). On death and dying. London, UK: Routlege.
Derek Sellman, editor of Nursing Philosophy, and to its anonymous Kuhn, T. (1970). The structure of scientific revolutions, 2nd ed. (p. 1962).
Chicago, IL: Chicago University Press. Original edition.
reviewers for their crucial input into the final presentation.
Lenz, E. R., Suppe, F., Gift, A. G., Pugh, L. C., & Milligan, R. A. (1995).
Collaborative development of middle‐range theories: Toward a
theory of unpleasant symptoms. ANS: Advances in Nursing Science,
ORCID
17(3):1–13.
Maciejewski, P. K., Zhang, B., Block, S. D., & Prigerson, H. G. (2007). An
Mark Risjord  http://orcid.org/0000-0003-2129-2009
empirical examination of the stage theory of grief. Journal of the
American Medical Association, 297(7), 716. https://doi.org/10.1001/
jama.297.7.716
REFERENCES Masters, K. (Ed.) (2015). Nursing theories: A framework for practice.
Burlington, NJ: Jones and Bartlett Learning.
Ajzen, I. (1985). From intentions to actions: A theory of planned behav‐
Meleis, A. I. (2007). Theoretical nursing: Development and progress, 4th ed.
ior. In J. Kuhl, & J. Beckmann (Eds.), Action control: From cognition to
(p. 1985). Philadelphia, PA: Lippincott, Williams & Wilkins. Original
behavior. New York, NY: Springer‐Verlag.
edition.
Ajzen, I. (1991). The theory of planned behavior. Organizational be‐
Neuman, B. (1982). The Neuman systems model: Application to nursing edu‐
havior and human decision processes, 50, 179–211. https://doi.
cation and practice. Norwalk, CT: Appleton‐Century‐Crofts.
org/10.1016/0749-5978(91)90020-T
Parse, R. R. (2005). Parse's criteria for evaluation of theory with a compar‐
Ajzen, I. (2005). Attitudes, personality, and behavior, 2nd ed. Berkshire,
ison of Fawcett's and Parse's approaches. Nursing Science Quarterly,
UK: Open University Press.
18(2), 135–137. https://doi.org/10.1177/0894318405275860
Bender, M. (2018). Models versus theories as a primary carrier of nurs‐
Peterson, S., & Bredow, T. (Eds.) (2008). Middle‐range theories: Application
ing knowledge: A philosophical argument. Nursing Philosophy, 19(1),
to nursing research, 2nd ed. Philadelphia, PA: Lippincott Williams &
e12198.
Wilkins.
Benner, P. (1984). From novice to expert. Menlo Park, CA: Addison‐Wesley.
Risjord, M. (2010). Nursing knowledge: Science, practice, and philosophy.
Donaldson, S. K. (2000). Breakthroughs in scientific research: The dis‐
London, UK: Wiley‐Blackwell.
cipline of nursing, 1960–1999. In J. J. Fitzpatrick, & J. Goeppinger
Tomey, A. M., & Alligood, M. R. (Eds.) (1998). Nursing theorists and their
(Eds.), Annual review of nursing research (pp. 247–311). New York, NY:
work, 4th ed. St. Louis, MO: Mosby.
Springer Publishing Company.
Dreyfus, S. E., & Dreyfus, H. L. (1980). A five stage model of the men‐
tal activities involved in directed skill acquisition. Washington DC:
Operations Research Center. How to cite this article: Risjord M. Middle‐range theories as
Fawcett, J. (2005a). Contemporary nursing knowledge: Analysis and eval‐ models: New criteria for analysis and evaluation. Nurs Philos.
uation of nursing models and theories, 2nd ed. Philadelphia, PA: F. A.
2018;e12225. https://doi.org/10.1111/nup.12225
Davis.
Fawcett, J. (2005b). Criteria for evaluation of theory. Nursing Science
Quarterly, 18(2), 131–135.