This article was downloaded by: [Monash University Library]

On: 05 February 2015, At: 00:20

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,
UK

Journal of the American

Statistical Association
Publication details, including instructions for authors
and subscription information:
http://www.tandfonline.com/loi/uasa20

Business Failures: Another

Example of the Analysis of
Failure Data
a
K. S. Lomax
a
University of Manchester
Published online: 11 Apr 2012.

To cite this article: K. S. Lomax (1954) Business Failures: Another Example of the
Analysis of Failure Data, Journal of the American Statistical Association, 49:268,
847-852

To link to this article: http://dx.doi.org/10.1080/01621459.1954.10501239

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the
information (the “Content”) contained in the publications on our platform.
However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness,
or suitability for any purpose of the Content. Any opinions and views
expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the
Content should not be relied upon and should be independently verified with
primary sources of information. Taylor and Francis shall not be liable for any
losses, actions, claims, proceedings, demands, costs, expenses, damages,
and other liabilities whatsoever or howsoever caused arising directly or
indirectly in connection with, in relation to or arising out of the use of the
Content.
 This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan,
sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden. Terms & Conditions of access and use can be found at
http://www.tandfonline.com/page/terms-and-conditions
Downloaded by [Monash University Library] at 00:20 05 February 2015
 BUSINESS FAILURES: ANOTHER EXAMPLE OF THE
ANALYSIS OF FAILURE DATA
K. S. LOMAX
University of Manchester

The analyses of failure data given by Davis [II all involve

essentially constant or increasing conditional probabilities
of failure. For business failures, however, it is reasonable to
expect monotonically decreasing conditional probabilities.
An analysis of data on failures of four types of business in
Poughkeepsie, New York, from 1844 to 1926 [21 confirms this
Downloaded by [Monash University Library] at 00:20 05 February 2015

expectation. The conditional probabilities of failure for these

four series are well described by both exponential and hyper-
bolic functions.

interesting and stimulating paper by D. J. Davis [1] on failure

T
HE
data is most suggestive to the economist.
Broadly, Davis analyzes three types of failure theory:
(a) The normal theory of failure, in which the failure probability
density function is Gaussian.
(b) Human mortality, characterized by rapid increase of the con-
ditional density function after middle-age.
(c) Exponential theory of failure, in which the conditional density
function is constant.
In (a) uniformly and in (b) after the very early years of life the con-
ditional density function of failure probability with time is strictly
monotonic increasing. In (c) it is constant.
The economist immediately thinks of business failures in which it is
reasonable to expect the conditional density function strictly to de-
crease monotonically. The purpose of this note, then, is to draw atten-
tion to a fourth category of failure theory:
(d) Business mortality. It is fairly well established that with most
types of business the early years are the most difficult. It is then
that mortality is highest. The longer a business survives, gener-
ally, other things being equal, the smaller becomes the prob-
ability of failure.
Take, for example, the useful and comprehensive data compiled by
R. G. and A. R. Hutchinson and Mabel Newcomer [2]. Their Table I,
showing the length of life for business enterprises established in Pough-
keepsie between 1844 and 1926, can serve as basis for calculation of
F(t), for different values of t, where
F(t) = cumulative probability of failure in the interval (0, t).
847
 848 AMERICAN STATISTICAL ASSOCIATION JOURNAL, DECEMBER 1954

The results of these calculations, omitting wholesale businesses since

the sample there was small in comparison with the other categories, are
shown in Table 1.
TABLE 1
CUMULATIVE PROBABILITY OF BUSINESS FAILURE
IN POUGHKEEPSIE, 1844-1926

Age in Manu-
years Retail Craft Service
facture
Downloaded by [Monash University Library] at 00:20 05 February 2015

1 0.296 0.231 0.307 0.327

2 0.438 0.346 0.454 0.457
3 0.532 0.469 0.551 0.551
4 0.594 0.547 0.607 0.618
5 0.643 0.602 0.660 0.669
6 0.684 0.655 0.697 0.708
7 0.715 0.678 0.727 0.743
8 0.741 0.702 0.753 0.769
9 0.762 0.726 0.772 0.792
10 0.782 0.746 0.791 0.812
Source: Calculated from Hutchinson, Hutchinson, and Newcomer [21.

In all these cases the graph of F(t) takes the shape shown in Figure 1.

F(t)

FIG. 1

Now, if
Jet) = probability density function of failure time
= probability of failure in infinitesimal interval (t, t+dt),
 BUSINESS FAILURES: ANALYSIS OF FAILURE DATA 849
then jet) =F'(t) and two methods are available for estimation of j(t)
from the above data. One is to measure the slopes of the F(t) graphs
at the different values of t. The other is to use the difference formula
F'(t) = fJ.F(t) - tfJ. 2F(t) + !fJ.
3F(t) - ....
The former method seems to be the more satisfactory here, and the
results of applying it to the data of Table 1 are shown in Table 2.
TABLE 2
PROBABILITY DENSITY OF BUSINESS FAILURE IN
Downloaded by [Monash University Library] at 00:20 05 February 2015

POUGHKEEPSIE, 1844-1926

Age of Manu-
years Retail Craft Service
facture
I
0 0.57 0.365 0.5 0.5
1 0.175 0.152 0.213 0.192
2 0.107 0.119 0.102 0.106
3 0.071 0.090 0.072 0.081
4 0.056 0.067 0.056 0.059
5 0.045 0.051 0.044 0.044
6 0.036 0.037 0.034 0.038
7 0.028 0.030 0.028 0.031
8 0.023 0.023 0.024 0.025
9 0.020 0.014 0.020 0.021
10 0.018 0.007 0.017 0.019
Source: Computed from Table 1.

Thus, jet) can generally be represented as in Figure 2.

j(t)

FIG. 2
 850 AMERICAN STATISTICAL ASSOCIATION JOURNAL, DECEMBER 19114

From the values of f(t) and F(t), following Davis, we calculate Z(t)
the conditional density function of failure probability with time, in
other words, the instantaneous probability rate of failure at time t
conditional upon non-failure prior to i:
Z(t) _ f(t)
1 - F(t)
These results are shown in Table 3.
TABLE 3
Downloaded by [Monash University Library] at 00:20 05 February 2015

CONDITIONAL PROBABILITY DENSITY OF BUSINESS

FAILURE, POUGHKEEPSIE, 1844-1926

Age in Manu-
years Retail Craft Service
facture

0 0.57 0.365 0.5 0.5

1 0.249 0.198 0.307 0.285
2 0.190 0.182 0.187 0.195
3 0.152 0.169 0.160 0.180
4 0.138 0.148 0.142 0.154
5 0.126 0.128 0.129 0.133
6 0.114 0.107 0.112 0.130
7 0.098 0.093 0.103 0.121
8 0.089 0.077 0.097 0.108
9 0.084 0.051 0.088 0.101
10 0.083 0.028 0.081 0.101

Source: Calculated from Tables 1 and 2.

Z(t) is of the fOl'm"'shown in Figure 3.

Z(t)

t
FIG. 3
 BUSINESS FAILURES: ANALYSIS OF FAILURE DATA 851
There is really little purpose to be served by searching for analytical
expressions representing this behavior. This could only be useful if it
were feasible to obtain general support from extraneous sources for a
particular form of relationship. The only such support in this case is
in relation to such trivialities as
F(t), !(t), Z(t)::t>1,
F(O) = 0; leO) = Z(O),
Downloaded by [Monash University Library] at 00:20 05 February 2015

F(t) monotonic increasing.

It is, however, of interest to record that a good fit to the Z(t) values
can be obtained, in each case, either by the exponential function
Z(t) = ae:"
or the hyperbola
b
Z(t) = - - .
t+a
The latter appears to be the more appropriate for the Retail, Craft,
and Service groups, while in the case of Manufacturing trades the ex-
ponential gives the better fit. These above functions were fitted to the
data in the transformations
log. Z(t) = log. a - bt, linear in t and log Z(t)
and
1 1 a 1
--=-t+-, linear in t and
Z(t) b b Z(t)

The correlation coefficients corresponding to these linear forms are

shown in Table 4.
TABLE 4
CORRELATION COEFFICIENTS FOR FUNCTIONS
FITTED TO DATA OF TABLE 3

Type of business Exponential Hyperbola

Retail 0.91 0.99

Manufacture 0.96 0.83
Craft 0.93 0.99
Service 0.91 0.98

One advantage of the hyperbolic "law" is that the expressions for

J(t) and F(t) remain fairly simple.
 852 AMERICAN STATISTICAL ASSOCIATION JOURNAL, DECEMBER 1954

If
b
Z(t) =--,
t+a
then

F(t) l_(_a )b
t + a
and
Downloaded by [Monash University Library] at 00:20 05 February 2015

b ( a
f(t) = - - -
)b+l ;
a t +a
whereas
Z(t) = ae:"
leads to
F(t) 1- ea/bCe-bl-l)

and
f(t) = ae-bt+a/b(e-bt-l).

Both alternatives conform to the desirable boundary conditions and

monotonic behavior exhibited by the data.
I hope shortly to carry out a more detailed analysis of business mor-
tality covering British as well as American data. A cursory examination
has already indicated that British experience does not always com-
pletely accord with American.
REFERENCES
[1] Davis, D. J., "An analysis of some failure data," J ournal of the American Sta-
tistical Association, 47 (1952), 113-.50.
[2] Hutchinson, Ruth Gillette and Arthur IL, and Newcomer, Mabel, "Study
in business mortality," American Economic Review, 28 (1938), 497-.5H.