Description
Eliminating alternative causal explanations plays an important role in establishing causality.
We analyze two strategies researchers use to eliminate alternatives to their preferred
causal explanations: providing persuasive evidence against other plausible explanations
and developing a preferred explanation in such a way as to limit the number of alternatives
against which evidence needs to be provided. Although positivist accounting research aims
at objectivity in the use of these strategies, we argue that subjectivity plays an important
role as well. We identify and discuss relatively more objective and more subjective components
of these strategies for validating and developing causal explanations.
Subjectivity in developing and validating causal explanations
in positivist accounting research
Joan Luft, Michael D. Shields
?
Broad College of Business, Michigan State University, N270 Business Complex, East Lansing, MI 48824, USA
a b s t r a c t
Eliminating alternative causal explanations plays an important role in establishing causal-
ity. We analyze two strategies researchers use to eliminate alternatives to their preferred
causal explanations: providing persuasive evidence against other plausible explanations
and developing a preferred explanation in such a way as to limit the number of alternatives
against which evidence needs to be provided. Although positivist accounting research aims
at objectivity in the use of these strategies, we argue that subjectivity plays an important
role as well. We identify and discuss relatively more objective and more subjective compo-
nents of these strategies for validating and developing causal explanations.
Ó 2013 Elsevier Ltd. All rights reserved.
Introduction
The accounting research that is sometimes labeled as
positivist aims at empirically validating general causal
explanations of accounting-related phenomena—that is,
causal explanations that apply to many instances of a given
phenomenon. This research aims at objectivity, in the
sense that empirical results and the inferences drawn from
them are meant to be independent of the characteristics of
the individual researcher. Thus, results of such research are
intended to be:
replicable by other researchers in the same setting;
reliable across settings that meet the conditions stated
by the relevant theory; and
persuasive within a community of researchers (that is,
the results have the power to change the beliefs of other
researchers in the community).
Paradoxically, the objective development and validation of
causal explanations in this literature are often dependent
on subjective judgments and decisions. We identify key
sources of subjectivity and trace their in?uence on devel-
oping and validating causal explanations.
The remainder of our article is organized as follows: The
next two sections lay the groundwork for our analysis by
providing de?nitions of key terms and identifying the role
of eliminating alternative explanations in establishing cau-
sality. The following sections analyze how researchers
eliminate alternatives to their preferred causal explana-
tions by validating their preferred causal explanations
through persuasive evidence against other plausible alter-
native explanations and/or by developing their preferred
causal explanations in such a way as to limit the number
of alternatives against which evidence needs to be pro-
vided. In these sections we identify important subjective
judgments and decisions that researchers make in both
validating and developing their preferred causal explana-
tions. The ?nal section concludes.
De?nitions
Positivist research
The accounting research that is sometimes labeled as
positivist investigates elements of accounting practice that
are common to many instances rather than the unique
con?guration of common and non-common elements that
0361-3682/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.aos.2013.09.001
?
Corresponding author. Tel.: +1 517 432 2915.
E-mail address: [email protected] (M.D. Shields).
Accounting, Organizations and Society 39 (2014) 550–558
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
occur in any single instance. Researchers attempt to draw
inferences about these common elements from a limited
sample of observations. Thus the validity of inferences
from the particular to the general is a core concern for this
research.
The term ‘‘positivist’’ has been used loosely in account-
ing research, as in other contemporary social science re-
search, often to denote quantitative hypothesis-testing
research. Scholars who do such research often do not ac-
cept the classic positivist program of treating the social sci-
ences as identical to natural sciences and discovering
stable ‘‘laws.’’
1
Nor do they share the position of some ear-
lier major advocates of positivism, from Comte (Andreski,
1974) to Friedman (1953), that social science should aim
at predictions based on observed regularities rather than
an understanding of causal processes.
Objective and subjective
Objectivity in developing and validating causal explana-
tions is often considered important in positivist research.
We use the term ‘‘objectivity’’ in the sense of epistemic,
not ontological, objectivity (Searle, 1995). ‘‘Assertions
(knowledge and judgments) can be considered [epistemi-
cally] objective to the extent that the community in ques-
tion has agreed-upon criteria for evaluating the
justi?cation or evidence for those assertions.’’ (Shapiro,
1997, p. 166). Many accounting phenomena are ontologi-
cally subjective, in the sense that they are socially con-
structed, but analysis of these phenomena can be
epistemically objective. As an example of this distinction,
paper money is ontologically subjective: it is money be-
cause people treat it as money, not because of properties
of the paper that are independent of human judgments
and decisions. But money is epistemically objective: there
are agreed-on criteria for evaluating whether a particular
piece of paper with ‘‘5 dollars’’ on it is really money rather
than a counterfeit, a toy, or a note about the price of tea.
The objective–subjective distinction is a continuum, not
a dichotomy. At the purely objective end of the continuum,
a large community agrees completely about the criteria for
evaluating assertions (e.g., inferences) and applies these
criteria in uniform ways. At the purely subjective end,
there is no agreement about criteria for claiming that one
assertion is more valid than another: diverse individual
judgments cannot be reconciled or ranked. In the
middle range of the continuum are degrees of agreement
that vary both with respect to the breadth of the commu-
nity that agree on criteria and the completeness of their
agreement.
Causality
We assume that the concept of causality in accounting
research is consistent with a probabilistic version of the
counterfactual-conditional account of causality. In this
concept of causality, ‘‘If event x and event y are distinct
actual events, then event y causally depends on event x if
and only if, if event x had not occurred, then the
probability of event y’s happening would be less than if
event x had happened’’. (Menzies, 2009) Thus, for example,
developing an argument that managers’ use of a
particular accounting practice x causes high levels of
performance y means developing an argument that high
levels of performance would be less probable if x were
not used.
This de?nition of causality may seem to exclude many
variables used in accounting research, because accounting
variables are often represented as facts (e.g., precision of
accounting information) rather than events. But in many
cases the ‘‘fact’’ is a summary representation of an event
or set of related events. For example, to say that accounting
information in a given setting has high precision as a mea-
sure of managers’ efforts is to say that a certain set of
events—managers’ effort choices and accountants’ record-
ing and analysis of indicators of these effort choices—has
occurred in this setting.
2
The role of alternative causal explanations
The concept of causality described above has important
implications for developing and validating causal explana-
tions in accounting research, because it makes the role of
alternative explanations for y salient. For example, evidence
that organizations with high performance
use account-
ing practice x more often than low performers does not by
itself provide strong support for an argument that x has a
causal in?uence on y, because it does not provide strong
evidence about other plausible counterfactual condition-
als. That is, it does not provide evidence that the high-per-
forming organizations probably would have had lower
performance if they had not used x, and that the low-per-
forming organizations probably would have had higher
performance if they had used x, other things equal. It is
possible in principle that the high-performing organiza-
tions would have had high performance even without
using x—thus causality cannot be claimed—because there
is an alternative explanation: the higher performance of
these organizations was caused by some other factor z that
tended to co-occur with x.
Because the counterfactual by de?nition is an event that
does not occur, researchers do not have direct evidence of
what would have happened to a given set of organizations
at a given time if they had used different accounting, all
else equal. Much of the process of validating a causal
explanation consists of ?nding or creating high-quality
proxies for these non-existent counterfactuals, such as:
1
For example, Mayntz (2004, p. 239) identi?es a large body of social-
science research that aims ‘‘to avoid the vain search for social laws;’’ and
Engel (2013, p. 6) summarizing behavioral economic research, argues,
‘‘Behavioral researchers have no reason to expect natural laws. They may at
best ?nd typical patterns.’’
2
Some versions of the counterfactual–conditional approach to causation
also allow for the use of variables that cannot be represented as events:
‘‘The [counterfactual conditional] de?nition of causal dependence . . . takes
the primary relata of causal dependence to be events. . . . However, very
different conceptions of events are compatible with the basic de?nition.
Indeed, it even seems possible to formulate it in terms of facts rather than
events (For instance, see Mellor, 1995, 2004.)’’ (Menzies 2009).
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 551
comparing the performance of matched pairs of
organizations that are as similar as possible on all
performance-relevant features except the speci?c
accounting practice;
comparing performance of the same organizations
before and after implementation of the accounting
practice, during time periods that include as few other
confounding events as possible; or
randomly assigning the accounting practice to some
organizations and not others.
3
Thus, an important part of validating a researcher’s pre-
ferred causal explanation (i.e., the explanation that is the
focus of the validation process in a study) consists of elim-
inating alternative causal explanations.
4
Table 1 summa-
rizes the ways in which researchers can aim to eliminate
alternative explanations, and limitations to the objective
accomplishment of these aims. We next provide a brief over-
view of Table 1, followed by more extensive analysis in the
remaining sections of this article.
Researchers can eliminate alternative causal explana-
tions by explicitly providing persuasive evidence against
them(‘‘Validating’’ in Table 1) and/or by reducing the num-
ber of alternatives against which researchers need to pro-
vide evidence (e.g., by specifying the preferred causal
explanationnarrowlysothat fewer alternative explanations
are plausible) (‘‘Developing’’ in Table 1). The success of val-
idating preferred causal arguments by gathering persuasive
evidence can be limited in two important ways. First, re-
search designs that provide strong evidence against one
alternative often provide weak evidence against some other
plausible alternatives: hence a single study with a given de-
sign is unlikely to provide strong evidence against all plau-
sible alternatives, and researchers must trade off one kind
of validity against another. Second, the number of plausible
alternative explanations can be quite large, and providing
evidence against them all can require more data collection,
more complex research designs, and/or more extensive
knowledge on the part of researchers than is feasible.
The strategy of reducing the number of alternatives ex
ante when developing a preferred causal explanation can
help researchers to cope with the problem of large num-
bers of alternatives, but it has the potential to create addi-
tional limitations to the persuasiveness of a study. First,
there is a tradeoff: narrower speci?cation of a preferred
causal explanation can make it easier for researchers to
persuade others that the evidence they have gathered val-
idates their preferred explanation, but narrower speci?ca-
tion can also make it more dif?cult to persuade others that
the study is interesting. Second, plausible alternatives are
sometimes excluded merely because researchers are
unaware of them, especially when these alternatives are
outside the scope of the researchers’ preferred theoretical
perspectives. Third, a study can fail to be persuasive to
readers with different prior beliefs because it does not
report the limitations to inference that result from exclud-
ing alternatives ex ante.
Excluding alternative causal explanations through the
processes of validating and/or developing a preferred cau-
sal explanation includes some judgments and decisions
that are relatively objective. That is, they are located nearer
the objective end of the objective–subjective continuum
and can lead to agreement among researchers with
initially different views about the accounting issue but
similar views about what constitutes persuasive evidence.
For example, researchers with different views about the
effectiveness of monetary incentives may be able to agree
that a given study showing no effect has been able to elim-
inate the alternative explanation ‘‘low statistical power of
test’’ by using a large sample.
As we argue below, however, there are also subjective
judgments and decisions in these research processes about
which researchers are not likely to agree so easily. Multiple
types of validity exist, and a single study typically cannot
maximize all types simultaneously; the tradeoff decisions
that researchers must make in consequence are subjective
to a considerable degree. Moreover, the success of even a
relatively objective process like providing persuasive
evidence against alternatives depends on choices made
during the development of a preferred causal explanation.
Alternatives that are excluded during development will not
be exposed to the potentially more objective processes of
validation; and researchers may disagree, on relatively
subjective bases, about the appropriateness of excluding
speci?c alternatives in the development process.
Validating preferred causal explanations
Because the research we address aims at validating gen-
eral causal explanations based on sets of speci?c observa-
tions, the validity of inference from the particular
observations to a broader population is crucial in assessing
how well a study validates the causal explanations it
makes and invalidates all plausible alternative causal
explanations. The predictive validity framework in Fig. 1
(adapted from Runkel and McGrath (1972)) provides a
widely accepted organizing framework for validating such
inferences.
5
Inferences about the conceptual (theoretical)
relation between the independent and dependent variables
(link 1) are made by validating link 4 between the
operationalized independent and dependent variables. Valid
inference about the conceptual relation also requires that
links 2 and 3 between the conceptual and operational
3
Although random assignment is often not feasible at the level of
independent organizations in the natural environment, it can be used with
organizational subunits in ?eld experiments or with laboratory ‘‘organiza-
tions’’ consisting of small numbers of individuals.
4
Note that ‘‘explanation’’ does not necessarily mean ‘‘single-factor
explanation,’’ and thus excluding alternative explanations does not mean
rejecting complex multi-factor causality. An explanation can be a model
specifying multiple causal factors and the forms of their relations with each
other and the dependent variable of interest (see Luft & Shields, 2003 for
more detail). Alternative explanations would then be models specifying
different sets of causal factors and/or forms of relation.
5
This framework was introduced to the accounting literature by Libby
(1976) and used by Libby, Bloom?eld, and Nelson (2002) to analyze
experimental research in ?nancial accounting. Note that, in order to present
validity issues clearly, this ?gure represents a simple causal relation
without interactions, endogeneity, etc. The causal diagram can readily be
modi?ed to include more complex relations (see Luft & Shields, 2003;
2007).
552 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
variables are valid and that other variables that can in?u-
ence the dependent variable are controlled for in link 5.
Validating these links requires researchers to provide
evidence that an apparent association between the opera-
tionalized independent and dependent variables is more
consistent with their preferred causal explanation (link 1)
than with other alternatives such as the following:
The apparent association between the operationalized
variables is not caused by a real association but by vio-
lations of the assumptions of the statistical tests
employed.
The real cause of variation in the dependent variable is
not the conceptual independent variable in the
researcher’s preferred causal explanation but a different
variable that the (poorly chosen) operationalized inde-
pendent variable actually captures.
The operationalized variables are associated with each
other, not because x causes y (as in the researcher’s pre-
ferred causal explanation), but because y causes x, or x
and y have a common cause z.
The study makes a general claim that x causes y, but a
more valid alternative explanation is that x causes y
only under certain conditions that were present in the
context of the study but are not universally present.
In terms of the widely used threats-to-validity frame-
work (Shadish, Cook, & Campbell, 2002), these alternative
explanations represent threats to statistical conclusion
validity, construct validity, internal validity, and external
validity, respectively. Because these threats are different,
created by different alternative explanations, the research
processes that reduce one threat will not necessarily
reduce—and may in fact increase—other threats. The exis-
tence of such tradeoffs creates important limitations to
researchers’ ability to validate their preferred causal expla-
nations through relatively objective process of providing
persuasive evidence (limitation A1 in Table 1).
Tradeoffs among validity types
The design of any particular study—for example, its
choice of research method and data sources—involves
tradeoffs among the different types of threats to validity.
For example, the choice of using a laboratory experiment
typically reduces threats to internal validity through
random assignment of sampling units to experimental
conditions but can raise questions about construct or
external validity.
6
Similarly, when researchers use archival
Table 1
Validating and developing researchers’ preferred causal explanations as a process of eliminating alternative causal explanations.
A. Validating preferred causal explanations
Aim:
Provide persuasive evidence that invalidates all plausible alternative causal explanations
Limitations to objective achievement of aim:
(1) Validity tradeoffs. Providing more persuasive evidence against one alternative causal explanation often entails providing less persuasive
evidence against some other plausible alternatives
(2) Numerous plausible alternative explanations. Providing evidence against all plausible alternatives would require more data collection, more
complex research designs, and/or more extensive theoretical knowledge than is feasible
B. Developing preferred causal explanations
Aim:
Reduce the number of alternative causal explanations against which persuasive evidence must be gathered, by narrow speci?cation of context,
causal-chain segment, etc.
Limitations to objective achievement of aim:
(1) Reduction in researchers’ ability to persuade others that their preferred causal explanations are interesting when they are very narrowly
speci?ed
(2) Exclusion of plausible alternatives merely because they are outside of the researchers’ preferred theoretical perspectives
(3) Failure to report limits of inference. Studies that provide evidence consistent with a preferred causal explanation do not necessarily provide
evidence against other explanations if these other explanations have been excluded in the development process
Fig. 1. Predictive validity framework (adapted from Runkel and McGrath (1972)).
6
For example, questions can arise about whether performance reports
provided to laboratory experiment participants capture the same theoret-
ical properties as a set of performance reports in natural environments
(construct validity). Questions can also arise about how large a set of
performance reports in natural environments actually have these proper-
ties (external validity).
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 553
and survey data, their sample choices create validity trade-
off choices. For example, a more diverse sample can increase
external validity by enabling researchers to test for interac-
tions, but can increase threats to statistical conclusion valid-
ity and construct validity.
7
In consequence of these design choices, the validation
process will be able to exclude some alternative causal
explanations persuasively, at the cost of providing only
limited evidence against some other alternatives. Judg-
ments about acceptable tradeoffs among threats to validity
are more subjective than judgments about whether a spe-
ci?c threat (e.g., low statistical power, an interaction) has
been eliminated. There is no common measure of validity
across the types that would allow researchers to provide
evidence that they are giving up only v units of one kind
of validity to gain v + n of another kind of validity. More-
over, as Shadish et al. (2002) argue, the relative importance
of different kinds of validity can vary across research con-
texts, such as basic versus applied research.
8
But there is no
standard set of weights for the different validity types that
would move researchers with different initial judgments to-
ward agreement about whether the gain is greater than the
loss in any particular tradeoff among validity types.
Because there are neither summary measures nor
agreed-on weights for the different types of validity, judg-
ments about the acceptability of particular tradeoffs are
relatively subjective. Perhaps in consequence (since tradi-
tion can substitute for objective validation as a basis for
agreement), these judgments are much in?uenced by the
traditions of sub-communities of researchers. For example,
some sub-communities are more likely than others to ac-
cept p = .06 as adequate support for a hypothesis. Similarly,
some sub-communities are more likely to be more con-
cerned about the threats to external validity in laboratory
experiments or single-site ?eld studies, while others are
likely to be more concerned about threats to internal and
construct validity in large-sample archival and survey
studies.
Numerous plausible alternative explanations
Identifying all plausible alternative explanations is a
challenge because the social sciences offer a very diverse
and extensive range of theories that are potentially rele-
vant to accounting. Economics, psychology, and sociology
theories are commonly employed in accounting research
(Chapman, Hopwood, & Shields, 2007), and other
disciplines such as anthropology, history, political science,
and neuroscience can also play a role. Moreover, the diver-
sity of relevant theories within each discipline is large. (See
Birnberg, Luft, & Shields, 2007 for an example of the diver-
sity of psychology theories relevant to accounting.)
9
Not all social science theories will be direct competitors
in explaining a given accounting phenomenon, but often
the number of possible explanations is nontrivially large.
Ittner, Larcker, and Meyer (2003) provide a good example,
identifying eight plausible explanations (two based on eco-
nomics theories and six based on psychology theories) for
the magnitudes of incentive weights on performance mea-
sures in a large bank’s balanced scorecard. A longer list of
plausible alternatives could be composed for many
accounting research questions. For example, Table 2
provides a (non-exhaustive) list of 14 plausible causal
explanations (E1–E14) of why some decision makers
would rely more heavily on apparently relevant accounting
information in making a particular decision than others do.
The plausible alternatives to a researcher’s preferred
causal explanation can be so numerous and diverse that
the process of validating the preferred explanation by pro-
viding evidence against all identi?ed alternatives can, in
principle, become unmanageable (limitation A2 in Table 1).
More plausible explanations provided by more diverse the-
ories often mean that more different independent variables
are potentially relevant. Using a larger number of variables
is likely to require more data collection, more complex
research designs to take into account the relations among
these variables, and/or more knowledge on the part of
researchers to deal with the resulting theoretical and
methodological issues. Meetings these requirements can
be costly or infeasible, and hence researchers often use
the process of developing their preferred causal explana-
tions to reduce the number of alternative plausible expla-
nations against which they need to provide evidence.
Developing preferred causal explanations
A broad-based literature search will often generate a
large number of plausible causal explanations related to
a particular research question, like the 14 alternatives that
appear in Table 2. How many of these are in fact competing
explanations against which evidence must be gathered de-
pends on how the researchers’ preferred causal explana-
tions are developed and thus how they are speci?ed in
their ?nal form. As explained in more detail below, some
criteria for excluding alternatives through development
are shared by many researchers, even though they differ
about speci?c accounting issues; and thus eliminating
alternative explanations by these means can be regarded
as relatively objective. However, as with validation, subjec-
tive judgments play a key role in developing preferred cau-
sal explanations and can create limitations to researchers’
success in persuading others who initially hold different
views about whether the researchers’ preferred explana-
tion is valid.
7
Larger samples can have more uncontrolled variation that reduces the
power of tests (statistical conclusion validity), for example when data are
collected from a larger number of ?rms and/or industries that vary in ways
not captured adequately by control variables. Larger samples can also
decrease the likelihood that a given operational variable is really capturing
the same construct in the diverse units of observation in the sample (e.g.,
does activity-based costing have the same essential properties in all the
different organizations in the sample?).
8
For example, ‘‘Basic researchers have high interest in construct validity
because of the key role that constructs play in theory construction and
testing. Applied researchers tend to have more interest in external validity
. . .’’ (Shadish et al. 2002, p. 100).
9
Balakrishnan and Penno (2013) illustrate the combinatorial explosion
of alternative causal explanations when many explanatory factors are
available and researchers do not include all possible factors in their
preferred model.
554 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
Plausible alternative explanations can be eliminated
relatively objectively by specifying the preferred causal
explanation more narrowly, so that explanations which
would be plausible alternatives to a broadly stated pre-
ferred explanation become logically inapplicable to the
narrower version of the explanation. For example, the
statement of the preferred causal explanation can explic-
itly limit the context to which it is intended to apply—orga-
nizations or markets of a certain size or structure,
individuals with a certain level of expertise or motivation,
short-term or long-term variation only,
10
etc.—in order to
exclude explanations based on variation in these individual,
market, or organizational characteristics.
Research settings are then chosen to be consistent with
the narrowly speci?ed context. For example, researchers
who are interested in the effects of noise in accounting
information (E1 in Table 2) might want to avoid the need
to measure and model individual variation in risk attitudes
(E2), estimation error (E3), and/or mental models (E4), be-
cause their research designs are already complex, or they
doubt that the available theoretical or empirical models
and/or operationalization of constructs will allow them
to control effectively for individual variation from all these
sources.
To avoid dealing with these variables, researchers can
develop a noise-based explanation that explicitly narrows
its applicability by stating assumptions about alternative
explanations that limit their in?uence. For example, the
explanation can be speci?ed to apply only to settings
where (a) risk aversion is high enough that the uncertainty
created by noise will matter, (b) estimation error is low
enough that individuals can distinguish higher from lower
noise reliably, and/or (c) individuals’ mental models are
similar enough that the implications of different noise lev-
els for individual decisions are reasonably similar across
individuals and predictable by the researcher. The research
setting then needs to be matched to such assumptions: for
example, a setting might be found where individuals rely
on a trusted system for key estimates rather than making
their own estimates with their own idiosyncratic errors.
Another relatively objective way to reduce the number
of plausible alternatives through development of the pre-
ferred explanation is to specify only a narrow segment of
the causal chain, to which only some alternative explana-
tions are logically applicable. For example, accounting
information received by different sets of users can be dif-
ferentially noisy (E1) because it is produced by different
technologies (E14): there is a causal chain leading from
technology to noise to information use. Researchers may
specify that they are only interested in the effects of noise
and not in its causes. Given this speci?cation, technology is
not a factor that competes with noise to explain the use or
non-use of accounting information, and researchers do not
need to collect information on technology variation to val-
idate their preferred causal explanation.
Reducing the number of plausible alternatives through
narrow speci?cation often contributes to the effectiveness
and ef?ciency of research design. It enables researchers to
document the explanatory power of their preferred causal
explanation without using inordinately complex research
designs and statistical analyses to control for or eliminate
alternatives. Narrow speci?cation also can create limita-
tions, however (see Table 1, Part B). Narrow speci?cation
often raises questions about how interesting a limited con-
text or limited segment of a causal chain is (limitation B1).
Sometimes researchers exclude alternative explanations in
Table 2
Selected alternative explanations for why some decision makers rely more on apparently relevant accounting information than others.
Difference between users and non-users of relevant accounting information Relevant theory
E1. Accounting information received by users is less noisy than that received by non-users Economics,
psychology
E2. Users of (noisy) accounting information are less risk-averse than non-users Economics,
psychology
E3. Non-users overestimate and/or users underestimate the noise in accounting information Psychology
E4. Accounting information plays a smaller role in non-users’ than in users’ mental models Psychology
E5. Leadership style of users and/or their superiors is more accounting-oriented Psychology
E6. Users (non-users) work in environments where reliance on accounting is (is not) consistent with social or professional
norms
Sociology
E7. Users have more accounting training than non-users and therefore are more con?dent that they can use the information
effectively
Psychology
E8. Users belong to a national culture that is more compatible with strong reliance on accounting information than non-users Psychology
E9. Non-users have better access to high-quality information that substitutes for accounting than users do Economics
E10. Users have better access to complementary information that improves the usability of the accounting information Economics
E11. Users have more ego-involvement with the accounting because they have been more involved in creating the speci?c
accounting measures than non-users have
Psychology
E12. Users (non-users) work in settings where powerful coalitions that would be advantaged (disadvantaged) by reliance on
accounting have encouraged (limited) its use in decisions
Economics,
sociology
E13. Users, more than non-users, work in settings where the use of accounting is a signal of skills that are important for
promotion to higher-level jobs
Economics
E14. Users work in settings with more advanced technology, which provides more up-to-date and/or more transparently
presented versions of the accounting information
Various theories
Notes: 1. E = Explanation.
2. See Hartmann (2000) for a review of literature that proposes and investigates many of these explanations.
10
See Luft and Shields (2003, 2007) for guidelines on making time-frame
and level of analysis choices.
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 555
the development process not because they are logically
implausible but merely because they are outside the
researchers’ preferred theoretical perspectives (limitation
B2). Finally, researchers sometimes fail to report
adequately the limits to inference that result from their
exclusion of alternative explanations in the development
process (limitation B3).
Choices of which alternatives to exclude and which lim-
itations to report often depend on researchers’ choices of
theoretical perspectives, which are highly subjective deci-
sions. Researchers often make these decisions early in their
careers, based on their preferences, abilities, and experi-
ences, as well as on institutional forces such as the avail-
ability of training and research support. There is no
objective way of determining (for example) whether sociol-
ogy is in general a ‘‘better’’ theoretical perspective as a basis
for accounting research than psychology, or vice versa.
11
Because a choice of theoretical perspective is typically a
long-term choice to build expertise in one particular set of
theories and research methods and not others, it tends to
constrain subsequent, more study-speci?c choices. Thus,
some of the research choices that follow from a choice of
theoretical perspective are also highly subjective and can-
not meaningfully be labeled right or wrong. Just as it is not
‘‘wrong’’ to be an economist and ‘‘right’’ to be an anthro-
pologist, so the choice of primary objects of study (e.g.,
individual versus social processes, markets versus organi-
zations) cannot be objectively right or wrong in general:
such choices are not subject to agreed-on criteria that
would determine which of these objects of study it is ‘‘bet-
ter’’ to know about. In the following subsections we ana-
lyze the three limitations presented in Table 1, Part B,
indicating how each of these is linked to researchers’ sub-
jective choices of theoretical perspectives.
Narrow speci?cation and interestingness
Choice of theoretical perspective is a strong subjective
in?uence on researchers’ judgments about the interesting-
ness of any given research question and the causal expla-
nation that is proposed to answer it. In Davis’ (1971)
widely cited de?nition, interesting studies are those that
share most of their audience’s assumption-ground but
deny some particular assumption.
12
Thus, judgments of
the interestingness of research questions are highly audi-
ence-speci?c: ‘‘Propositions are interesting or uninteresting
only in relation to the assumption-ground of some audi-
ence.’’ (Davis, 1971, p. 32). In such cases there is no objective
method, transcending audience assumption-grounds, of
determining which audience’s preferred questions are
‘‘more interesting.’’
As researchers narrow the applicability of their pre-
ferred causal explanations in order to improve their
chances of success in validation, they risk making their
studies less interesting to some potential audiences, be-
cause narrower explanations may no longer be a good ?t
to the assumption-ground of these audiences. For example,
suppose a preferred causal explanation focuses only on the
individual level of analysis and excludes explanations at
organizational and social levels. (Note that this does not
mean denying the relevance of organizational and social
levels of explanation; it simply means that these levels
are outside the scope of investigation in the particular
study.) An individual-level explanation is likely not to
seem interesting, and thus not valuable and persuasive,
to researchers whose preferred theoretical perspective
includes the assumption that explanations should focus
primarily on higher levels of analysis.
Researchers often strive to interest and persuade rela-
tively broad audiences (not just their few closest col-
leagues), and thus they can be willing to make a tradeoff
in developing their preferred causal explanation. Rather
than maximizing their chances of validation success by
specifying their preferred explanations very narrowly, they
trade off chances of validation success against chances of
interesting a broad(er) audience. But just as there is often
no objective method of determining which audience’s pre-
ferred questions are more interesting, there is also no
objective method of determining what the optimum point
is in such tradeoffs. Hence beliefs can sometimes differ in
objectively unresolvable ways about whether a given study
has made an appropriate tradeoff of this kind.
In the two following sections, we argue that excluding
alternative explanations in the development process some-
times runs the risk of slipping from justi?able narrow
speci?cation into two less justi?able research practices:
excluding plausible alternatives merely because they are
outside of the researcher’s preferred theoretical
perspective, and failing to report the limitations of the
inferences that can be made when alternatives are
excluded through narrow speci?cation of preferred causal
explanations.
Exclusions based on researchers’ preferred theoretical
perspective
Researchers trained in one theoretical perspective
sometimes have limited acquaintance with other perspec-
tives. Insofar as plausible alternative causal explanations
are excluded from consideration in a study simply on the
basis of convenience—that is, alternatives are excluded be-
cause the researcher is neither aware of themnor equipped
to deal with them—it is unlikely to inspire con?dence in
the validity of the study’s inferences. Insofar as exclusions
are based on a belief that causal explanations from other
theoretical perspectives are so generally implausible or
uninteresting that they do not need to be considered
explicitly, it is also likely to limit the persuasive power of
the study.
11
Max Weber identi?es the paradox that scientists (among whom he
includes natural scientists, social scientists, and scholars of the humanities)
all believe that their particular scienti?c activities are especially worth-
while, ‘‘But they cannot prove scienti?cally that this is the case.’’ (Weber,
1958, p. 145).
12
Similarly, Bartunek, Rynes, and Ireland’s (2006) survey of editorial
board members of the Academy of Management Journal found that the most
frequently given reason for why an article is interesting is that it is
counterintuitive, in that it challenges established theories, is contrary to
folk wisdom, or creates an ‘‘aha’’ moment. But see Towry (2012, p. 26) for
important caveats to this view: for example, the ‘‘risk that we will
encourage poorly developed, convoluted theories, as authors embark on a
‘search for surprise.’’’
556 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
Accounting research increasingly does not take this
approach but explicitly considers multiple theoretical
perspectives in individual studies. For example, Libby
et al. (2002) describe how research studies that integrate
theories from economics and psychology have contributed
to a better understanding of ?nancial accounting by spec-
ifying more clearly the mechanisms affecting individual
and market behavior. In management accounting research,
Covaleski, Evans, Luft, and Shields (2003, 2007) identify
opportunities and provide guidelines for budgeting
research that integrates theories from economics, psychol-
ogy and sociology.
Failure to report limits of inference
Studies that do not adequately report their limitations
are not likely to be persuasive, because they claim broader
applicability or more certainty than can be justi?ed by the
evidence they provide. The limitations that need to be re-
ported depend in part on the choices researchers make in
developing their preferred causal explanations. For exam-
ple, when researchers specify a preferred causal explana-
tion narrowly so that it applies only to a particular
context (e.g., time frame, organization type), they need to
make clear to readers that their study makes no inferences
about other contexts. Failure to report such limitations can
easily arise when plausible alternatives are outside of
researchers’ (subjectively) chosen theoretical perspectives.
If these alternatives are excluded in development simply
because researchers are unaware of them or unprepared
to deal with them, then researchers will also be unaware
of the resulting limitations.
Even when researchers are aware of plausible
alternatives and have excluded them via narrow speci?-
cation, they sometimes fail to make it clear that their
studies provide no evidence about these alternative
explanations. For example, evidence of the importance
of individual-level variables is not by itself evidence
against the importance of culture or vice versa. Individ-
ual-level studies often hold culture constant and thus
provide no evidence about effects of cultural variation.
Similarly, evidence of strong effects of monetary incen-
tives in settings where social norms have little vari-
ance—or evidence of strong social-norm effects in
settings with little monetary-incentive variance—does
not enable researchers to make inferences about the rela-
tive importance of money versus social norms. Whether a
study is informative about causal explanations other than
the researcher’s preferred explanation depends on
whether other explanations were even considered and
whether the (often limited) research setting chosen was
one in which these other explanations could plausibly
have had signi?cant effects.
Conclusion
Philosophers have commented on the subjectivity of
empirical inquiry even in the natural sciences:
There are, of course, general maxims for empirical
enquiry—for example, the very fact that we speak of
empirical enquiry re?ects one of them: ‘Don’t try to
?gure out in a purely a priori way how nature works.’. . .
But which theories we should test . . .; when a theory has
been suf?ciently tested to warrant provisional accep-
tance, and when it has been tested enough to be relied
on, at least until a better theory comes along; these are
all matters which in practice scientists decide partly on
the basis of tradition . . . and partly on the basis of
intuition. (Magee, 1978, p. 232).
Subjective intuition can be a source of creativity, and tradi-
tion can support potentially useful regularities of scholarly
practice when no logically or evidentially compelling
reasons for regularity are available.
13
But reliance on rela-
tively objective—explicit, shared, logically and evidentially
based—criteria remains an ideal in positivist accounting
research.
A widely accepted strategy for implementing the posi-
tivist ideal relies on using the predictive validity frame-
work (Fig. 1) and explicitly and systematically addressing
threats to the validity of empirical inferences. Addressing
these threats means validating a researcher’s preferred ca-
sual explanation by providing evidence against plausible
alternatives and/or developing the preferred causal expla-
nation in such as way as to limit the number of alternatives
about which evidence must be provided. Insofar as these
approaches to developing and validating causal explana-
tions make acceptance of a study’s inferences more persua-
sive to a broad community of researchers and thus less
researcher-dependent, they increase the objectivity of
research.
The positivist ideal of objectivity also includes explicit
awareness and reporting of the subjective judgments and
decisions involved in developing causal explanations and
making research-design choices. Because these judgments
and decisions can exclude alternative plausible explana-
tions from being considered and can reduce the evidence
gathered against these alternatives, they constitute impor-
tant limitations on validation of a study’s inferences.
Reporting these limitations—that is, reporting the (often
unavoidably) subjective nature of developing and validat-
ing causal explanations—can, perhaps paradoxically, in-
crease the objectivity of a research study. Researchers
who espouse different, and potentially con?icting, speci?c
theories are more likely to agree about the validity of the
inferences that can be drawn from a particular study—that
is, the inferences will be less researcher-dependent and
thus less subjective—if the limitations of these inferences
are fully reported.
Acknowledgments
We would like to thank Chris Chapman, Will Demere,
and Tyler Thomas for their comments on an earlier version
of our article.
13
For example, it is not evident that p < .05 is the optimal cutoff for
acceptable inferential results, but it is ef?cient for the scholarly community
to accept such a common cutoff rather than argue about it in the review
process of every hypothesis-testing paper submitted to journals.
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 557
References
Andreski, B. (Ed.). (1974). The essential Comte: Selected from the cours de
philosophie positive. New York: Barnes & Noble.
Balakrishnan, R., & Penno, M. (2013). Causality in the context of analytical
models and numerical experiments. Accounting, Organizations, and
Society.
Bartunek, J., Rynes, S., & Ireland, R. D. (2006). What makes management
research interesting, and why does it matter? Academy of
Management Journal, 49, 9–15.
Birnberg, J., Luft, J., & Shields, M. (2007). Psychology theory in
management accounting research. In C. Chapman, A. Hopwood, &
M. Shields (Eds.). Handbook of management accounting research (Vol.
1). Oxford: Elsevier.
Chapman, C., Hopwood, A., & Shields, M. (Eds.). (2007). Handbook of
management accounting research (Vol. 1). Oxford: Elsevier.
Covaleski, M., Evans, J., Luft, J., & Shields, M. (2003). Budgeting research:
Three theoretical perspectives and criteria for selective integration.
Journal of Management Accounting Research, 15, 3–49.
Covaleski, M., Evans, J., Luft, J., & Shields, M. (2007). Budgeting research:
Three theoretical perspectives and criteria for selective integration. In
C. Chapman, A. Hopwood, & M. Shields (Eds.). Handbook of
management accounting research (Vol. 2). Oxford: Elsevier.
Davis, M. (1971). That’s interesting! Towards a phenomenology of
sociology and a sociology of phenomenology. Philosophy of the Social
Sciences, 1, 309–344.
Engel, C. (2013). Behavioral law and economics: Empirical methods. MPI
Collective Goods Preprint, No. 2013/1. .
Friedman, M. (1953). Essays in positive economics. Chicago: University of
Chicago Press.
Hartmann, F. (2000). The appropriateness of RAPM: Toward the further
development of theory. Accounting, Organizations and Society, 25,
451–482.
Ittner, C. D., Larcker, D. F., & Meyer, M. W. (2003). Subjectivity and the
weighting of performance measures: Evidence from a balanced
scorecard. The Accounting Review, 78, 725–756.
Libby, R. (1976). Discussion of cognitive changes induced by accounting
changes: Experimental evidence on the functional ?xation
hypothesis. Journal of Accounting Research Supplement, 14, 18–24.
Libby, R., Bloom?eld, R., & Nelson, M. (2002). Experimental research in
?nancial accounting. Accounting, Organizations and Society, 27,
775–810.
Luft, J., & Shields, M. (2003). Mapping management accounting: Graphics
and guidelines for theory-consistent empirical research. Accounting,
Organizations and Society, 28, 169–249.
Luft, J., & Shields, M. (2007). Mapping management accounting: Graphics
and guidelines for theory-consistent empirical research. In C.
Chapman, A. Hopwood, & M. Shields (Eds.). Handbook of
management accounting research (Vol. 1). Oxford: Elsevier.
Magee, B. (Ed.). (1978). Men of ideas. NY: Viking Press.
Mayntz, R. (2004). Mechanisms in the analysis of social macro-
phenomena. Philosophy of the Social Sciences, 34, 237–259.
Mellor, D. (1995). The facts of causation. London: Routledge.
Mellor, D. H. (2004). For facts as causes and effects. In J. Collins, E. Hall, &
L. Paul (Eds.). Causation and Counterfactuals. Cambridge, Mass: MIT
Press.
Menzies, P. (2009). Counterfactual theories of causation. In E. Zalta (Ed.).
The Stanford encyclopedia of philosophy. Fall 2009 Edition. .
Runkel, P., & McGrath, J. (1972). Research on human behavior: A systematic
guide to method. New York: Holt, Rinehart and Winston.
Searle, J. (1995). The social construction of reality. New York: The Free
Press.
Shadish, W., Cook, T., & Campbell, D. (2002). Experimental and quasi-
experimental designs for generalized causal inference. Belmont, CA:
Wadsworth.
Shapiro, B. P. (1997). Objectivity, relativism, and truth in external
?nancial reporting: What’s really at stake in the dispute. Accounting,
Organizations and Society, 22, 165–185.
Towry, K. (2012). Discussion of ‘‘Subordinates as the ?rst line of defense
against biased ?nancial reporting’’. Journal of Management Accounting
Research, 24, 25–30.
Weber, M. (1958). Science as a vocation. In H. Green, & C. Wright Mills
(Trans. & Eds.), From Max Weber: Essays in sociology. New York: Oxford
University Press.
558 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
doc_565473330.pdf
Eliminating alternative causal explanations plays an important role in establishing causality.
We analyze two strategies researchers use to eliminate alternatives to their preferred
causal explanations: providing persuasive evidence against other plausible explanations
and developing a preferred explanation in such a way as to limit the number of alternatives
against which evidence needs to be provided. Although positivist accounting research aims
at objectivity in the use of these strategies, we argue that subjectivity plays an important
role as well. We identify and discuss relatively more objective and more subjective components
of these strategies for validating and developing causal explanations.
Subjectivity in developing and validating causal explanations
in positivist accounting research
Joan Luft, Michael D. Shields
?
Broad College of Business, Michigan State University, N270 Business Complex, East Lansing, MI 48824, USA
a b s t r a c t
Eliminating alternative causal explanations plays an important role in establishing causal-
ity. We analyze two strategies researchers use to eliminate alternatives to their preferred
causal explanations: providing persuasive evidence against other plausible explanations
and developing a preferred explanation in such a way as to limit the number of alternatives
against which evidence needs to be provided. Although positivist accounting research aims
at objectivity in the use of these strategies, we argue that subjectivity plays an important
role as well. We identify and discuss relatively more objective and more subjective compo-
nents of these strategies for validating and developing causal explanations.
Ó 2013 Elsevier Ltd. All rights reserved.
Introduction
The accounting research that is sometimes labeled as
positivist aims at empirically validating general causal
explanations of accounting-related phenomena—that is,
causal explanations that apply to many instances of a given
phenomenon. This research aims at objectivity, in the
sense that empirical results and the inferences drawn from
them are meant to be independent of the characteristics of
the individual researcher. Thus, results of such research are
intended to be:
replicable by other researchers in the same setting;
reliable across settings that meet the conditions stated
by the relevant theory; and
persuasive within a community of researchers (that is,
the results have the power to change the beliefs of other
researchers in the community).
Paradoxically, the objective development and validation of
causal explanations in this literature are often dependent
on subjective judgments and decisions. We identify key
sources of subjectivity and trace their in?uence on devel-
oping and validating causal explanations.
The remainder of our article is organized as follows: The
next two sections lay the groundwork for our analysis by
providing de?nitions of key terms and identifying the role
of eliminating alternative explanations in establishing cau-
sality. The following sections analyze how researchers
eliminate alternatives to their preferred causal explana-
tions by validating their preferred causal explanations
through persuasive evidence against other plausible alter-
native explanations and/or by developing their preferred
causal explanations in such a way as to limit the number
of alternatives against which evidence needs to be pro-
vided. In these sections we identify important subjective
judgments and decisions that researchers make in both
validating and developing their preferred causal explana-
tions. The ?nal section concludes.
De?nitions
Positivist research
The accounting research that is sometimes labeled as
positivist investigates elements of accounting practice that
are common to many instances rather than the unique
con?guration of common and non-common elements that
0361-3682/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.aos.2013.09.001
?
Corresponding author. Tel.: +1 517 432 2915.
E-mail address: [email protected] (M.D. Shields).
Accounting, Organizations and Society 39 (2014) 550–558
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
occur in any single instance. Researchers attempt to draw
inferences about these common elements from a limited
sample of observations. Thus the validity of inferences
from the particular to the general is a core concern for this
research.
The term ‘‘positivist’’ has been used loosely in account-
ing research, as in other contemporary social science re-
search, often to denote quantitative hypothesis-testing
research. Scholars who do such research often do not ac-
cept the classic positivist program of treating the social sci-
ences as identical to natural sciences and discovering
stable ‘‘laws.’’
1
Nor do they share the position of some ear-
lier major advocates of positivism, from Comte (Andreski,
1974) to Friedman (1953), that social science should aim
at predictions based on observed regularities rather than
an understanding of causal processes.
Objective and subjective
Objectivity in developing and validating causal explana-
tions is often considered important in positivist research.
We use the term ‘‘objectivity’’ in the sense of epistemic,
not ontological, objectivity (Searle, 1995). ‘‘Assertions
(knowledge and judgments) can be considered [epistemi-
cally] objective to the extent that the community in ques-
tion has agreed-upon criteria for evaluating the
justi?cation or evidence for those assertions.’’ (Shapiro,
1997, p. 166). Many accounting phenomena are ontologi-
cally subjective, in the sense that they are socially con-
structed, but analysis of these phenomena can be
epistemically objective. As an example of this distinction,
paper money is ontologically subjective: it is money be-
cause people treat it as money, not because of properties
of the paper that are independent of human judgments
and decisions. But money is epistemically objective: there
are agreed-on criteria for evaluating whether a particular
piece of paper with ‘‘5 dollars’’ on it is really money rather
than a counterfeit, a toy, or a note about the price of tea.
The objective–subjective distinction is a continuum, not
a dichotomy. At the purely objective end of the continuum,
a large community agrees completely about the criteria for
evaluating assertions (e.g., inferences) and applies these
criteria in uniform ways. At the purely subjective end,
there is no agreement about criteria for claiming that one
assertion is more valid than another: diverse individual
judgments cannot be reconciled or ranked. In the
middle range of the continuum are degrees of agreement
that vary both with respect to the breadth of the commu-
nity that agree on criteria and the completeness of their
agreement.
Causality
We assume that the concept of causality in accounting
research is consistent with a probabilistic version of the
counterfactual-conditional account of causality. In this
concept of causality, ‘‘If event x and event y are distinct
actual events, then event y causally depends on event x if
and only if, if event x had not occurred, then the
probability of event y’s happening would be less than if
event x had happened’’. (Menzies, 2009) Thus, for example,
developing an argument that managers’ use of a
particular accounting practice x causes high levels of
performance y means developing an argument that high
levels of performance would be less probable if x were
not used.
This de?nition of causality may seem to exclude many
variables used in accounting research, because accounting
variables are often represented as facts (e.g., precision of
accounting information) rather than events. But in many
cases the ‘‘fact’’ is a summary representation of an event
or set of related events. For example, to say that accounting
information in a given setting has high precision as a mea-
sure of managers’ efforts is to say that a certain set of
events—managers’ effort choices and accountants’ record-
ing and analysis of indicators of these effort choices—has
occurred in this setting.
2
The role of alternative causal explanations
The concept of causality described above has important
implications for developing and validating causal explana-
tions in accounting research, because it makes the role of
alternative explanations for y salient. For example, evidence
that organizations with high performance
use account-ing practice x more often than low performers does not by
itself provide strong support for an argument that x has a
causal in?uence on y, because it does not provide strong
evidence about other plausible counterfactual condition-
als. That is, it does not provide evidence that the high-per-
forming organizations probably would have had lower
performance if they had not used x, and that the low-per-
forming organizations probably would have had higher
performance if they had used x, other things equal. It is
possible in principle that the high-performing organiza-
tions would have had high performance even without
using x—thus causality cannot be claimed—because there
is an alternative explanation: the higher performance of
these organizations was caused by some other factor z that
tended to co-occur with x.
Because the counterfactual by de?nition is an event that
does not occur, researchers do not have direct evidence of
what would have happened to a given set of organizations
at a given time if they had used different accounting, all
else equal. Much of the process of validating a causal
explanation consists of ?nding or creating high-quality
proxies for these non-existent counterfactuals, such as:
1
For example, Mayntz (2004, p. 239) identi?es a large body of social-
science research that aims ‘‘to avoid the vain search for social laws;’’ and
Engel (2013, p. 6) summarizing behavioral economic research, argues,
‘‘Behavioral researchers have no reason to expect natural laws. They may at
best ?nd typical patterns.’’
2
Some versions of the counterfactual–conditional approach to causation
also allow for the use of variables that cannot be represented as events:
‘‘The [counterfactual conditional] de?nition of causal dependence . . . takes
the primary relata of causal dependence to be events. . . . However, very
different conceptions of events are compatible with the basic de?nition.
Indeed, it even seems possible to formulate it in terms of facts rather than
events (For instance, see Mellor, 1995, 2004.)’’ (Menzies 2009).
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 551
comparing the performance of matched pairs of
organizations that are as similar as possible on all
performance-relevant features except the speci?c
accounting practice;
comparing performance of the same organizations
before and after implementation of the accounting
practice, during time periods that include as few other
confounding events as possible; or
randomly assigning the accounting practice to some
organizations and not others.
3
Thus, an important part of validating a researcher’s pre-
ferred causal explanation (i.e., the explanation that is the
focus of the validation process in a study) consists of elim-
inating alternative causal explanations.
4
Table 1 summa-
rizes the ways in which researchers can aim to eliminate
alternative explanations, and limitations to the objective
accomplishment of these aims. We next provide a brief over-
view of Table 1, followed by more extensive analysis in the
remaining sections of this article.
Researchers can eliminate alternative causal explana-
tions by explicitly providing persuasive evidence against
them(‘‘Validating’’ in Table 1) and/or by reducing the num-
ber of alternatives against which researchers need to pro-
vide evidence (e.g., by specifying the preferred causal
explanationnarrowlysothat fewer alternative explanations
are plausible) (‘‘Developing’’ in Table 1). The success of val-
idating preferred causal arguments by gathering persuasive
evidence can be limited in two important ways. First, re-
search designs that provide strong evidence against one
alternative often provide weak evidence against some other
plausible alternatives: hence a single study with a given de-
sign is unlikely to provide strong evidence against all plau-
sible alternatives, and researchers must trade off one kind
of validity against another. Second, the number of plausible
alternative explanations can be quite large, and providing
evidence against them all can require more data collection,
more complex research designs, and/or more extensive
knowledge on the part of researchers than is feasible.
The strategy of reducing the number of alternatives ex
ante when developing a preferred causal explanation can
help researchers to cope with the problem of large num-
bers of alternatives, but it has the potential to create addi-
tional limitations to the persuasiveness of a study. First,
there is a tradeoff: narrower speci?cation of a preferred
causal explanation can make it easier for researchers to
persuade others that the evidence they have gathered val-
idates their preferred explanation, but narrower speci?ca-
tion can also make it more dif?cult to persuade others that
the study is interesting. Second, plausible alternatives are
sometimes excluded merely because researchers are
unaware of them, especially when these alternatives are
outside the scope of the researchers’ preferred theoretical
perspectives. Third, a study can fail to be persuasive to
readers with different prior beliefs because it does not
report the limitations to inference that result from exclud-
ing alternatives ex ante.
Excluding alternative causal explanations through the
processes of validating and/or developing a preferred cau-
sal explanation includes some judgments and decisions
that are relatively objective. That is, they are located nearer
the objective end of the objective–subjective continuum
and can lead to agreement among researchers with
initially different views about the accounting issue but
similar views about what constitutes persuasive evidence.
For example, researchers with different views about the
effectiveness of monetary incentives may be able to agree
that a given study showing no effect has been able to elim-
inate the alternative explanation ‘‘low statistical power of
test’’ by using a large sample.
As we argue below, however, there are also subjective
judgments and decisions in these research processes about
which researchers are not likely to agree so easily. Multiple
types of validity exist, and a single study typically cannot
maximize all types simultaneously; the tradeoff decisions
that researchers must make in consequence are subjective
to a considerable degree. Moreover, the success of even a
relatively objective process like providing persuasive
evidence against alternatives depends on choices made
during the development of a preferred causal explanation.
Alternatives that are excluded during development will not
be exposed to the potentially more objective processes of
validation; and researchers may disagree, on relatively
subjective bases, about the appropriateness of excluding
speci?c alternatives in the development process.
Validating preferred causal explanations
Because the research we address aims at validating gen-
eral causal explanations based on sets of speci?c observa-
tions, the validity of inference from the particular
observations to a broader population is crucial in assessing
how well a study validates the causal explanations it
makes and invalidates all plausible alternative causal
explanations. The predictive validity framework in Fig. 1
(adapted from Runkel and McGrath (1972)) provides a
widely accepted organizing framework for validating such
inferences.
5
Inferences about the conceptual (theoretical)
relation between the independent and dependent variables
(link 1) are made by validating link 4 between the
operationalized independent and dependent variables. Valid
inference about the conceptual relation also requires that
links 2 and 3 between the conceptual and operational
3
Although random assignment is often not feasible at the level of
independent organizations in the natural environment, it can be used with
organizational subunits in ?eld experiments or with laboratory ‘‘organiza-
tions’’ consisting of small numbers of individuals.
4
Note that ‘‘explanation’’ does not necessarily mean ‘‘single-factor
explanation,’’ and thus excluding alternative explanations does not mean
rejecting complex multi-factor causality. An explanation can be a model
specifying multiple causal factors and the forms of their relations with each
other and the dependent variable of interest (see Luft & Shields, 2003 for
more detail). Alternative explanations would then be models specifying
different sets of causal factors and/or forms of relation.
5
This framework was introduced to the accounting literature by Libby
(1976) and used by Libby, Bloom?eld, and Nelson (2002) to analyze
experimental research in ?nancial accounting. Note that, in order to present
validity issues clearly, this ?gure represents a simple causal relation
without interactions, endogeneity, etc. The causal diagram can readily be
modi?ed to include more complex relations (see Luft & Shields, 2003;
2007).
552 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
variables are valid and that other variables that can in?u-
ence the dependent variable are controlled for in link 5.
Validating these links requires researchers to provide
evidence that an apparent association between the opera-
tionalized independent and dependent variables is more
consistent with their preferred causal explanation (link 1)
than with other alternatives such as the following:
The apparent association between the operationalized
variables is not caused by a real association but by vio-
lations of the assumptions of the statistical tests
employed.
The real cause of variation in the dependent variable is
not the conceptual independent variable in the
researcher’s preferred causal explanation but a different
variable that the (poorly chosen) operationalized inde-
pendent variable actually captures.
The operationalized variables are associated with each
other, not because x causes y (as in the researcher’s pre-
ferred causal explanation), but because y causes x, or x
and y have a common cause z.
The study makes a general claim that x causes y, but a
more valid alternative explanation is that x causes y
only under certain conditions that were present in the
context of the study but are not universally present.
In terms of the widely used threats-to-validity frame-
work (Shadish, Cook, & Campbell, 2002), these alternative
explanations represent threats to statistical conclusion
validity, construct validity, internal validity, and external
validity, respectively. Because these threats are different,
created by different alternative explanations, the research
processes that reduce one threat will not necessarily
reduce—and may in fact increase—other threats. The exis-
tence of such tradeoffs creates important limitations to
researchers’ ability to validate their preferred causal expla-
nations through relatively objective process of providing
persuasive evidence (limitation A1 in Table 1).
Tradeoffs among validity types
The design of any particular study—for example, its
choice of research method and data sources—involves
tradeoffs among the different types of threats to validity.
For example, the choice of using a laboratory experiment
typically reduces threats to internal validity through
random assignment of sampling units to experimental
conditions but can raise questions about construct or
external validity.
6
Similarly, when researchers use archival
Table 1
Validating and developing researchers’ preferred causal explanations as a process of eliminating alternative causal explanations.
A. Validating preferred causal explanations
Aim:
Provide persuasive evidence that invalidates all plausible alternative causal explanations
Limitations to objective achievement of aim:
(1) Validity tradeoffs. Providing more persuasive evidence against one alternative causal explanation often entails providing less persuasive
evidence against some other plausible alternatives
(2) Numerous plausible alternative explanations. Providing evidence against all plausible alternatives would require more data collection, more
complex research designs, and/or more extensive theoretical knowledge than is feasible
B. Developing preferred causal explanations
Aim:
Reduce the number of alternative causal explanations against which persuasive evidence must be gathered, by narrow speci?cation of context,
causal-chain segment, etc.
Limitations to objective achievement of aim:
(1) Reduction in researchers’ ability to persuade others that their preferred causal explanations are interesting when they are very narrowly
speci?ed
(2) Exclusion of plausible alternatives merely because they are outside of the researchers’ preferred theoretical perspectives
(3) Failure to report limits of inference. Studies that provide evidence consistent with a preferred causal explanation do not necessarily provide
evidence against other explanations if these other explanations have been excluded in the development process
Fig. 1. Predictive validity framework (adapted from Runkel and McGrath (1972)).
6
For example, questions can arise about whether performance reports
provided to laboratory experiment participants capture the same theoret-
ical properties as a set of performance reports in natural environments
(construct validity). Questions can also arise about how large a set of
performance reports in natural environments actually have these proper-
ties (external validity).
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 553
and survey data, their sample choices create validity trade-
off choices. For example, a more diverse sample can increase
external validity by enabling researchers to test for interac-
tions, but can increase threats to statistical conclusion valid-
ity and construct validity.
7
In consequence of these design choices, the validation
process will be able to exclude some alternative causal
explanations persuasively, at the cost of providing only
limited evidence against some other alternatives. Judg-
ments about acceptable tradeoffs among threats to validity
are more subjective than judgments about whether a spe-
ci?c threat (e.g., low statistical power, an interaction) has
been eliminated. There is no common measure of validity
across the types that would allow researchers to provide
evidence that they are giving up only v units of one kind
of validity to gain v + n of another kind of validity. More-
over, as Shadish et al. (2002) argue, the relative importance
of different kinds of validity can vary across research con-
texts, such as basic versus applied research.
8
But there is no
standard set of weights for the different validity types that
would move researchers with different initial judgments to-
ward agreement about whether the gain is greater than the
loss in any particular tradeoff among validity types.
Because there are neither summary measures nor
agreed-on weights for the different types of validity, judg-
ments about the acceptability of particular tradeoffs are
relatively subjective. Perhaps in consequence (since tradi-
tion can substitute for objective validation as a basis for
agreement), these judgments are much in?uenced by the
traditions of sub-communities of researchers. For example,
some sub-communities are more likely than others to ac-
cept p = .06 as adequate support for a hypothesis. Similarly,
some sub-communities are more likely to be more con-
cerned about the threats to external validity in laboratory
experiments or single-site ?eld studies, while others are
likely to be more concerned about threats to internal and
construct validity in large-sample archival and survey
studies.
Numerous plausible alternative explanations
Identifying all plausible alternative explanations is a
challenge because the social sciences offer a very diverse
and extensive range of theories that are potentially rele-
vant to accounting. Economics, psychology, and sociology
theories are commonly employed in accounting research
(Chapman, Hopwood, & Shields, 2007), and other
disciplines such as anthropology, history, political science,
and neuroscience can also play a role. Moreover, the diver-
sity of relevant theories within each discipline is large. (See
Birnberg, Luft, & Shields, 2007 for an example of the diver-
sity of psychology theories relevant to accounting.)
9
Not all social science theories will be direct competitors
in explaining a given accounting phenomenon, but often
the number of possible explanations is nontrivially large.
Ittner, Larcker, and Meyer (2003) provide a good example,
identifying eight plausible explanations (two based on eco-
nomics theories and six based on psychology theories) for
the magnitudes of incentive weights on performance mea-
sures in a large bank’s balanced scorecard. A longer list of
plausible alternatives could be composed for many
accounting research questions. For example, Table 2
provides a (non-exhaustive) list of 14 plausible causal
explanations (E1–E14) of why some decision makers
would rely more heavily on apparently relevant accounting
information in making a particular decision than others do.
The plausible alternatives to a researcher’s preferred
causal explanation can be so numerous and diverse that
the process of validating the preferred explanation by pro-
viding evidence against all identi?ed alternatives can, in
principle, become unmanageable (limitation A2 in Table 1).
More plausible explanations provided by more diverse the-
ories often mean that more different independent variables
are potentially relevant. Using a larger number of variables
is likely to require more data collection, more complex
research designs to take into account the relations among
these variables, and/or more knowledge on the part of
researchers to deal with the resulting theoretical and
methodological issues. Meetings these requirements can
be costly or infeasible, and hence researchers often use
the process of developing their preferred causal explana-
tions to reduce the number of alternative plausible expla-
nations against which they need to provide evidence.
Developing preferred causal explanations
A broad-based literature search will often generate a
large number of plausible causal explanations related to
a particular research question, like the 14 alternatives that
appear in Table 2. How many of these are in fact competing
explanations against which evidence must be gathered de-
pends on how the researchers’ preferred causal explana-
tions are developed and thus how they are speci?ed in
their ?nal form. As explained in more detail below, some
criteria for excluding alternatives through development
are shared by many researchers, even though they differ
about speci?c accounting issues; and thus eliminating
alternative explanations by these means can be regarded
as relatively objective. However, as with validation, subjec-
tive judgments play a key role in developing preferred cau-
sal explanations and can create limitations to researchers’
success in persuading others who initially hold different
views about whether the researchers’ preferred explana-
tion is valid.
7
Larger samples can have more uncontrolled variation that reduces the
power of tests (statistical conclusion validity), for example when data are
collected from a larger number of ?rms and/or industries that vary in ways
not captured adequately by control variables. Larger samples can also
decrease the likelihood that a given operational variable is really capturing
the same construct in the diverse units of observation in the sample (e.g.,
does activity-based costing have the same essential properties in all the
different organizations in the sample?).
8
For example, ‘‘Basic researchers have high interest in construct validity
because of the key role that constructs play in theory construction and
testing. Applied researchers tend to have more interest in external validity
. . .’’ (Shadish et al. 2002, p. 100).
9
Balakrishnan and Penno (2013) illustrate the combinatorial explosion
of alternative causal explanations when many explanatory factors are
available and researchers do not include all possible factors in their
preferred model.
554 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
Plausible alternative explanations can be eliminated
relatively objectively by specifying the preferred causal
explanation more narrowly, so that explanations which
would be plausible alternatives to a broadly stated pre-
ferred explanation become logically inapplicable to the
narrower version of the explanation. For example, the
statement of the preferred causal explanation can explic-
itly limit the context to which it is intended to apply—orga-
nizations or markets of a certain size or structure,
individuals with a certain level of expertise or motivation,
short-term or long-term variation only,
10
etc.—in order to
exclude explanations based on variation in these individual,
market, or organizational characteristics.
Research settings are then chosen to be consistent with
the narrowly speci?ed context. For example, researchers
who are interested in the effects of noise in accounting
information (E1 in Table 2) might want to avoid the need
to measure and model individual variation in risk attitudes
(E2), estimation error (E3), and/or mental models (E4), be-
cause their research designs are already complex, or they
doubt that the available theoretical or empirical models
and/or operationalization of constructs will allow them
to control effectively for individual variation from all these
sources.
To avoid dealing with these variables, researchers can
develop a noise-based explanation that explicitly narrows
its applicability by stating assumptions about alternative
explanations that limit their in?uence. For example, the
explanation can be speci?ed to apply only to settings
where (a) risk aversion is high enough that the uncertainty
created by noise will matter, (b) estimation error is low
enough that individuals can distinguish higher from lower
noise reliably, and/or (c) individuals’ mental models are
similar enough that the implications of different noise lev-
els for individual decisions are reasonably similar across
individuals and predictable by the researcher. The research
setting then needs to be matched to such assumptions: for
example, a setting might be found where individuals rely
on a trusted system for key estimates rather than making
their own estimates with their own idiosyncratic errors.
Another relatively objective way to reduce the number
of plausible alternatives through development of the pre-
ferred explanation is to specify only a narrow segment of
the causal chain, to which only some alternative explana-
tions are logically applicable. For example, accounting
information received by different sets of users can be dif-
ferentially noisy (E1) because it is produced by different
technologies (E14): there is a causal chain leading from
technology to noise to information use. Researchers may
specify that they are only interested in the effects of noise
and not in its causes. Given this speci?cation, technology is
not a factor that competes with noise to explain the use or
non-use of accounting information, and researchers do not
need to collect information on technology variation to val-
idate their preferred causal explanation.
Reducing the number of plausible alternatives through
narrow speci?cation often contributes to the effectiveness
and ef?ciency of research design. It enables researchers to
document the explanatory power of their preferred causal
explanation without using inordinately complex research
designs and statistical analyses to control for or eliminate
alternatives. Narrow speci?cation also can create limita-
tions, however (see Table 1, Part B). Narrow speci?cation
often raises questions about how interesting a limited con-
text or limited segment of a causal chain is (limitation B1).
Sometimes researchers exclude alternative explanations in
Table 2
Selected alternative explanations for why some decision makers rely more on apparently relevant accounting information than others.
Difference between users and non-users of relevant accounting information Relevant theory
E1. Accounting information received by users is less noisy than that received by non-users Economics,
psychology
E2. Users of (noisy) accounting information are less risk-averse than non-users Economics,
psychology
E3. Non-users overestimate and/or users underestimate the noise in accounting information Psychology
E4. Accounting information plays a smaller role in non-users’ than in users’ mental models Psychology
E5. Leadership style of users and/or their superiors is more accounting-oriented Psychology
E6. Users (non-users) work in environments where reliance on accounting is (is not) consistent with social or professional
norms
Sociology
E7. Users have more accounting training than non-users and therefore are more con?dent that they can use the information
effectively
Psychology
E8. Users belong to a national culture that is more compatible with strong reliance on accounting information than non-users Psychology
E9. Non-users have better access to high-quality information that substitutes for accounting than users do Economics
E10. Users have better access to complementary information that improves the usability of the accounting information Economics
E11. Users have more ego-involvement with the accounting because they have been more involved in creating the speci?c
accounting measures than non-users have
Psychology
E12. Users (non-users) work in settings where powerful coalitions that would be advantaged (disadvantaged) by reliance on
accounting have encouraged (limited) its use in decisions
Economics,
sociology
E13. Users, more than non-users, work in settings where the use of accounting is a signal of skills that are important for
promotion to higher-level jobs
Economics
E14. Users work in settings with more advanced technology, which provides more up-to-date and/or more transparently
presented versions of the accounting information
Various theories
Notes: 1. E = Explanation.
2. See Hartmann (2000) for a review of literature that proposes and investigates many of these explanations.
10
See Luft and Shields (2003, 2007) for guidelines on making time-frame
and level of analysis choices.
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 555
the development process not because they are logically
implausible but merely because they are outside the
researchers’ preferred theoretical perspectives (limitation
B2). Finally, researchers sometimes fail to report
adequately the limits to inference that result from their
exclusion of alternative explanations in the development
process (limitation B3).
Choices of which alternatives to exclude and which lim-
itations to report often depend on researchers’ choices of
theoretical perspectives, which are highly subjective deci-
sions. Researchers often make these decisions early in their
careers, based on their preferences, abilities, and experi-
ences, as well as on institutional forces such as the avail-
ability of training and research support. There is no
objective way of determining (for example) whether sociol-
ogy is in general a ‘‘better’’ theoretical perspective as a basis
for accounting research than psychology, or vice versa.
11
Because a choice of theoretical perspective is typically a
long-term choice to build expertise in one particular set of
theories and research methods and not others, it tends to
constrain subsequent, more study-speci?c choices. Thus,
some of the research choices that follow from a choice of
theoretical perspective are also highly subjective and can-
not meaningfully be labeled right or wrong. Just as it is not
‘‘wrong’’ to be an economist and ‘‘right’’ to be an anthro-
pologist, so the choice of primary objects of study (e.g.,
individual versus social processes, markets versus organi-
zations) cannot be objectively right or wrong in general:
such choices are not subject to agreed-on criteria that
would determine which of these objects of study it is ‘‘bet-
ter’’ to know about. In the following subsections we ana-
lyze the three limitations presented in Table 1, Part B,
indicating how each of these is linked to researchers’ sub-
jective choices of theoretical perspectives.
Narrow speci?cation and interestingness
Choice of theoretical perspective is a strong subjective
in?uence on researchers’ judgments about the interesting-
ness of any given research question and the causal expla-
nation that is proposed to answer it. In Davis’ (1971)
widely cited de?nition, interesting studies are those that
share most of their audience’s assumption-ground but
deny some particular assumption.
12
Thus, judgments of
the interestingness of research questions are highly audi-
ence-speci?c: ‘‘Propositions are interesting or uninteresting
only in relation to the assumption-ground of some audi-
ence.’’ (Davis, 1971, p. 32). In such cases there is no objective
method, transcending audience assumption-grounds, of
determining which audience’s preferred questions are
‘‘more interesting.’’
As researchers narrow the applicability of their pre-
ferred causal explanations in order to improve their
chances of success in validation, they risk making their
studies less interesting to some potential audiences, be-
cause narrower explanations may no longer be a good ?t
to the assumption-ground of these audiences. For example,
suppose a preferred causal explanation focuses only on the
individual level of analysis and excludes explanations at
organizational and social levels. (Note that this does not
mean denying the relevance of organizational and social
levels of explanation; it simply means that these levels
are outside the scope of investigation in the particular
study.) An individual-level explanation is likely not to
seem interesting, and thus not valuable and persuasive,
to researchers whose preferred theoretical perspective
includes the assumption that explanations should focus
primarily on higher levels of analysis.
Researchers often strive to interest and persuade rela-
tively broad audiences (not just their few closest col-
leagues), and thus they can be willing to make a tradeoff
in developing their preferred causal explanation. Rather
than maximizing their chances of validation success by
specifying their preferred explanations very narrowly, they
trade off chances of validation success against chances of
interesting a broad(er) audience. But just as there is often
no objective method of determining which audience’s pre-
ferred questions are more interesting, there is also no
objective method of determining what the optimum point
is in such tradeoffs. Hence beliefs can sometimes differ in
objectively unresolvable ways about whether a given study
has made an appropriate tradeoff of this kind.
In the two following sections, we argue that excluding
alternative explanations in the development process some-
times runs the risk of slipping from justi?able narrow
speci?cation into two less justi?able research practices:
excluding plausible alternatives merely because they are
outside of the researcher’s preferred theoretical
perspective, and failing to report the limitations of the
inferences that can be made when alternatives are
excluded through narrow speci?cation of preferred causal
explanations.
Exclusions based on researchers’ preferred theoretical
perspective
Researchers trained in one theoretical perspective
sometimes have limited acquaintance with other perspec-
tives. Insofar as plausible alternative causal explanations
are excluded from consideration in a study simply on the
basis of convenience—that is, alternatives are excluded be-
cause the researcher is neither aware of themnor equipped
to deal with them—it is unlikely to inspire con?dence in
the validity of the study’s inferences. Insofar as exclusions
are based on a belief that causal explanations from other
theoretical perspectives are so generally implausible or
uninteresting that they do not need to be considered
explicitly, it is also likely to limit the persuasive power of
the study.
11
Max Weber identi?es the paradox that scientists (among whom he
includes natural scientists, social scientists, and scholars of the humanities)
all believe that their particular scienti?c activities are especially worth-
while, ‘‘But they cannot prove scienti?cally that this is the case.’’ (Weber,
1958, p. 145).
12
Similarly, Bartunek, Rynes, and Ireland’s (2006) survey of editorial
board members of the Academy of Management Journal found that the most
frequently given reason for why an article is interesting is that it is
counterintuitive, in that it challenges established theories, is contrary to
folk wisdom, or creates an ‘‘aha’’ moment. But see Towry (2012, p. 26) for
important caveats to this view: for example, the ‘‘risk that we will
encourage poorly developed, convoluted theories, as authors embark on a
‘search for surprise.’’’
556 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
Accounting research increasingly does not take this
approach but explicitly considers multiple theoretical
perspectives in individual studies. For example, Libby
et al. (2002) describe how research studies that integrate
theories from economics and psychology have contributed
to a better understanding of ?nancial accounting by spec-
ifying more clearly the mechanisms affecting individual
and market behavior. In management accounting research,
Covaleski, Evans, Luft, and Shields (2003, 2007) identify
opportunities and provide guidelines for budgeting
research that integrates theories from economics, psychol-
ogy and sociology.
Failure to report limits of inference
Studies that do not adequately report their limitations
are not likely to be persuasive, because they claim broader
applicability or more certainty than can be justi?ed by the
evidence they provide. The limitations that need to be re-
ported depend in part on the choices researchers make in
developing their preferred causal explanations. For exam-
ple, when researchers specify a preferred causal explana-
tion narrowly so that it applies only to a particular
context (e.g., time frame, organization type), they need to
make clear to readers that their study makes no inferences
about other contexts. Failure to report such limitations can
easily arise when plausible alternatives are outside of
researchers’ (subjectively) chosen theoretical perspectives.
If these alternatives are excluded in development simply
because researchers are unaware of them or unprepared
to deal with them, then researchers will also be unaware
of the resulting limitations.
Even when researchers are aware of plausible
alternatives and have excluded them via narrow speci?-
cation, they sometimes fail to make it clear that their
studies provide no evidence about these alternative
explanations. For example, evidence of the importance
of individual-level variables is not by itself evidence
against the importance of culture or vice versa. Individ-
ual-level studies often hold culture constant and thus
provide no evidence about effects of cultural variation.
Similarly, evidence of strong effects of monetary incen-
tives in settings where social norms have little vari-
ance—or evidence of strong social-norm effects in
settings with little monetary-incentive variance—does
not enable researchers to make inferences about the rela-
tive importance of money versus social norms. Whether a
study is informative about causal explanations other than
the researcher’s preferred explanation depends on
whether other explanations were even considered and
whether the (often limited) research setting chosen was
one in which these other explanations could plausibly
have had signi?cant effects.
Conclusion
Philosophers have commented on the subjectivity of
empirical inquiry even in the natural sciences:
There are, of course, general maxims for empirical
enquiry—for example, the very fact that we speak of
empirical enquiry re?ects one of them: ‘Don’t try to
?gure out in a purely a priori way how nature works.’. . .
But which theories we should test . . .; when a theory has
been suf?ciently tested to warrant provisional accep-
tance, and when it has been tested enough to be relied
on, at least until a better theory comes along; these are
all matters which in practice scientists decide partly on
the basis of tradition . . . and partly on the basis of
intuition. (Magee, 1978, p. 232).
Subjective intuition can be a source of creativity, and tradi-
tion can support potentially useful regularities of scholarly
practice when no logically or evidentially compelling
reasons for regularity are available.
13
But reliance on rela-
tively objective—explicit, shared, logically and evidentially
based—criteria remains an ideal in positivist accounting
research.
A widely accepted strategy for implementing the posi-
tivist ideal relies on using the predictive validity frame-
work (Fig. 1) and explicitly and systematically addressing
threats to the validity of empirical inferences. Addressing
these threats means validating a researcher’s preferred ca-
sual explanation by providing evidence against plausible
alternatives and/or developing the preferred causal expla-
nation in such as way as to limit the number of alternatives
about which evidence must be provided. Insofar as these
approaches to developing and validating causal explana-
tions make acceptance of a study’s inferences more persua-
sive to a broad community of researchers and thus less
researcher-dependent, they increase the objectivity of
research.
The positivist ideal of objectivity also includes explicit
awareness and reporting of the subjective judgments and
decisions involved in developing causal explanations and
making research-design choices. Because these judgments
and decisions can exclude alternative plausible explana-
tions from being considered and can reduce the evidence
gathered against these alternatives, they constitute impor-
tant limitations on validation of a study’s inferences.
Reporting these limitations—that is, reporting the (often
unavoidably) subjective nature of developing and validat-
ing causal explanations—can, perhaps paradoxically, in-
crease the objectivity of a research study. Researchers
who espouse different, and potentially con?icting, speci?c
theories are more likely to agree about the validity of the
inferences that can be drawn from a particular study—that
is, the inferences will be less researcher-dependent and
thus less subjective—if the limitations of these inferences
are fully reported.
Acknowledgments
We would like to thank Chris Chapman, Will Demere,
and Tyler Thomas for their comments on an earlier version
of our article.
13
For example, it is not evident that p < .05 is the optimal cutoff for
acceptable inferential results, but it is ef?cient for the scholarly community
to accept such a common cutoff rather than argue about it in the review
process of every hypothesis-testing paper submitted to journals.
J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558 557
References
Andreski, B. (Ed.). (1974). The essential Comte: Selected from the cours de
philosophie positive. New York: Barnes & Noble.
Balakrishnan, R., & Penno, M. (2013). Causality in the context of analytical
models and numerical experiments. Accounting, Organizations, and
Society.
Bartunek, J., Rynes, S., & Ireland, R. D. (2006). What makes management
research interesting, and why does it matter? Academy of
Management Journal, 49, 9–15.
Birnberg, J., Luft, J., & Shields, M. (2007). Psychology theory in
management accounting research. In C. Chapman, A. Hopwood, &
M. Shields (Eds.). Handbook of management accounting research (Vol.
1). Oxford: Elsevier.
Chapman, C., Hopwood, A., & Shields, M. (Eds.). (2007). Handbook of
management accounting research (Vol. 1). Oxford: Elsevier.
Covaleski, M., Evans, J., Luft, J., & Shields, M. (2003). Budgeting research:
Three theoretical perspectives and criteria for selective integration.
Journal of Management Accounting Research, 15, 3–49.
Covaleski, M., Evans, J., Luft, J., & Shields, M. (2007). Budgeting research:
Three theoretical perspectives and criteria for selective integration. In
C. Chapman, A. Hopwood, & M. Shields (Eds.). Handbook of
management accounting research (Vol. 2). Oxford: Elsevier.
Davis, M. (1971). That’s interesting! Towards a phenomenology of
sociology and a sociology of phenomenology. Philosophy of the Social
Sciences, 1, 309–344.
Engel, C. (2013). Behavioral law and economics: Empirical methods. MPI
Collective Goods Preprint, No. 2013/1. .
Friedman, M. (1953). Essays in positive economics. Chicago: University of
Chicago Press.
Hartmann, F. (2000). The appropriateness of RAPM: Toward the further
development of theory. Accounting, Organizations and Society, 25,
451–482.
Ittner, C. D., Larcker, D. F., & Meyer, M. W. (2003). Subjectivity and the
weighting of performance measures: Evidence from a balanced
scorecard. The Accounting Review, 78, 725–756.
Libby, R. (1976). Discussion of cognitive changes induced by accounting
changes: Experimental evidence on the functional ?xation
hypothesis. Journal of Accounting Research Supplement, 14, 18–24.
Libby, R., Bloom?eld, R., & Nelson, M. (2002). Experimental research in
?nancial accounting. Accounting, Organizations and Society, 27,
775–810.
Luft, J., & Shields, M. (2003). Mapping management accounting: Graphics
and guidelines for theory-consistent empirical research. Accounting,
Organizations and Society, 28, 169–249.
Luft, J., & Shields, M. (2007). Mapping management accounting: Graphics
and guidelines for theory-consistent empirical research. In C.
Chapman, A. Hopwood, & M. Shields (Eds.). Handbook of
management accounting research (Vol. 1). Oxford: Elsevier.
Magee, B. (Ed.). (1978). Men of ideas. NY: Viking Press.
Mayntz, R. (2004). Mechanisms in the analysis of social macro-
phenomena. Philosophy of the Social Sciences, 34, 237–259.
Mellor, D. (1995). The facts of causation. London: Routledge.
Mellor, D. H. (2004). For facts as causes and effects. In J. Collins, E. Hall, &
L. Paul (Eds.). Causation and Counterfactuals. Cambridge, Mass: MIT
Press.
Menzies, P. (2009). Counterfactual theories of causation. In E. Zalta (Ed.).
The Stanford encyclopedia of philosophy. Fall 2009 Edition. .
Runkel, P., & McGrath, J. (1972). Research on human behavior: A systematic
guide to method. New York: Holt, Rinehart and Winston.
Searle, J. (1995). The social construction of reality. New York: The Free
Press.
Shadish, W., Cook, T., & Campbell, D. (2002). Experimental and quasi-
experimental designs for generalized causal inference. Belmont, CA:
Wadsworth.
Shapiro, B. P. (1997). Objectivity, relativism, and truth in external
?nancial reporting: What’s really at stake in the dispute. Accounting,
Organizations and Society, 22, 165–185.
Towry, K. (2012). Discussion of ‘‘Subordinates as the ?rst line of defense
against biased ?nancial reporting’’. Journal of Management Accounting
Research, 24, 25–30.
Weber, M. (1958). Science as a vocation. In H. Green, & C. Wright Mills
(Trans. & Eds.), From Max Weber: Essays in sociology. New York: Oxford
University Press.
558 J. Luft, M.D. Shields / Accounting, Organizations and Society 39 (2014) 550–558
doc_565473330.pdf