Description
Ponemon [Ponemon, L. (1988). A cognitive–developmental approach to the analysis of Certified Public Accountants’
ethical judgments. PhDdissertation, Union College of Union University; Accounting, Organizations and Society 17
(1992) 239] contends that a selection–socialization process exists in which (1) senior-level public accountants in the
United States have less capacity to comprehend and resolve basic ethical problems than do accountants in subordinate
ranks, and (2) the senior-level accountants perpetuate their ranks by promoting persons of low moral reasoning capacity.
An empirical reanalysis of the selection–socialization
hypothesis: a research note
Stephen B. Sco?eld
a,,
, Thomas J. Phillips Jr.
b,
*, Charles D. Bailey
c
a
Department of Accounting & CIS, Manning Center for Professional Ethics, Texas A&M University, Kingsville, TX 78363, USA
b
School of Professional Accountancy, Louisiana Tech University, Ruston, LA 71272, USA
c
School of Accountancy, Fogelman College of Business and Economics, The University of Memphis, Memphis, TN 38152, USA
Abstract
Ponemon [Ponemon, L. (1988). A cognitive–developmental approach to the analysis of Certi?ed Public Accountants’
ethical judgments. PhD dissertation, Union College of Union University; Accounting, Organizations and Society 17
(1992) 239] contends that a selection–socialization process exists in which (1) senior-level public accountants in the
United States have less capacity to comprehend and resolve basic ethical problems than do accountants in subordinate
ranks, and (2) the senior-level accountants perpetuate their ranks by promoting persons of low moral reasoning capa-
city. Using large, national random samples from 1995 and 2001, and correcting for inherent problems in the key study
supporting this hypothesis, the current study yields results that materially contradict that earlier research. The sample
sizes provide ample statistical power to conclude that, in the United States, capacity for moral judgment does not di?er
across rank (position) and this is not a factor in promotion in the public accounting profession as a whole. The study
examines both political conservatism and gender as potential confounding variables, but ?nds that they do not drive
the results. A sample of attorneys also gives no hint of selection–socialization. Instead, di?erences in moral cognition
seem to be most relevant when people are making a choice of profession and less so after entering a particular
profession.
Published by Elsevier Ltd.
Introduction
One stream of research on ethics in the
accounting profession deals with a process of
selection and promotion in public accounting
?rms. Ponemon (1988, 1992) ?nds that higher-
ranking CPAs in public practice exhibit dimin-
ished capacity for moral judgment. He attributes
this pattern to discrimination in promotion. That
is, the process of promotion to the rank of man-
ager, and then to partner, selects against those
having a more highly developed capacity for
moral judgment. Ponemon concludes, ‘‘As a
result, those at the top are ill equipped to deal with
and resolve the multitude of ethical con?icts that
abound in everyday practice’’ (Ponemon, 1992, p.
254). That paper has been cited in 26 publications
as of September 2002, according to the Social
Sciences Citation Index, and was the ?fth or sixth
most frequently cited paper in Behavioral Research
in Accounting during the 1989–1998 period (Meyer
& Rigsby, 2001). The general ?nding that partners
in public accounting ?rms have lower ethical rea-
0361-3682/03/$ - see front matter Published by Elsevier Ltd.
doi:10.1016/S0361-3682(03)00021-7
Accounting, Organizations and Society 29 (2004) 543–563
www.elsevier.com/locate/aos
* Corresponding author. Tel.: +1-318-257-2822; fax: +1-
318-257-4253.
E-mail addresses: [email protected] (T.J. Phillips
Jr.), [email protected] (C.D. Bailey).
soning than sta? members seems to have passed
into the accepted wisdom.
Two reasons motivate empirical reexamination
of the selection–socialization hypothesis. First,
Ponemon (1988/1992) argues not that the selec-
tion–socialization process occurs only in a certain
few ?rms, but that the process is a pervasive char-
acteristic of the public accounting culture in the
United States. Ponemon’s data indicate that
between the ranks of supervisor and partner, the
moral reasoning capacity of CPAs in public prac-
tice drops 1.59 standard deviations.
1
We believe
that, in order for de?cient moral reasoning to have
such a dominant e?ect on the promotion decision
to partner (via the selection–socialization process),
this criterion would have to be weighted more
heavily than the combined weight placed on other
likely criteria, such as technical and administrative
competence, appearance, energy, intelligence,
work motivation and other personality and psy-
chological traits. This presses the limits of cred-
ibility and warrants additional evidence. Further,
Ponemon reports results concerning stages of
ethical-reasoning development that are contrary to
developmental theory, with the ‘‘conventional’’
level that one expects from most adults being
suppressed and lower than the subjects’ more
advanced, ‘‘principled’’ reasoning.
Second, we believe that errors in Ponemon
(1988/1992) undermine the conclusions. As dis-
cussed later, the demographics of the data are not
consistent with those of the population from
which the data were drawn, and certain statistical
controls important to the interpretation of the
results are not present.
Cognitive developmental theory and ethical beha-
vior in the accounting profession
Much of the contemporary research pertaining
to ethical behavior in the accounting profession
relies heavily on the cognitive developmental
paradigm (Louwers, Ponemon, & Radtke, 1997;
chap. 6). To allow measurement of moral judg-
ment via a paper-and-pencil test, rather than the
labor-intensive Moral Judgment Interview, James
Rest (see Rest & Narva´ ez, 1994, for background)
developed the De?ning Issues Test (DIT), which
has been used in most studies involving large
samples. The DIT presents the subject with ethical
dilemmas (six in the full set or three in a reduced
set) and asks the subject to rate, then rank, the
importance of factors to consider in making an
ethical decision about each dilemma.
About 500 published studies have used the DIT
during the past quarter century (Rest, Narva´ ez,
Bebeau, & Thoma, 1999, p. 62). The most-used
measure derived from the DIT has been the ‘‘P’’
score, which represents the percentage of respon-
ses that represent ‘‘principled’’ reasoning.
2
Recently, after a period of validation, Rest and
his colleagues at the University of Minnesota
released a revised instrument called the DIT2,
which uses contemporary ethical scenarios and has
demonstrated slightly greater reliability and valid-
ity than the original DIT (Rest, Narva´ ez, Bebeau
et al., 1999; Rest, Narva´ ez, Thoma, & Bebeau,
1999).
The DIT’s ease of application has encouraged
widespread use, perhaps to the neglect of other
productive approaches (Reiter, 2000). Moral
judgment is but one link in the chain that leads to
moral behavior; these links include the moral sen-
sitivity to recognize that a moral question exists,
moral judgment of right and wrong, moral moti-
vation (versus competing motives), and moral
character to persist against obstacles and do the
‘‘right’’ thing (Rest, Narva´ ez, Bebeau et al., 1999,
101; Rest & Narva´ ez, 1994, 23). Nonetheless, the
impact of moral judgment upon the ?nal action is
real: high DIT scores correlate with a variety of
‘‘prosocial’’ behaviors (with r’s in the 0.30–0.40s),
and Rest and Narva´ ez (1994) summarize research
1
He reports a 15.57 point drop in the mean De?ning Issues
Test ‘‘P’’ score (representing higher-level or ‘‘principled’’ moral
reasoning) between supervisors (47.74) and partners (32.17).
The standard deviation for supervisors is 9.79, so that 15.57
points is 1.59 standard deviations. See Ponemon, 1992, p. 247.
2
The P score normally is expressed in terms of a percentage
of responses that are principled responses, rather than a raw
score (Rest, 1990). Inadvertent reporting of raw scores has led
to some spurious results and ‘‘researchers are cautioned to
convert all raw scores to percentage scores’’ (Rest, Narva´ ez,
Bebeau et al., 1999, 116). All of our mentions of P scores refer
to P percentage scores, consistent with past research.
544 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
that links moral development to performance in
the professions. Each of the other links in the
chain deserves more attention from researchers,
yet the preeminence of moral judgment in existing
research cannot be denied, and the integrity of the
published ?ndings on moral judgment is a matter
of concern to researchers and practitioners.
Prior research on moral development in the
accounting profession
Several studies have shown that accountants
exhibit lower moral development scores than other
professional groups (Armstrong, 1987; Ponemon
& Glazer, 1990; St. Pierre, Nelson, & Gabbin,
1990). Of perhaps greater concern, however, are
the cross-sectional and longitudinal studies that
appear to show that the promotion and retention
processes in public accounting ?rms tend to retain
and promote people with less capacity for moral
judgment. In the ?rst such study, Ponemon (1988,
1992), measured the P scores of a sample of ‘‘CPA
practitioners’’ drawn at random from the roster of
the American Institute of Certi?ed Public
Accountants (AICPA). P scores were inversely
proportional to position in the highest ranks of
accounting ?rms. A separate, longitudinal phase
of the study ties the pattern to discrimination
in the promotion and retention process against
candidates with a more highly developed moral
reasoning capacity.
Ponemon and Gabhart (1993) performed a
similar study of P scores for CPAs in several multi-
national accounting ?rms in the northeastern
United States and for a ?rm in Ontario, Canada.
The US accountants’ P scores decreased with
professional rank, while scores for the Canadian
accountants increased. Ponemon and Gabhart
speculate that the di?erence in patterns might be
attributable to di?erences between socialization
pressures in the professional cultures of the United
States and Canada.
Shaub (1994) studied a sample of auditors from
several southwestern o?ces of a multinational
?rm. Shaub’s data also exhibit a decline in P score
but much less than in Ponemon’s data, and the
decline begins at the senior level rather than the
supervisor level.
Bernardi and Arnold (1997), in a cross-sectional
study of managers and seniors at ?ve of the ‘‘Big
6’’ ?rms, ?nd some support for the selection–
socialization hypothesis. However, in a long-
itudinal extension of the study (Arnold &
Bernardi, 2001) they ?nd no such e?ect in actual
promotion decisions. Je?rey and Weatherholt
(1996) found no di?erence in P scores of auditors
across ranks in public accounting.
Evidence in the US public accounting milieu of
a pervasive selection process that chooses those
with lower ethical reasoning ability is disturbing.
Consequent pressure on individuals to conform to
the expectations of a corporate or professional
culture could lead to the compromise of one’s
ethical code, e?ectively undermining the ?nancial
reporting process. While some recent studies have
provided limited disagreement with Ponemon’s
?ndings, it remains largely accepted that part-
ners have lower ethical reasoning capacity than
subordinates.
Critique of Ponemon 1988/1992
Ponemon’s selection–socialization hypothesis is
instinctively appealing. Many professionals have
felt pressures to conform to the expectations of a
corporate or professional culture and perhaps to
compromise one’s ethical code. In his dissertation,
Ponemon (1988) recounts his personal experience
interviewing with the ‘‘Big 8’’ ?rms—wearing a
bright green suit. His perception that he did not
conform to the ?rm uniform increased his ‘‘ten-
sion level to paranoiac proportions,’’ and the job
o?er at one ?rm came ‘‘contingent upon your
burning that green suit’’ (p. 100). While this issue
may not be an ethical one, the metaphor is clear.
To place high value on conformance to society is
to move toward Kohlberg’s (1969) ‘‘conventional’’
reasoning, which is below the level of principled
reasoning.
Two discrepancies, however, undermine Pone-
mon’s (1988/1992) results: the demographics of
the data analyzed in the study di?er greatly from
the population from which the sample was drawn,
and no e?ective control exists for the possible
e?ects of ?rm size.
S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563 545
Discrepancy with the population demographics
Ponemon described his sample as a ‘‘simple
random sample of CPA practitioners. . . selected
from the AICPA’s computerized membership ros-
ter’’ (Ponemon, 1988, p. 163) and reported that
only CPAs in public accounting practice were
sampled (Ponemon, 1992, p. 246). As of 31 July
1994, the ‘‘AICPA membership . . . in public
accounting practice’’ consisted of ‘‘partners’’ and
?rm sta? in the numbers shown in Fig. 1. Com-
parable detail for the 1987/1988 population has
not been maintained, although reconstructed
records indicate that the pro?le of the 1987/1988
membership was materially the same as the pro?le
of the 1994/1995 membership.
3
In addition, sum-
mary statistics published by the AICPA exhibit no
material change between 1988 and 1994 in mem-
bership distribution by ?rm size.
4
Fig. 1. AICPA membership in public practice membership structure by ?rm size. Source:AICPA communications.
3
AICPA communications with the ?rst author.
4
AICPAhttp://www.aicpa.org [Member Matters] [AICPA
Membership Breakdown], 1996.
546 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
Demographics of the sample in Ponemon (1988/
1992) do not match the demographics of CPA
practitioners in the AICPA. Firms with fewer than
50 professionals constitute 78% of the population
(see bottom of Fig. 1) but only 17% of the records
in Ponemon (1988/1992). Firms with more than 50
professionals constitute only 22% of the popula-
tion but make up 83% of Ponemon’s respondents.
These proportions di?er so substantially that the
probability that the proportions of responses in
Ponemon’s study are a random departure from the
dimensions of the population of the AICPA
membership in public accounting practice is
remote (
2
=74.10, p249 Professionals
Mean S.D. N Mean S.D. N
Descriptive statistics
Sta? 39.2 15.5642 79 40.928 13.996 18
Senior 41.126 15.2303 195 39.277 15.4821 60
Supervisor 41.342 17.7387 67 43.657 15.8395 21
Manager 43.675 15.5111 128 44.926 14.373 65
Partner 39.088 14.8552 146 42.668 11.9353 40
Total 40.949 15.5743 615 42.338 14.4422 204
Source Type III SS df Mean square F Sig. Partial Z
2
F Sig. Partial Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected model 11,691.58 6 1948.60 8.63 0.000 0.079 4.50 0.000 0.121
Intercept 39,947.41 1 39,947.41 176.97 0.000 0.225 75.00 0.000 0.276
SEX 2754.24 1 2754.24 12.20 0.001 0.020 3.84 0.051 0.019
CNS 6668.50 1 6668.50 29.54 0.000 0.046 14.94 0.000 0.070
POS 1553.53 4 388.38 1.72 0.144 0.011 1.39 0.238 0.027
Error 137,240.37 608 225.72
Total 1,180,165.57 615
Corrected total 148,931.96 614 Adjusted R
2
=0.069 Adjusted R
2
=0.094
Table 5
Analysis of covariance, 2001 sample
Recoded position All ?rm sizes >249 Professionals
Mean SD N Mean SD N
Descriptive statistics
Below Manager 38.476 12.663 63 38.313 13.499 32
Manager 37.130 14.430 119 37.123 14.155 65
Partner 34.664 14.550 76 37.852 13.867 27
Total 36.733 14.077 258 37.589 13.824 124
Source Type III SS df Mean square F Sig. Partial Z
2
F Sig. Partial Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected model 3719.86 4 929.96 4.98 0.001 0.073 2.041 0.093 0.064
Intercept 19,407.10 1 19,407.10 140.01 0.000 0.291 48.193 0.000 0.288
SEX 64.64 1 64.64 0.35 0.557 0.001 1.018 0.315 0.008
POLCON 2775.33 1 2775.33 14.88 0.000 0.056 5.985 0.016 0.048
RECPOS 466.22 2 233.11 1.25 0.288 0.010 0.419 0.659 0.007
Error 47,205.20 253 186.58
Total 399,040.00 258
Corrected total 50,925.00 257
Adjusted R
2
=0.058 Adjusted R
2
=0.033
554 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
sample sizes.
17
The results show political con-
servatism (using the DIT2 scale) to explain about
5.6% of the variance and to be signi?cant at p
0.995.
For the 2001 sample from ?rms with >249
professionals, gender again is not signi?cant; the
DIT2 conservatism scale explains 4.8% of the
variance and is signi?cant at p=0.016; and reco-
ded position (RECPOS) is not signi?cant. The
power of this test of the position e?ect is only 0.30
for a small e?ect but about 0.70 for a medium
e?ect and 0.98 for a large e?ect.
‘‘Big 5’’ firms. In the 2001 mailing, respondents
indicated whether they worked for a ‘‘Big 5’’ ?rm
or ever had been so employed. Responses inclu-
ded 42 current ‘‘Big 5’’ managers, 15 partners,
and 17 at lower ranks. The results, shown in
Table 6, are consistent with the larger sample,
with position again not signi?cant. Power to
detect a medium e?ect is 0.36 and to detect a large
e?ect is 0.76.
In summary, no di?erences in P scores appear
across professional positions, in either the 1995 or
2001 samples, that would be consistent with the
selection–socialization hypothesis. The analyses
included all ?rms combined as well as separate
analyses of large ?rms.
Question 2: might political di?erences across ranks
or age groups have driven Ponemon’s observed
di?erences in P scores?
We plotted scores on the Collins–Hayes con-
servatism scale against position in the ?rm, and
also against age, for the 1995 sample (all except
single proprietors). No trends are evident, and
both trend lines are essentially ?at and the slopes
not signi?cant. The same is true for the DIT2
political measure from the 2001 sample plotted
against position and against age. Both tests for
both years are quite powerful: power to detect a
very small correlation (r
2
=0.01) is 0.79 for 1995
and 0.47 for 2001. Power to detect a medium e?ect
(r
2
=0.09) exceeds 0.995 for both years. Thus, we
can conclude with considerable con?dence that no
relationship exists between political conservatism
and either age or ?rm position for US CPAs.
20
17
Ponemon’s sta?, senior, and supervisor groups were simi-
lar as to P scores (Ponemon, 1992, 247), and di?erences among
these groups are not relevant to the selection–socialization
hypothesis. Our uncombined sample sizes are clearly too small
for some comparisons; for example, in a ‘‘Big-5
00
ANCOVA,
only one subject would be in the ‘‘sta?’’ position. Regardless,
ANCOVAs utilizing all ranks consistently found no di?erence
across ranks.
18
Given the absence of other DIT2 data from accountants,
speculation is premature. However, the gender e?ects for
accountants reported in the past are unexplained. Perhaps
something about the DIT1, but not the DIT2, led to the slightly
higher scores among female accountants.
19
Reducing the cuto? point to 100 professionals produced
similar results, but with slightly higher power as a function of
the larger sample size. Only 26 respondents fell into the 100–
249 member range.
20
The attorneys showed very similar patterns. Plotted
against age, conservatism again shows a ?at trend
(R
2
=0.00000). By positions, the partners are marginally less
conservative than subordinates (t=1.67, p=0.05, two-tailed).
Although individuals may grow more conservative with age,
this trend may be o?set by the increasingly conservative poli-
tical climate, such that, in our cross-sectional data, the con-
servatism levels are rather constant.
S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563 555
Accordingly, it seems that the variations Ponemon
found in P scores could not have been the result of
political di?erences across ranks or age groups.
Question 3: might confounding between big-?rm
partners and single proprietors explain Ponemon’s
results?
The 1995 sample included 84 partners employed
by ?rms with >249 professionals
21
and 111 single
proprietors. The mean P score of the large-?rm
partners was 39.21(S.D. 12.22), and the mean for
single proprietors was 37.48 (S.D. 15.20). A t-test is
not signi?cant (t=0.88, p=0.38), and the power of
the test to detect a medium e?ect is >0.90. Thus,
incorrect inclusion of sole proprietors in Ponemon’s
sample of ?rm ‘‘partners’’ does not appear to
account for the partners’ depressed P scores.
Question 4: can di?erences between those leaving
big ?rms and those remaining explain the results?
Only the 2001 data allows examination of dif-
ferences in P scores among CPAs never employed
by the ‘‘Big 5’’, those staying, and those leaving, as
we did not collect relevant data in 1995. Table 7
shows the results of an ANCOVA comparing the
means of those currently employed by the ‘‘Big 5’’,
those never so employed, and those who have left.
The analysis includes as covariates gender and
political conservatism.
22
Only the political con-
servatism variable is signi?cant, and the ?rm-sta-
tus factor is not. The power of this test to detect a
small e?ect is 0.54, and to detect a medium e?ect
is 0.94. To emphasize that no ‘‘tendency’’ exists
towards a pattern consistent with a selection–
socialization bias against ethical big-?rm mem-
bers, note that the highest average P score belongs
to those currently employed by the ‘‘Big 5’’, and
the smallest belongs to those who have left.
As a complementary analysis, we plotted P scores
against year of departure from a ‘‘Big 5/6/8’’ ?rm,
for those who ever have worked for such a ?rm
(n=127). For those still employed by the ‘‘Big 5’’,
we use the year 2001. (Note that everyone in our
sample still was employed in public accounting.)
The trend line is ?at, with r
2
approaching zero.
The power of this regression to detect a relation-
21
Comparisons that include partners in smaller ?rms yield
the same result.
Table 6
Analysis of covariance, ‘‘Big 5’’ ?rms
Position Mean S.D. N
Descriptive statistics (dependent variable: P%)
Manager 36.667 15.740 42
Partner 41.600 13.119 15
Total 37.965 15.140 57
Source Type III sum of squares df Mean square F Sig. Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected Model 2101.654
a
3 700.551 3.459 0.023 0.164
Intercept 6966.802 1 6966.80 34.398 0.000 0.394
SEX 1.699 1 1.699 0.008 0.927 0.000
POLCON 1719.013 1 1719.01 8.488 0.005 0.138
POS 427.860 1 427.860 2.113 0.152 0.038
Error 10,734.3 53 202.534
Total 94,992.0 57
Corrected Total 12,835.9 56
a
R
2
=0.164 (adjusted R
2
=0.116).
22
As with the earlier analyses, age and professional
experience were not signi?cant and are omitted.
556 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
ship in the population with true explained variance
(r
2
) of 0.04 exceeds 0.70, and the power to detect a
relationship with true r
2
of 0.09 exceeds 0.96.
Question 5: does selection–socialization exist in
law ?rms, a similar professional setting?
Of the usable sample of 148 attorneys, 80 were
partners, 63 associates, and only ?ve at lower ranks.
Thus we limited the analysis to partners and associ-
ates. As with accountants, age is not signi?cant in an
ANCOVA, while gender is signi?cant and explains
about 6% of the variance (with the women again
scoring higher). Interestingly, the Collins–Hayes
conservatism score is not a signi?cant covariate for
the attorneys,
23
although their response to the
simple question of whether their political orienta-
tion is ‘‘conservative’’ does explain about 5% of
the variance. As with the accountants, the position
factor is not signi?cant (p=0.77, Z
2
=0.001).
Power is 0.44 to detect a small position e?ect, and
0.84 to detect a medium e?ect.
Attorneys also di?ered from accountants in their
P scores. For the key rank of partner, (80 attorneys
and 146 accountants) average P scores were 45.37
for attorneys and 39.09 for accountants (t=3.15, p
15 years
Experience 38.841 0.000 0.001 0.000 0.000
‘‘Big’’ Firm
Partner >15 years
Experience 43.797 0.000 0.000 0.000 0.000
24
As noted earlier, comparisons to P scores from the DIT2
are not appropriate, precluding use of our 2001 sample in these
tests.
558 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
sponding increase in the far more rudimentary
Stage 3 type reasoning starting at the manager
ranks and continuing into the partner rank. It is
expected that adults would have moved well
beyond the ‘‘self-interest’’ schema of stages 2 and
3 (Rest, Narva´ ez, Bebeau et al., 1999, pp. 36, 54).
Ponemon’s data show higher percentages of the
self-interest schema throughout the ranks, culmi-
nating in a remarkable increase to over 46% in the
partners, dominating their importance decisions
when responding to DIT scenarios. Furthermore,
ethical development, although not strictly ruling
out the usage of more primitive schemas, would
not be expected to show inversions in which an
intermediate stage or schema would appear to be
‘‘skipped.’’ For adults with stunted development,
one would not expect their principled thinking (P)
to exceed their ‘‘conventional,’’ maintaining-
norms thinking (Stage 4). Yet this is the pattern
for both managers and partners in Ponemon’s
data.
The 1995 data, summarized graphically in Fig. 2,
display no signi?cant di?erence between P scores
across professional ranks. The pro?les of scores
for all DIT items are consistent from one profes-
sional level to the next. Stage 2 reasoning and
Stage 3 reasoning are level across all ranks, with a
small proportion (
Ponemon [Ponemon, L. (1988). A cognitive–developmental approach to the analysis of Certified Public Accountants’
ethical judgments. PhDdissertation, Union College of Union University; Accounting, Organizations and Society 17
(1992) 239] contends that a selection–socialization process exists in which (1) senior-level public accountants in the
United States have less capacity to comprehend and resolve basic ethical problems than do accountants in subordinate
ranks, and (2) the senior-level accountants perpetuate their ranks by promoting persons of low moral reasoning capacity.
An empirical reanalysis of the selection–socialization
hypothesis: a research note
Stephen B. Sco?eld
a,,
, Thomas J. Phillips Jr.
b,
*, Charles D. Bailey
c
a
Department of Accounting & CIS, Manning Center for Professional Ethics, Texas A&M University, Kingsville, TX 78363, USA
b
School of Professional Accountancy, Louisiana Tech University, Ruston, LA 71272, USA
c
School of Accountancy, Fogelman College of Business and Economics, The University of Memphis, Memphis, TN 38152, USA
Abstract
Ponemon [Ponemon, L. (1988). A cognitive–developmental approach to the analysis of Certi?ed Public Accountants’
ethical judgments. PhD dissertation, Union College of Union University; Accounting, Organizations and Society 17
(1992) 239] contends that a selection–socialization process exists in which (1) senior-level public accountants in the
United States have less capacity to comprehend and resolve basic ethical problems than do accountants in subordinate
ranks, and (2) the senior-level accountants perpetuate their ranks by promoting persons of low moral reasoning capa-
city. Using large, national random samples from 1995 and 2001, and correcting for inherent problems in the key study
supporting this hypothesis, the current study yields results that materially contradict that earlier research. The sample
sizes provide ample statistical power to conclude that, in the United States, capacity for moral judgment does not di?er
across rank (position) and this is not a factor in promotion in the public accounting profession as a whole. The study
examines both political conservatism and gender as potential confounding variables, but ?nds that they do not drive
the results. A sample of attorneys also gives no hint of selection–socialization. Instead, di?erences in moral cognition
seem to be most relevant when people are making a choice of profession and less so after entering a particular
profession.
Published by Elsevier Ltd.
Introduction
One stream of research on ethics in the
accounting profession deals with a process of
selection and promotion in public accounting
?rms. Ponemon (1988, 1992) ?nds that higher-
ranking CPAs in public practice exhibit dimin-
ished capacity for moral judgment. He attributes
this pattern to discrimination in promotion. That
is, the process of promotion to the rank of man-
ager, and then to partner, selects against those
having a more highly developed capacity for
moral judgment. Ponemon concludes, ‘‘As a
result, those at the top are ill equipped to deal with
and resolve the multitude of ethical con?icts that
abound in everyday practice’’ (Ponemon, 1992, p.
254). That paper has been cited in 26 publications
as of September 2002, according to the Social
Sciences Citation Index, and was the ?fth or sixth
most frequently cited paper in Behavioral Research
in Accounting during the 1989–1998 period (Meyer
& Rigsby, 2001). The general ?nding that partners
in public accounting ?rms have lower ethical rea-
0361-3682/03/$ - see front matter Published by Elsevier Ltd.
doi:10.1016/S0361-3682(03)00021-7
Accounting, Organizations and Society 29 (2004) 543–563
www.elsevier.com/locate/aos
* Corresponding author. Tel.: +1-318-257-2822; fax: +1-
318-257-4253.
E-mail addresses: [email protected] (T.J. Phillips
Jr.), [email protected] (C.D. Bailey).
soning than sta? members seems to have passed
into the accepted wisdom.
Two reasons motivate empirical reexamination
of the selection–socialization hypothesis. First,
Ponemon (1988/1992) argues not that the selec-
tion–socialization process occurs only in a certain
few ?rms, but that the process is a pervasive char-
acteristic of the public accounting culture in the
United States. Ponemon’s data indicate that
between the ranks of supervisor and partner, the
moral reasoning capacity of CPAs in public prac-
tice drops 1.59 standard deviations.
1
We believe
that, in order for de?cient moral reasoning to have
such a dominant e?ect on the promotion decision
to partner (via the selection–socialization process),
this criterion would have to be weighted more
heavily than the combined weight placed on other
likely criteria, such as technical and administrative
competence, appearance, energy, intelligence,
work motivation and other personality and psy-
chological traits. This presses the limits of cred-
ibility and warrants additional evidence. Further,
Ponemon reports results concerning stages of
ethical-reasoning development that are contrary to
developmental theory, with the ‘‘conventional’’
level that one expects from most adults being
suppressed and lower than the subjects’ more
advanced, ‘‘principled’’ reasoning.
Second, we believe that errors in Ponemon
(1988/1992) undermine the conclusions. As dis-
cussed later, the demographics of the data are not
consistent with those of the population from
which the data were drawn, and certain statistical
controls important to the interpretation of the
results are not present.
Cognitive developmental theory and ethical beha-
vior in the accounting profession
Much of the contemporary research pertaining
to ethical behavior in the accounting profession
relies heavily on the cognitive developmental
paradigm (Louwers, Ponemon, & Radtke, 1997;
chap. 6). To allow measurement of moral judg-
ment via a paper-and-pencil test, rather than the
labor-intensive Moral Judgment Interview, James
Rest (see Rest & Narva´ ez, 1994, for background)
developed the De?ning Issues Test (DIT), which
has been used in most studies involving large
samples. The DIT presents the subject with ethical
dilemmas (six in the full set or three in a reduced
set) and asks the subject to rate, then rank, the
importance of factors to consider in making an
ethical decision about each dilemma.
About 500 published studies have used the DIT
during the past quarter century (Rest, Narva´ ez,
Bebeau, & Thoma, 1999, p. 62). The most-used
measure derived from the DIT has been the ‘‘P’’
score, which represents the percentage of respon-
ses that represent ‘‘principled’’ reasoning.
2
Recently, after a period of validation, Rest and
his colleagues at the University of Minnesota
released a revised instrument called the DIT2,
which uses contemporary ethical scenarios and has
demonstrated slightly greater reliability and valid-
ity than the original DIT (Rest, Narva´ ez, Bebeau
et al., 1999; Rest, Narva´ ez, Thoma, & Bebeau,
1999).
The DIT’s ease of application has encouraged
widespread use, perhaps to the neglect of other
productive approaches (Reiter, 2000). Moral
judgment is but one link in the chain that leads to
moral behavior; these links include the moral sen-
sitivity to recognize that a moral question exists,
moral judgment of right and wrong, moral moti-
vation (versus competing motives), and moral
character to persist against obstacles and do the
‘‘right’’ thing (Rest, Narva´ ez, Bebeau et al., 1999,
101; Rest & Narva´ ez, 1994, 23). Nonetheless, the
impact of moral judgment upon the ?nal action is
real: high DIT scores correlate with a variety of
‘‘prosocial’’ behaviors (with r’s in the 0.30–0.40s),
and Rest and Narva´ ez (1994) summarize research
1
He reports a 15.57 point drop in the mean De?ning Issues
Test ‘‘P’’ score (representing higher-level or ‘‘principled’’ moral
reasoning) between supervisors (47.74) and partners (32.17).
The standard deviation for supervisors is 9.79, so that 15.57
points is 1.59 standard deviations. See Ponemon, 1992, p. 247.
2
The P score normally is expressed in terms of a percentage
of responses that are principled responses, rather than a raw
score (Rest, 1990). Inadvertent reporting of raw scores has led
to some spurious results and ‘‘researchers are cautioned to
convert all raw scores to percentage scores’’ (Rest, Narva´ ez,
Bebeau et al., 1999, 116). All of our mentions of P scores refer
to P percentage scores, consistent with past research.
544 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
that links moral development to performance in
the professions. Each of the other links in the
chain deserves more attention from researchers,
yet the preeminence of moral judgment in existing
research cannot be denied, and the integrity of the
published ?ndings on moral judgment is a matter
of concern to researchers and practitioners.
Prior research on moral development in the
accounting profession
Several studies have shown that accountants
exhibit lower moral development scores than other
professional groups (Armstrong, 1987; Ponemon
& Glazer, 1990; St. Pierre, Nelson, & Gabbin,
1990). Of perhaps greater concern, however, are
the cross-sectional and longitudinal studies that
appear to show that the promotion and retention
processes in public accounting ?rms tend to retain
and promote people with less capacity for moral
judgment. In the ?rst such study, Ponemon (1988,
1992), measured the P scores of a sample of ‘‘CPA
practitioners’’ drawn at random from the roster of
the American Institute of Certi?ed Public
Accountants (AICPA). P scores were inversely
proportional to position in the highest ranks of
accounting ?rms. A separate, longitudinal phase
of the study ties the pattern to discrimination
in the promotion and retention process against
candidates with a more highly developed moral
reasoning capacity.
Ponemon and Gabhart (1993) performed a
similar study of P scores for CPAs in several multi-
national accounting ?rms in the northeastern
United States and for a ?rm in Ontario, Canada.
The US accountants’ P scores decreased with
professional rank, while scores for the Canadian
accountants increased. Ponemon and Gabhart
speculate that the di?erence in patterns might be
attributable to di?erences between socialization
pressures in the professional cultures of the United
States and Canada.
Shaub (1994) studied a sample of auditors from
several southwestern o?ces of a multinational
?rm. Shaub’s data also exhibit a decline in P score
but much less than in Ponemon’s data, and the
decline begins at the senior level rather than the
supervisor level.
Bernardi and Arnold (1997), in a cross-sectional
study of managers and seniors at ?ve of the ‘‘Big
6’’ ?rms, ?nd some support for the selection–
socialization hypothesis. However, in a long-
itudinal extension of the study (Arnold &
Bernardi, 2001) they ?nd no such e?ect in actual
promotion decisions. Je?rey and Weatherholt
(1996) found no di?erence in P scores of auditors
across ranks in public accounting.
Evidence in the US public accounting milieu of
a pervasive selection process that chooses those
with lower ethical reasoning ability is disturbing.
Consequent pressure on individuals to conform to
the expectations of a corporate or professional
culture could lead to the compromise of one’s
ethical code, e?ectively undermining the ?nancial
reporting process. While some recent studies have
provided limited disagreement with Ponemon’s
?ndings, it remains largely accepted that part-
ners have lower ethical reasoning capacity than
subordinates.
Critique of Ponemon 1988/1992
Ponemon’s selection–socialization hypothesis is
instinctively appealing. Many professionals have
felt pressures to conform to the expectations of a
corporate or professional culture and perhaps to
compromise one’s ethical code. In his dissertation,
Ponemon (1988) recounts his personal experience
interviewing with the ‘‘Big 8’’ ?rms—wearing a
bright green suit. His perception that he did not
conform to the ?rm uniform increased his ‘‘ten-
sion level to paranoiac proportions,’’ and the job
o?er at one ?rm came ‘‘contingent upon your
burning that green suit’’ (p. 100). While this issue
may not be an ethical one, the metaphor is clear.
To place high value on conformance to society is
to move toward Kohlberg’s (1969) ‘‘conventional’’
reasoning, which is below the level of principled
reasoning.
Two discrepancies, however, undermine Pone-
mon’s (1988/1992) results: the demographics of
the data analyzed in the study di?er greatly from
the population from which the sample was drawn,
and no e?ective control exists for the possible
e?ects of ?rm size.
S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563 545
Discrepancy with the population demographics
Ponemon described his sample as a ‘‘simple
random sample of CPA practitioners. . . selected
from the AICPA’s computerized membership ros-
ter’’ (Ponemon, 1988, p. 163) and reported that
only CPAs in public accounting practice were
sampled (Ponemon, 1992, p. 246). As of 31 July
1994, the ‘‘AICPA membership . . . in public
accounting practice’’ consisted of ‘‘partners’’ and
?rm sta? in the numbers shown in Fig. 1. Com-
parable detail for the 1987/1988 population has
not been maintained, although reconstructed
records indicate that the pro?le of the 1987/1988
membership was materially the same as the pro?le
of the 1994/1995 membership.
3
In addition, sum-
mary statistics published by the AICPA exhibit no
material change between 1988 and 1994 in mem-
bership distribution by ?rm size.
4
Fig. 1. AICPA membership in public practice membership structure by ?rm size. Source:AICPA communications.
3
AICPA communications with the ?rst author.
4
AICPAhttp://www.aicpa.org [Member Matters] [AICPA
Membership Breakdown], 1996.
546 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
Demographics of the sample in Ponemon (1988/
1992) do not match the demographics of CPA
practitioners in the AICPA. Firms with fewer than
50 professionals constitute 78% of the population
(see bottom of Fig. 1) but only 17% of the records
in Ponemon (1988/1992). Firms with more than 50
professionals constitute only 22% of the popula-
tion but make up 83% of Ponemon’s respondents.
These proportions di?er so substantially that the
probability that the proportions of responses in
Ponemon’s study are a random departure from the
dimensions of the population of the AICPA
membership in public accounting practice is
remote (
2
=74.10, p249 Professionals
Mean S.D. N Mean S.D. N
Descriptive statistics
Sta? 39.2 15.5642 79 40.928 13.996 18
Senior 41.126 15.2303 195 39.277 15.4821 60
Supervisor 41.342 17.7387 67 43.657 15.8395 21
Manager 43.675 15.5111 128 44.926 14.373 65
Partner 39.088 14.8552 146 42.668 11.9353 40
Total 40.949 15.5743 615 42.338 14.4422 204
Source Type III SS df Mean square F Sig. Partial Z
2
F Sig. Partial Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected model 11,691.58 6 1948.60 8.63 0.000 0.079 4.50 0.000 0.121
Intercept 39,947.41 1 39,947.41 176.97 0.000 0.225 75.00 0.000 0.276
SEX 2754.24 1 2754.24 12.20 0.001 0.020 3.84 0.051 0.019
CNS 6668.50 1 6668.50 29.54 0.000 0.046 14.94 0.000 0.070
POS 1553.53 4 388.38 1.72 0.144 0.011 1.39 0.238 0.027
Error 137,240.37 608 225.72
Total 1,180,165.57 615
Corrected total 148,931.96 614 Adjusted R
2
=0.069 Adjusted R
2
=0.094
Table 5
Analysis of covariance, 2001 sample
Recoded position All ?rm sizes >249 Professionals
Mean SD N Mean SD N
Descriptive statistics
Below Manager 38.476 12.663 63 38.313 13.499 32
Manager 37.130 14.430 119 37.123 14.155 65
Partner 34.664 14.550 76 37.852 13.867 27
Total 36.733 14.077 258 37.589 13.824 124
Source Type III SS df Mean square F Sig. Partial Z
2
F Sig. Partial Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected model 3719.86 4 929.96 4.98 0.001 0.073 2.041 0.093 0.064
Intercept 19,407.10 1 19,407.10 140.01 0.000 0.291 48.193 0.000 0.288
SEX 64.64 1 64.64 0.35 0.557 0.001 1.018 0.315 0.008
POLCON 2775.33 1 2775.33 14.88 0.000 0.056 5.985 0.016 0.048
RECPOS 466.22 2 233.11 1.25 0.288 0.010 0.419 0.659 0.007
Error 47,205.20 253 186.58
Total 399,040.00 258
Corrected total 50,925.00 257
Adjusted R
2
=0.058 Adjusted R
2
=0.033
554 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
sample sizes.
17
The results show political con-
servatism (using the DIT2 scale) to explain about
5.6% of the variance and to be signi?cant at p
0.995.
For the 2001 sample from ?rms with >249
professionals, gender again is not signi?cant; the
DIT2 conservatism scale explains 4.8% of the
variance and is signi?cant at p=0.016; and reco-
ded position (RECPOS) is not signi?cant. The
power of this test of the position e?ect is only 0.30
for a small e?ect but about 0.70 for a medium
e?ect and 0.98 for a large e?ect.
‘‘Big 5’’ firms. In the 2001 mailing, respondents
indicated whether they worked for a ‘‘Big 5’’ ?rm
or ever had been so employed. Responses inclu-
ded 42 current ‘‘Big 5’’ managers, 15 partners,
and 17 at lower ranks. The results, shown in
Table 6, are consistent with the larger sample,
with position again not signi?cant. Power to
detect a medium e?ect is 0.36 and to detect a large
e?ect is 0.76.
In summary, no di?erences in P scores appear
across professional positions, in either the 1995 or
2001 samples, that would be consistent with the
selection–socialization hypothesis. The analyses
included all ?rms combined as well as separate
analyses of large ?rms.
Question 2: might political di?erences across ranks
or age groups have driven Ponemon’s observed
di?erences in P scores?
We plotted scores on the Collins–Hayes con-
servatism scale against position in the ?rm, and
also against age, for the 1995 sample (all except
single proprietors). No trends are evident, and
both trend lines are essentially ?at and the slopes
not signi?cant. The same is true for the DIT2
political measure from the 2001 sample plotted
against position and against age. Both tests for
both years are quite powerful: power to detect a
very small correlation (r
2
=0.01) is 0.79 for 1995
and 0.47 for 2001. Power to detect a medium e?ect
(r
2
=0.09) exceeds 0.995 for both years. Thus, we
can conclude with considerable con?dence that no
relationship exists between political conservatism
and either age or ?rm position for US CPAs.
20
17
Ponemon’s sta?, senior, and supervisor groups were simi-
lar as to P scores (Ponemon, 1992, 247), and di?erences among
these groups are not relevant to the selection–socialization
hypothesis. Our uncombined sample sizes are clearly too small
for some comparisons; for example, in a ‘‘Big-5
00
ANCOVA,
only one subject would be in the ‘‘sta?’’ position. Regardless,
ANCOVAs utilizing all ranks consistently found no di?erence
across ranks.
18
Given the absence of other DIT2 data from accountants,
speculation is premature. However, the gender e?ects for
accountants reported in the past are unexplained. Perhaps
something about the DIT1, but not the DIT2, led to the slightly
higher scores among female accountants.
19
Reducing the cuto? point to 100 professionals produced
similar results, but with slightly higher power as a function of
the larger sample size. Only 26 respondents fell into the 100–
249 member range.
20
The attorneys showed very similar patterns. Plotted
against age, conservatism again shows a ?at trend
(R
2
=0.00000). By positions, the partners are marginally less
conservative than subordinates (t=1.67, p=0.05, two-tailed).
Although individuals may grow more conservative with age,
this trend may be o?set by the increasingly conservative poli-
tical climate, such that, in our cross-sectional data, the con-
servatism levels are rather constant.
S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563 555
Accordingly, it seems that the variations Ponemon
found in P scores could not have been the result of
political di?erences across ranks or age groups.
Question 3: might confounding between big-?rm
partners and single proprietors explain Ponemon’s
results?
The 1995 sample included 84 partners employed
by ?rms with >249 professionals
21
and 111 single
proprietors. The mean P score of the large-?rm
partners was 39.21(S.D. 12.22), and the mean for
single proprietors was 37.48 (S.D. 15.20). A t-test is
not signi?cant (t=0.88, p=0.38), and the power of
the test to detect a medium e?ect is >0.90. Thus,
incorrect inclusion of sole proprietors in Ponemon’s
sample of ?rm ‘‘partners’’ does not appear to
account for the partners’ depressed P scores.
Question 4: can di?erences between those leaving
big ?rms and those remaining explain the results?
Only the 2001 data allows examination of dif-
ferences in P scores among CPAs never employed
by the ‘‘Big 5’’, those staying, and those leaving, as
we did not collect relevant data in 1995. Table 7
shows the results of an ANCOVA comparing the
means of those currently employed by the ‘‘Big 5’’,
those never so employed, and those who have left.
The analysis includes as covariates gender and
political conservatism.
22
Only the political con-
servatism variable is signi?cant, and the ?rm-sta-
tus factor is not. The power of this test to detect a
small e?ect is 0.54, and to detect a medium e?ect
is 0.94. To emphasize that no ‘‘tendency’’ exists
towards a pattern consistent with a selection–
socialization bias against ethical big-?rm mem-
bers, note that the highest average P score belongs
to those currently employed by the ‘‘Big 5’’, and
the smallest belongs to those who have left.
As a complementary analysis, we plotted P scores
against year of departure from a ‘‘Big 5/6/8’’ ?rm,
for those who ever have worked for such a ?rm
(n=127). For those still employed by the ‘‘Big 5’’,
we use the year 2001. (Note that everyone in our
sample still was employed in public accounting.)
The trend line is ?at, with r
2
approaching zero.
The power of this regression to detect a relation-
21
Comparisons that include partners in smaller ?rms yield
the same result.
Table 6
Analysis of covariance, ‘‘Big 5’’ ?rms
Position Mean S.D. N
Descriptive statistics (dependent variable: P%)
Manager 36.667 15.740 42
Partner 41.600 13.119 15
Total 37.965 15.140 57
Source Type III sum of squares df Mean square F Sig. Z
2
Tests of between-subjects e?ects (dependent variable: P%)
Corrected Model 2101.654
a
3 700.551 3.459 0.023 0.164
Intercept 6966.802 1 6966.80 34.398 0.000 0.394
SEX 1.699 1 1.699 0.008 0.927 0.000
POLCON 1719.013 1 1719.01 8.488 0.005 0.138
POS 427.860 1 427.860 2.113 0.152 0.038
Error 10,734.3 53 202.534
Total 94,992.0 57
Corrected Total 12,835.9 56
a
R
2
=0.164 (adjusted R
2
=0.116).
22
As with the earlier analyses, age and professional
experience were not signi?cant and are omitted.
556 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
ship in the population with true explained variance
(r
2
) of 0.04 exceeds 0.70, and the power to detect a
relationship with true r
2
of 0.09 exceeds 0.96.
Question 5: does selection–socialization exist in
law ?rms, a similar professional setting?
Of the usable sample of 148 attorneys, 80 were
partners, 63 associates, and only ?ve at lower ranks.
Thus we limited the analysis to partners and associ-
ates. As with accountants, age is not signi?cant in an
ANCOVA, while gender is signi?cant and explains
about 6% of the variance (with the women again
scoring higher). Interestingly, the Collins–Hayes
conservatism score is not a signi?cant covariate for
the attorneys,
23
although their response to the
simple question of whether their political orienta-
tion is ‘‘conservative’’ does explain about 5% of
the variance. As with the accountants, the position
factor is not signi?cant (p=0.77, Z
2
=0.001).
Power is 0.44 to detect a small position e?ect, and
0.84 to detect a medium e?ect.
Attorneys also di?ered from accountants in their
P scores. For the key rank of partner, (80 attorneys
and 146 accountants) average P scores were 45.37
for attorneys and 39.09 for accountants (t=3.15, p
15 years
Experience 38.841 0.000 0.001 0.000 0.000
‘‘Big’’ Firm
Partner >15 years
Experience 43.797 0.000 0.000 0.000 0.000
24
As noted earlier, comparisons to P scores from the DIT2
are not appropriate, precluding use of our 2001 sample in these
tests.
558 S.B. Sco?eld et al. / Accounting, Organizations and Society 29 (2004) 543–563
sponding increase in the far more rudimentary
Stage 3 type reasoning starting at the manager
ranks and continuing into the partner rank. It is
expected that adults would have moved well
beyond the ‘‘self-interest’’ schema of stages 2 and
3 (Rest, Narva´ ez, Bebeau et al., 1999, pp. 36, 54).
Ponemon’s data show higher percentages of the
self-interest schema throughout the ranks, culmi-
nating in a remarkable increase to over 46% in the
partners, dominating their importance decisions
when responding to DIT scenarios. Furthermore,
ethical development, although not strictly ruling
out the usage of more primitive schemas, would
not be expected to show inversions in which an
intermediate stage or schema would appear to be
‘‘skipped.’’ For adults with stunted development,
one would not expect their principled thinking (P)
to exceed their ‘‘conventional,’’ maintaining-
norms thinking (Stage 4). Yet this is the pattern
for both managers and partners in Ponemon’s
data.
The 1995 data, summarized graphically in Fig. 2,
display no signi?cant di?erence between P scores
across professional ranks. The pro?les of scores
for all DIT items are consistent from one profes-
sional level to the next. Stage 2 reasoning and
Stage 3 reasoning are level across all ranks, with a
small proportion (