Académique Documents
Professionnel Documents
Culture Documents
a.
b.
The conditions of the test should be consistent with the underlying assumptions or axioms of the theory.
The test should define clearly and with good reason the test results that support the theory and the test
results that would contradict the theory.
c. The test should be unbiased in the sense of providing a reasonable probability that it could uncover
evidence that could confirm or contradict the theory.
d. The test should define and measure accurately the artifacts for the theorys variables.
3. Twenty Years of Management Accounting Field Research: Accomplishments and Shortcomings
Papers that were published in the 1980s are widely associated with contemporary field research in management
accounting : Hopwood (1983: 302) argued that accounting had divorced itself from the organization and called for
more research on the design and use of accounting systems within the context of an organization Kaplan (1983,
1984) argued that innovative cost accounting and management control practices were invisible to traditional
academic research, which tended to study central tendencies. in large populations Ferreira & Merchant (1992)
evaluate the quality of field research in management accounting by comparing the 82 studies to a set of fieldwork
standards (Bruns & Kaplan, 1987) and categorize the papers according to their subject matter, research design,
data presentation and interpretation, and practical implications They conclude that published field studies in
management accounting have explored interesting questions, usually in conjunction with emerging and
innovative practices, and that most of the studies have progressed from description to testing or developing
theory. Merchant & Van der Stede (2005) conduct a survey of revisions to leading accounting textbooks during
the same period. They trace the five major areas of innovation in management accounting practice to research
that had its genesis in the field and conclude that field research has been exceptionally valuable at producing
usable knowledge as compared to other research methods during the last 20 years. In spite of these
contributions, field research is not without weaknesses. Ferreira & Merchant (1992) attribute problems with field
research to poor research design, poor data presentation, and flawed data interpretation. Ferreira and Merchant
also criticize field studies as being inadequately connected to prior research literatures. For example, in failing to
select sites that differ in controlled ways from previous studies, each field study can easily become a one off
study that does not contribute to a stream of research and cumulative knowledge. Using a smaller set of primarily
North American journals, Shields (1997) finds that the field research method has a low frequency of use; 10 out
of 152 articles published during the 19901996 period used field study methods. He posits that the lack of
published field research stems from inadequate skills in conducting field research, a dearth of colleagues from
whom to obtain feedback and with whom one might collaborate, long research cycle times that are in conflict with
short-tenure clocks, limited access to field sites, and editorial biases against field research. Bennis & OToole
(2005:99) trace broader failings in management education to failings of the research community. They criticize
the form of scientific method commonly employed by management scholars for involving minimal time in the field
discovering the actual problems facing managers and conclude that,....because they are at arms length from
actual practice, they often fail to reflect the way business works in real life. Unfortunately, Bennis and OToole
are less optimistic than Young or Shields that extant academic institutions have the means or motivation to
support scholarship that is more connected to the practice of management.
4. Key Decisions, Choices, and Contributions of Quantitative Field Research
Table 1 categorizes the field research papers the author use to illustrate their arguments according to the nature
of the data analysis and the papers research purpose.
was critical to the quality of the data and proper specification of the tests (Atkinson & Shaffir, 1998).
One condition for successful theory testing is the fidelity with which the measured variables represent
the variables of the theory. In how many studies does the theory and literature review seem utterly
disconnected from the empirical analysis because the variable measures do not correspond to the
theory? In the end, even if the researcher finds significant results, the reader does not believe that
the theory has been tested.
Author have found it equally challenging in quantitative field Studies as when labels that companies
assign to measured data do not correspond to academic definitions or when survey questions
inadvertently use language that has special meaning inside a company that differs from general use.
Author have found that it typically takes several conversations as well as extensive exploratory data
analysis for the researcher to be able to relate a companys measure of capacity to economic notions
of, for example, theoretical capacity, normal capacity, actual capacity, and peak load capacity. Young
& Selto (1993) is an example of researchers conducting an open-minded search for data on
performance. Their objective was to identify and measure performance improvements associated
with implementation of just-in-time manufacturing methods. During the first 6-months in the field, they
identified six different archival measures of performance appropriate for their site: cycle time
efficiency, yield rate, defective rate, production schedule adherence, product cost efficiency, and
number of manufacturing process problems. In sum, theory is emergent in field studies aimed at
grounded theory development and is a strong preconditioning factor for field research aimed at theory
testing. However, even theory testing requires dynamic adaptation of the researcher. Once the theory
is identified, quantitative field studies aimed at theory testing must grapple with pragmatic issues of
research design, data collection, and analysis in an effort to ensure that the tests are aligned with the
theory.
4.2. Site Selection
Selecting a site suitable for theory testing occurs simultaneously with the practicalities of ensuring that
data exist to allow the proposed hypotheses to be tested. A sequence of (1) identifying
(opportunistically) a site where researchers can obtain access, (2) searching for data, and (3)
selecting a theory to test is generally not recommended. Practicalities are such that research only
rarely progresses linearly from theory to identifying the properties of an ideal site, to negotiating
access to that site, and to discovering data that meet all of the researchers needs.
In some field research, it may be possible to unearth latent data through interviews, direct
observations, and surveys. However, if the researcher is interested in using archival data that are
measured and captured by the firm, then site selection requires a simultaneous evaluation of the
availability of measured data. Chenhall (1997) and Davila (2005) are excellent examples of selecting
a group of firms that are in equilibrium with respect to a certain feature of interest. For Chenhall
(1997), it was important to obtain an adequate sample of firms that practiced total quality
management techniques and used manufacturing performance measures to evaluate managerial
performance. This example illustrates an approach to identifying multiple field research sites that
resemble an ideal profile that is derived from the theory to be tested. Another example of a field
study that selects a large number of firms that are distinct with respect to certain characteristics is
Davila (2005). To generate a large enough sample of firms (95) to enable empirical testing Davila
capitalized on the disproportionate representation of small, young, technology-oriented firms in
Californias Silicon Valley. In Anderson & Lanen (2002), the authors became involved with the firm
when the companys top manufacturing manager sought to understand the impact of product mix
complexity on manufacturing performance. The company was committed to an involved relationship
with educational and research programs of the university, and had identified product complexity as a
topic of interest. However, in spite of the aligned interests of managers and researchers, extensive
plant visits and interviews at six production facilities revealed that the data generated in the factories
were too aggregate and imprecise to support empirical analysis that would extend the literature on
complexity costs. The first example is of a linear search for a firm that permits testing of a rather
specific theory. The success of this approach owes much to the extensive network of practicing
managers to whom senior faculty mentors had access as well as to the willingness of these
colleagues to help the researcher find and gain access to an exemplar organization that permitted
powerful tests of the theory. Thus, we concur with Youngs (1999) counsel that young aspiring field
researchers ally themselves with senior faculty and administrators who know and are involved with
practicing managers. During the research design stage, the researcher must determine the proper
unit of analysis for the theory being tested, and then assess the probability that they will be able to
detect a small effect. If the theory is at the level of the firm, the researcher should consider whether
they can create a large enough sample to rule out a possible explanation of no findings due to low
statistical power. In sum, the researcher pursuing a quantitative field research study must consider
various factors when selecting an appropriate site including the availability of data, the
appropriateness of the company for the study, the appropriate unit of analysis, and whether adequate
statistical power is likely to be obtained for testing the theory. For quantitative field studies that have
theory testing as a goal, careful site selection is critical to meeting the first and third conditions that
Atkinson & Shaffir (1998) require of successful theory testing in the field: that the conditions of the test
(i.e., the setting, the unit of analysis) are consistent with the axioms of the theory, and that the test
provides reasonable probability (i.e., power) to confirm or contradict the theory.
4.3. Data Identification, Collection, and Preparation
4.3.1. Overview of Data
The natural state of the data that are needed for the research question will determine what data
collection tools are necessary as well as the most likely set of analysis methods to employ. However,
this overstates the case, because at times it is possible to identify good proxies for latent data in
measured data, and similarly, there are opportunities to create latent data from measured data.
Researchers also often substitute latent data that emerge from measurement interventions for
measured data and for latent data that are created from measured data, as for example, when survey
respondents are asked to report otherwise unverifiable performance or are asked to opine on matters
that are measured or measurable using pre-existing data. This approach to data collection has proven
the most difficult in publishing field-based research. As might be expected, it opens the researcher to
questions about the likely effects of measurement intervention and is often associated with concerns
about endogeneity and circularity that are not easily addressed.The most important question is not
how to obtain a quantitative measure for each construct that is demanded by the research question,
but how to obtain many measures for each construct (which may include multiple respondents to an
interview or survey) and how to obtain measures that have different natural states (which requires
supplementing interviews and surveys with measured data or latent data derived from measured
data). Together these approaches increase the likelihood of quantitative field research meeting
Atkinson and Shaffirs fourth condition for theory testing: that variable measures represent with high
fidelity the constructs that are critical to the test of the theory.
4.3.2. Collecting Measured Data in the Field Setting
Although the researcher is not involved in the construction of measured data, the researcher still has
an important role to play in assembling an appropriate database and in preparing the data for use.
Even when the data are archival accounting records, the researcher must typically make important
selections from a large database of potential measures. Although the temptation is to take
everything, this only delays the careful culling of important measures. Atkinson & Shaffir (1998: 44)
quoting Gusfield (1995) say: Data that do not stay close to the events, actions, or texts being studied
are always suspectyWhat Geertz called thick description is the ideal, not always achieved but always
to be aimed at. Furthermore, if we are trying to understand management accounting in its natural
context, then we need to understand the key actors interpretation and perception of the situation. A
somewhat unique opportunity when measured data are employed is the possibility of longitudinal
analysis.25 Ittner et al. (1997), in their study of activity- based costing, provide an illustration of a
longitudinal study that uses measured data. Using timeseries data on monthly cost, revenue, and
operations, the authors generated insights about the relation between the cost hierarchy
classifications and costs. These relations are often hard to detect in cross-sectional studies where
variances are averaged out of the sample. Longitudinal analysis brings a different challenge namely,
ensuring that the data have been collected and used consistently over time. It is important for the
researcher to determine whether what is measured has changed over time (e.g., as when accounting
classifications change or boundaries within the organization change). the survey questions were
consistent across time, sometimes the scale changed (e.g., a change from a 5-point to a 7-point
scale, or a change of endpoint anchors for the scale). In addition to the content and structure of the
data, it is also important to consider how the use and visibility of the data has changed over time.
Relatedly, perhaps the measure has had consistent visibility; however, during a portion of the time
period it was linked directly to compensation. Archival research that is not associated with sustained
direct contact with the data generating process is unlikely to uncover these issues that can seriously
compromise empirical tests.
4.3.3. Collecting Survey Data in the Field Setting
If a survey is used to collect data in the field, researchers must consider the timing of the survey in
relation to the timing of other fieldwork. For example, if the researcher combines interviews with a
survey, which should come first? Should the survey precede the more personal interaction with
informants, or follow it? Studies that are predominantly survey-based, may prefer the survey to be
informed by the interviews. The interviews set the stage and define terminology so that survey
respondents are ready to efficiently provide responses (Ferreira & Merchant, 1992). Often there is
an iterative process with data collection alternating between interviews and various stages of
development, refinement, and use of the survey instrument. Using knowledge gained during their
preliminary visits, the researchers adapted the survey to fit a modern technology firm. This was
followed by a pre-test on several groups of plant workers and managers to ensure that the language
was meaningful for their particular research site The survey was revised and administered, at which
time the researchers realized that many of the survey respondents were not fluent in English. In
hindsight, Young and Selto acknowledge that they would have used simpler language if this
had been known. One might use the survey to collect the more easily measured and defined items,
leaving the interview to cover more complex concepts. Another consideration in sequencing different
modes of data collection from a single informant is that the sequence may affect responses and
introduce unwanted bias. By doing a survey first, the interview can be done more efficiently since the
respondent will already understand the researchers interests. The interview can take the place of
open-ended questions on the survey. If one assumes that surveys suffer less from respondent
reactivity, the sequence of survey first, interview second may be less prone to the researcher
affecting all of the data in undesirable ways. However, if the researcher includes complex concepts in
the survey, it becomes more compelling to use the interview first as a means of building shared
understanding of key concepts. Then the survey can be used to elicit respondent opinions related to
the concepts with some assurance that measurement error has been mitigated. In this manner the
strength of having researcher involvement to explain complex issues is harnessed, but at the cost that
undesirable researcher effects could influence both the survey and the interview. Surveys can also be
used to supplement archival data; thus, researchers need to consider opportunities to establish
construct validity with survey data and measured data. Since the survey data typically are drawn from
multiple observers while the archival data may present a single measure that applies to the entire
organization, researchers must decide how many survey respondents are needed and who will be
adequately informed about the phenomena in question. Finally, when data are collected using surveys
in the field the researcher must consider the distribution mechanism. Survey administration is always
fraught with concerns about response bias and response rate, and there are methods for mitigating
these concerns for mail surveys (e.g., Dillman, 1978). Alternatively, the researcher may decide to
administer the survey in person. This has several desirable properties since researchers can mitigate
one common criticism of surveysthat an unknown and/or unwanted person completed the survey. In
addition, if respondents are allowed time during their work hours to complete the survey they may
infer that the survey is more important than they might otherwise.