akuntansi perilaku

561

THE ACCOUNTING REVIEWVol. 76, No. 4October 2001pp. 561–587

Why Does Fixation Persist?Experimental Evidence on the

Judgment Performance Effects ofExpensing Intangibles

Joan L. LuftMichael D. Shields

Michigan State University

ABSTRACT: This study shows experimentally that when individuals use in-formation on intangibles expenditures to predict future profits, expensing (vs.capitalizing) the expenditures significantly reduces the accuracy, consistency,consensus, and self-insight of individuals’ subjective profit predictions. Theexperimental design allows us to eliminate several competing explanations forthis apparent fixation on accounting. Subjects do not base their judgments ona naıve prior belief that expensing precludes effects on future profits; a pre-experiment question shows that subjects expect intangibles expenditures willaffect future profits even when expensed. Moreover, subjects do not lack, orfail to use, data that would allow them to learn the exact expenditure-profitrelation. They receive data on intangibles expenditures and profits as a basisfor learning, and in some respects the learning is quite successful even whenintangibles are expensed; subjects’ profit predictions accurately reflect themean and standard deviation of actual profits. Nevertheless, consistent withpsychological theories of learning, subjects do not learn the exact magnitudeof the effect of intangibles on future profits as well when the intangibles areexpensed. Although the mean of their predictions is accurate, they do notdiscriminate well between cases with high and low actual profits. In conse-quence, their prediction accuracy, consistency, consensus, and self-insight arelower when intangibles are expensed. Thus, in this case, learning does notmitigate fixation on accounting, because accounting affects the learning pro-cess itself.

We appreciate the research assistance of Annie Farrell on this project and the comments of Mark Nelson(associate editor), two anonymous reviewers, Bob Libby, and the seminar participants at the Universities of Illinois,Michigan, Saskatchewan, South Carolina, Southern California, Texas at Austin, and Waterloo. An earlier versionof this paper was presented at the Second Intangibles Conference at NYU, May 1999 and the AAA Annual Meeting,August 1999.

Submitted November 1999Accepted May 2001

Bima C. Pratama

Highlight

562 The Accounting Review, October 2001

Keywords: accounting choice; intangibles; learning; profit prediction.

Data Availability: Please contact the first author.

I. INTRODUCTION

Both managerial and financial accounting research examine whether individuals fixateon accounting or can ‘‘see through’’ or ‘‘unscramble’’ the effects of alternativeaccounting methods (see reviews by Wilner and Birnberg 1986; Libby et al. forth-

coming; Kothari forthcoming). Our study addresses a key question that remains unresolvedin prior studies: Will opportunities to learn eliminate fixation?

A number of studies, both archival and experimental, suggest that fixation results fromlack of experience or relevant data, and therefore opportunities to acquire this experienceor data—i.e., to learn—should eliminate it (Chen and Schoderbek 2000; Gupta and King1997; Waller et al. 1999). Experiments in which student subjects have opportunities to learnsometimes reduce or eliminate fixation on accounting (Gupta and King 1997; Waller et al.1999). Professional financial analysts, however, continue to display fixation when predictingstock prices based on accounting information, although they presumably have opportunitiesto learn about the relation between accounting data and stock prices on the job, beforeparticipating in the experiment (Hopkins 1996; Hirst and Hopkins 1998; Hopkins et al.2000). The absence of learning opportunities for financial analysts within the experimentsremains problematic, however, especially because some of the accounting methods on whichthe experiments focus are relatively rare or novel (Lipe 1998).

In this study we test whether capitalization vs. expensing of intangibles expendituresresults in fixation even when individuals have opportunities to learn. Accounting for intan-gibles is a particularly valuable context for testing the persistence of fixation. First, it hasbroad practical implications for both financial and managerial accounting. Aboody and Lev(1998) and Chan et al. (1999) suggest that requirements to expense intangibles for externalreporting result in mispricing of some firms’ stock. Even though GAAP requires firms toexpense most expenditures for intangibles, a number of firms capitalize these expendituresfor internal reporting out of concern that expensing can mislead managers (Stewart 1991;Tully 1993, 1998).

Second, individuals with business training and experience might reasonably be expectedto see through the expensing of intangibles. Expensing vs. capitalizing intangibles is morewidely publicized and conceptually simpler than most of the other accounting issues forwhich fixation has been demonstrated: debt-equity swaps (Hand 1990), deferred tax assetadjustments (Chen and Schoderbek 2000), accounting for mandatorily redeemable preferredstock (Hopkins 1996), and reclassification adjustments in comprehensive income for un-realized gains or losses in marketable securities (Hirst and Hopkins 1998; Maines andMcDaniel 2000). By choosing a simple and familiar accounting issue, we provide a settingin which learning not to fixate should be relatively easy.

Subjects in our experiment receive data on intangibles expenditures (spending on aquality-improvement program) and gross profits at 20 similar manufacturing plants. In thesedata, the effect of the intangibles expenditures on profits in the current and two succeedingperiods is too small to be statistically significant, but the effect on profits three periods inthe future is large and statistically significant. The experiment tests whether subjects learnthe lagged effect of quality-improvement expenditures on profits from these data, andwhether they learn it equally well when the firm capitalizes expenditures (classifies themas investments in assets) or expenses them. The statistical predictive ability of the expen-diture data is identical whether the expenditures are expensed or capitalized, and subjects’incentives to learn the relation are identical in both conditions.

Bima C. Pratama

Highlight

Bima C. Pratama

Highlight

Luft and Shields—Effects of Expensing Intangibles 563

When the firm capitalizes expenditures (investment condition), subjects learn the rela-tion between intangible expenditures and profits relatively well. With expensed expenditures(expense condition), subjects do not learn the relation as well; on average, they underesti-mate the strength of the lagged relation by about half and make significantly greater pre-diction errors than do subjects in the investment condition. When they use what they havelearned about the effect of intangibles on profits to predict new cases, subjects in theexpense condition predict less consistently across cases and exhibit less consensus acrossindividuals than do subjects in the investment condition. Moreover, subjects in the expensecondition display less insight into their own judgment processes when asked to explainhow they made their predictions.

These results do not occur because subjects naively believe that expensing means theexpenditures have no future benefits. We verify in a pre-experiment question that all subjectsexpect intangibles expenditures to affect profits in future periods, even when the firm im-mediately expenses the intangibles. Moreover, nearly all subjects in the expense conditiondetect the three-period lagged effect of expenditures on profits in the data they receive.Compared to subjects in the investment condition, however, they make greater errors inestimating the magnitude of the lagged expenditure-profit effect, and they incorporate itmuch less effectively into their judgments, as their lower judgment consistency, consensus,and self-insight demonstrate.

These results show that learning is not necessarily a quick remedy for fixation onaccounting, because accounting can influence the learning process itself. This finding isconsistent with psychological theories of the learning process presented in Section II. Thisstudy adds to the repertory of explanations of how fixation occurs—and even persists de-spite apparent user sophistication about accounting (i.e., subjects’ awareness of the potentialfuture benefits of intangibles) and opportunities to eliminate fixation through learning.

The remainder of the paper proceeds as follows: Sections II through IV present thehypothesis motivation, design, and results of an experiment that tests the effects on judg-ment of capitalizing vs. expensing expenditures on intangibles. Section V discusses thestudy and its implications for future research.

II. HYPOTHESIS MOTIVATIONA wide range of individuals in an organization, from top management to lower-level

employees with small-scale spending authorization, make decisions about expenditures onintangibles. Individuals’ judgments about the effects of intangibles expenditures on profitsare key inputs into these decisions. This section describes what individuals must learn toperform well in this judgment task, how they learn it, and how we expect accountingclassification to affect the learning process.

What Must Individuals Learn?Expenditures on intangibles affect current and future profits through two pathways,

which we designate the accounting-calculation effect and the indirect economic-causal ef-fect. As an example of the accounting-calculation effect, if the firm expenses a $10,000expenditure on employee training, then the expenditure reduces current profits by $10,000and has no accounting-calculation effect on future profits. If the firm capitalizes the expen-diture and amortizes it over four periods, it reduces profits by $2,500 each period. Knowl-edge of basic accounting rules enables individuals to identify the magnitude and timing ofthese accounting-calculation effects.

The effect of training-program expenditures on profits includes not only these simplecalculations, but also changes in revenue and/or operating costs resulting from changes inemployee behavior after the training. When the expenditure yields a positive return, the


indirect economic-causal effect1 exceeds the accounting-calculation effect, and errors inestimating the economic-causal effect can affect profit predictions significantly. Individualsmay develop prior beliefs about the magnitude and timing of indirect economic-causaleffects based on their experience with other training programs or information about others’experience. However, individuals can estimate the actual profit effects of particular expen-ditures only by observing the relevant expenditures and profits, and then inferring therelation between them from these observations. Neither prior beliefs nor knowledge ofaccounting rules necessarily yield the same results as estimation of these indirect economic-causal relations from the data.

Field research has documented that managers use accounting reports to modify andrefine their beliefs about the relations between their actions (e.g., expenditure choices) andprofits. In a large-scale field study of nonfinancial managers’ use of accounting data,McKinnon and Bruns (1992, 206) found that managers use periodic accounting reports totest and modify their mental models of key causal relations underlying firm performance.

As managers review their success as reported in accounting reports, they are continu-ously at work, testing and perfecting their mental model of the relationship betweenactivities and success as measured by the management accounting system....In this way,part of the accounting model is incorporated by managers into their own models. Man-agers learn to associate actions with organizational performance and success.

This observation is consistent with the claim that managers can use reports from perform-ance measurement systems to test and modify the beliefs about cause-and-effect relationsembedded in a firm’s strategy and action plan—for example, the relation between employee-training expenditures and profits in a human-capital-intensive firm (Kaplan and Norton1996a, 65).

Regressions of profits on expenditures capture both accounting-calculation and indirecteconomic-causal effects of these expenditures simultaneously. Although managers some-times use such formal statistical analysis, subjective estimation is common in practice(McKinnon and Bruns 1992; Kaplan and Norton 1996b).2 Such subjective estimation is adifficult inferential task, however, especially when individuals must assess lagged effectssuch as the effect on future profits of spending on quality, customer satisfaction, researchand development, or employee training. Prior experimental research provides evidence that,as time lags between reported cause and effect increase, individuals are less able to detectcausal relations and use them in judgments and decisions (Sterman 1989a, 1989b; Diehland Sterman 1995). Little is known, however, about how accounting affects individuals’ability to detect lagged relations and use them appropriately in judgment.

Accounting and the Learning ProcessThis subsection uses basic psychology research about learning under uncertainty to

predict how accounting classifications affect the learning process. We expect this effect tooccur under the following two conditions only: (1) detecting expenditure-profit relations inthe data and using them appropriately in judgment requires considerable cognitive proc-essing effort: the relations are not so strong or salient that they are immediately transparent;

1 We call the economic-causal effects of expenditures ‘‘indirect’’ because they occur only through some interveningprocess. An employee training expense decreases current profit through the accounting-calculation effect, re-gardless of whether the training has any effect on employee behavior; but the expense affects future profit onlythrough the intervening process of behavior changes.

2 Individuals may use subjective approaches because they believe statistical analysis is too costly, the data quantityor quality is not sufficient to support reliable formal statistical analysis, or they believe they can outperformstatistical models.


(2) the amount of potentially relevant data available is large enough, relative to individuals’cognitive information-processing resources, that not all data receive the maximum possibleprocessing. In other words, the setting is such that attention and subjective processing effortare scarce resources (Birnberg and Shields 1984; Simon 1990).

These conditions have two possible alternative consequences. One is that individualswill spread limited attention roughly equally across all potentially relevant relations in thedata, resulting in learning that is imperfect, but no more imperfect for one relation than foranother. The second possibility is that individuals will allocate attention unequally acrosspotentially relevant relations in the data; thus, they may not learn equally well relationsthat are equally strong in the data. Research in psychology, summarized below, providesevidence consistent with the second view.

Suppose individuals must estimate the relation between Y (e.g., current profits) andmultiple Xi’s (expenditures on a particular type of intangible in current and prior periods).Any one, or all, of the Xi’s may affect Y, but the existence and magnitude of the relationare uncertain ex ante. In subjective estimation (unlike multiple regression), individuals tendto examine Xi � Y relations one at a time rather than simultaneously, probably because oflimited working memory (Brehmer 1979; Klayman 1988). Individuals use a variety ofsubjective strategies to estimate Xi � Y relations, some more effort-intensive and morenearly optimal than others (Hutchinson and Alba 1997). In particular, individuals tend toallocate more attention and use more effort-intensive processing for the Xi � Y relationsthey examine earlier; they tend to give less scrutiny to Xi � Y relations that they examinelater, resulting in significantly larger estimation errors (Brehmer 1974, 1979; Klayman1988). The order in which individuals examine Xi � Y relations is therefore key to judg-ment accuracy.

The order in which individuals examine Xi � Y relations often depends on domainknowledge (prior beliefs) activated by data labels or referents (Muchinsky and Dudycha1975; Sniezek 1986; Broniarczyk and Alba 1994). For example, suppose Y is total costs,X1 is customer satisfaction, and X2 is productivity. If individuals believe that productivityhas a strong effect on total costs and that customer satisfaction has a weak effect, then theyare likely to examine the productivity-total cost relation first and more carefully, and there-fore to estimate it more accurately than the customer satisfaction-total cost relation. Whenindividuals are unsure of the Xi � Y relation in the data, perhaps because they have notexamined the data intensively and lack strong prior beliefs about the relation, they tend tosystematically underweight Xi, rather than systematically overweight it or weight it in arandom but unbiased way (Sniezek 1986; Broniarczyk and Alba 1994). These processesare partly conscious and deliberately controlled, but also partly unconscious (Brehmer 1979;Broniarczyk and Alba 1994).

We expect individuals to attend earlier and more intensively to the lagged expenditure-profit relation when the firm capitalizes rather than expenses the expenditure. First, indi-viduals may interpret a firm’s decision to expense vs. capitalize an intangibles expenditureas a signal about the expected timing of the benefits from the expenditure, and may thereforedeliberately direct more attention to lagged relations.3 Second, the expense classificationmay automatically direct attention to the current-period relation, and individuals may there-fore examine lagged relations less closely without being aware how unequally they areallocating their attention. In addition, the expense classification suggests a negative effecton profits; when individuals expect the wrong sign for a relation, they often estimate themagnitude of the relation less accurately (Muchinsky and Dudycha 1975; Sniezek 1986).

3 This argument assumes that individuals believe that information they do not possess has influenced the firm’schoice between capitalizing and expensing the intangible expenditure.


Subjects in our experiment receive data on gross profits and expenditures on a quality-improvement program (employee training, etc.) at 20 similar manufacturing plants. We askthem to use these data to learn how quality-program expenditures affect gross profits. Wethen ask them to use what they have learned to predict gross profits at an additional 20plants, given information on quality-program expenditures at these plants. The expendituresare classified either as expenses or as investments in assets. When the expenditures areclassified as expenses, we expect individuals to allocate less attention to lagged effects;therefore they should learn this relation less accurately and be less certain of the strengthof the lagged effect on profit, and therefore should tend to underestimate it. Because thelagged effect is an important determinant of profits in our experimental setting, we expectindividuals’ profit predictions to be less accurate when the intangibles are classified asexpenses.

H1a: Individuals’ profit predictions will be less accurate when expenditures on intan-gibles are expensed than when they are capitalized.

Our experimental design allows us to partition individuals’ prediction errors into severalcomponents that can have different causes and different practical consequences. AppendixA shows a two-step breakdown of mean squared prediction error (MSE). The first step isthe partition shown in Panel A (Theil 1966; Lee and Yates 1992):

2 2MSE � (Y � Y ) � (S � S ) � 2(1 � r )S S (1)s e Ys Ye a Ys Ye

where s (s � subject) is the mean of individuals’ profit predictions, e (e � environment)Y Yis the mean of actual profits, SYs and SYe are their corresponding standard deviations, andra is the Pearson correlation between the predicted and actual profits.

The three components of the partition represent different judgment errors. The first isbias: overall optimism or pessimism in predictions. If classifying the expenditures on in-tangibles as investments causes individuals to predict higher profits on average, regardlessof the exact magnitude and timing of the quality expenditures, then the bias measure willcapture this effect.

The second component of the partition is variability. Individuals’ judgments may varymore or less around the mean than the actual outcomes do, perhaps because individualstend to overreact or underreact generally to information provided. For example, an individ-ual who uses the mean value of profits as the prediction for each case would exhibit nobias but extreme underreaction to the information available. If individuals expect qualityexpenditures to have more (or less) effect on profits when they are classified as investments,regardless of the timing of the expenditures, then the variability measure will capture thiseffect.

The third component is a function of ra, which the psychology literature calls achieve-ment (Lee and Yates 1992; Cooksey 1996), magnified by variability in actual and predictedoutcomes. If individuals fail to identify the relative importance of different lags in theexpenditure-profit relation or fail to use this knowledge consistently in their predictions—forexample, if individuals expect contemporaneous effects of intangibles expenditures to belarge and lagged effects to be small when in fact the converse occurs—then ra will capturethis effect.

We predict fixation will persist in this experiment because we expect individuals in theexpense condition to underestimate the strength of the lagged effect of intangibles expen-ditures on profits, relative to individuals in the investment condition. This implies that ra


will be smaller when intangibles are expensed, and that the difference in ra across experi-mental conditions will be the principal source of difference in prediction error.

H1b: Individuals’ profit-prediction achievement (ra) will be lower when intangibles areexpensed than when they are capitalized.

Our explanation of expense vs. investment effects does not imply that individuals’predictions will be more optimistic or pessimistic on average or more (or less) responsiveto all intangible-expenditure information. Therefore we do not hypothesize differences inthe bias or variability of individuals’ profit predictions.

Panels B and C in Appendix A show how to decompose ra further to identify sourcesof difference in achievement. This decomposition, the so-called lens model (Tucker 1964;Lee and Yates 1992), is based on comparisons between two regression models of therelation between intangibles expenditures and profits. Regressing actual profits on actualexpenditures, using the data that individuals receive to learn the expenditure-profit relation,creates an environmental model (Equation [2], Panel B). Regressing an individual’s profitpredictions on the expenditures individuals receive as a basis for prediction creates a policy-capturing model for that individual. If the individual’s subjective analysis of the learningdata leads to the same inferences about expenditure-profit relations as a statistical (regres-sion) analysis, and if the individual consistently applies these inferences in making predic-tions, then the two models will be the same.

The lens-model equation below identifies potential sources of low achievement (ra) interms of correlations between the environmental and policy-capturing models.

2 1 / 2 2 1 / 2r � R GR � C(1 � R ) (1 � R ) . (3)a e s e s

Panel C of Appendix A shows how we calculate the components of this equation. Re

(environmental predictability) measures the accuracy with which a statistical model predictsprofits, using the data available to individuals. If the relation between profits and intangiblesexpenditures is deterministic, then Re equals 1.0. The less of the variation in actual profitsthat the intangibles-expenditure data can explain, the lower Re is.4 G, which the psychologyliterature refers to as matching (Lee and Yates 1992; Cooksey 1996), captures the similarityin the relative magnitudes of the coefficients between an individual’s policy-capturing model(Equation [3], Appendix A) and the environmental model (Equation [2], Appendix A).5 Rs

(consistency or cognitive control) captures the degree to which individuals use the samemodel without error from prediction to prediction. If unmodeled relations exist in the un-derlying data (e.g., if nonlinearities or interactions exist in the data but the statistical modelemployed is linear additive), then C captures the extent to which individual judgment in-corporates these relations.

In our study, Re is identical across conditions by design, and no nonlinear relationsexist in the underlying data (C � 0). Therefore only two of the lens-model measures,

4 Re is the square root of the environmental model’s unadjusted R2.5 Because G is a correlation measure, it does not capture differences between the models with respect to intercepts

or to absolute magnitudes of the coefficients. If an individual’s policy-capturing model is Ys � 5 � 10X1 � 5X2

while the environmental model is Ye � 1 � 2X1 � X2, then predictions from the two models would be perfectlycorrelated and G would be 1.0, even though the individual’s predictions are quite inaccurate. The inaccuracywould be fully captured in the first two components of the MSE partition, however, and therefore would not alsobe included in ra or its components. (The individual’s predictions would display optimism in the bias measureand overreaction in the variability measure.) In contrast, if the individual’s policy-capturing model is Ys � 5� 1X1 � 2X2—that is, if the relative weights on X1 and X2 differ from those in the environmental model—Gwould be substantially lower.


matching (G) and consistency (Rs), can differ across conditions. We predict that expensingthe intangibles expenditures will reduce matching, because individuals will underweight thelagged effect of intangibles on profits to a greater degree in the expense than in the in-vestment condition.

H2a: Matching (G) in individuals’ profit predictions will be lower when intangibles areexpensed than when they are capitalized.

H2b: In predicting profits, individuals will underweight the significant lagged effect ofintangibles expenditures more when these expenditures are expensed than whenthey are capitalized.

Matching is important in practice because it indicates an understanding of the relativemagnitudes of contemporaneous and lagged effects of intangibles expenditures. Consistency(Rs) is also important in practice, for two reasons. First, individuals’ judgments about theeffect of various expenditures on future profits often affect resource allocation decisions.Inconsistency in these judgments can impair resource allocation even if average judgment(as captured by G) is good. For example, individuals who make inconsistent judgments willsometimes overestimate future profits from a given type of expenditure and therefore over-spend, while at other times they will underestimate future profits from the same expenditureand thus underspend. Second, efficient contracting requires the contracting individuals tobe predictable: principals can design contracts to induce agents to take the actions theprincipals desire only if principals can predict how agents will respond to the incentivesoffered (Baiman 1982, 1990; Sunder 1999). Our consistency measure is an indicator ofpredictability. Thus, when individuals’ judgments (and the actions that depend on thesejudgments) are inconsistent, contracts will not necessarily have the effect principals intend.

Sniezek (1986) shows that individuals make more consistent judgments when the labelor referent of the data (in our case, ‘‘investment’’ vs. ‘‘expense’’) prompts them to examineimportant relations in the data early in the judgment process. In contrast, when individualsfocus most of their attention on a relation that proves to be unimportant, they are likely tofeel uncertain that they have a good basis on which to make judgments (profit predictions).They may therefore try multiple judgment strategies across cases, resulting in inconsistentjudgments. In our setting, this is likely to occur when individuals allocate more attentionto looking for a current-period expenditure-profit relation that is, in fact, a poor basis forprofit prediction. In contrast, when they allocate more attention to learning the more pre-dictive lagged effect, they should be more certain they have a good basis for profit predic-tion and thus should be more likely to use that basis consistently, rather than alternatingamong different judgment strategies. In consequence, we expect judgments to be less con-sistent when intangibles expenditures with future value are expensed than when they arecapitalized.

H2c: Individuals’ profit predictions will be less consistent (lower Rs) when intangiblesare expensed than when they are capitalized.

We also expect the accounting classification to affect consensus and self-insight. Con-sistent with prior literature (Ashton 1985; Stewart et al. 1997), we define consensus assimilarity in predictions across individuals. Low judgment consensus implies time-consuming discussion (and perhaps additional data collection and analysis) when individ-uals must establish agreement before making expenditure decisions. Low consensus can


also imply costly conflict and, in the extreme, impasse and failure to make expendituresthat would benefit the organization.

We expect the investment classification to lead to higher consensus. We expect indi-viduals in the investment condition to be more successful in detecting and using the stronglagged effect of intangibles expenditures on profits, which is present in the data all indi-viduals see. This common basis for prediction should lead to higher consensus. In theexpense condition, in contrast, we expect individuals to be less successful in finding asatisfactory basis in the data for predicting profits. They must rely more on idiosyncraticprediction strategies and prior beliefs that vary across individuals, resulting in lower judg-ment consensus.

H3: Individuals’ profit-prediction consensus will be lower when intangibles are ex-pensed than when they are capitalized.

Consistent with prior literature, we define self-insight as the degree to which individ-uals’ ex post explanations of how they made judgments correspond to how they actuallymade judgments (Cook and Stewart 1975). The lower individuals’ self-insight, the lessaccurately they explain the basis of their judgments and therefore the more difficult andcostly it is likely to be to resolve disagreements among individuals about the profits theyexpect from a given set of expenditures.

Self-insight can be low because only part of the judgment process is conscious. Whenindividuals explain how they made judgments, part of their explanation is an account oftheir consciously directed thought processes, and part is their best guess about processesthat are unobservable and/or difficult to recall accurately (Nisbett and Wilson 1977). Absentcomplete knowledge of their own judgment processes, individuals tend to report judgmentprocesses they think would be reasonable under the circumstances (Nisbett and Wilson1977).6 Thus, if they believe an expense should have a strong effect on contemporaneousprofits, then they will tend to report ex post that they weighted current-period expendituresheavily in the expense condition—even if they did not in fact do so. We expect that in theexpense condition individuals will not in fact consistently place heavy weights on current-period expenditures. As predicted above, they will tend to be uncertain and to predictinconsistently. Thus when intangibles are expensed, individuals’ actual use of informationin predicting profits will correspond poorly with their ex post explanations. In the investmentcondition, in contrast, individuals are likely to believe an investment should have a stronglagged effect on profits, and therefore they will tend to report ex post that they weightedthe lagged effect heavily in predicting profits. If, as predicted, they actually weight thelagged effect heavily in predictions, then correspondence between their actual and reporteduse of data will be high.

H4: Individuals’ self-insight into their profit predictions will be lower when intangiblesare expensed than when they are capitalized.

III. DESIGN OF EXPERIMENTSubjects

Thirty-one M.B.A. students who had completed a course in management accountingand six undergraduates who had completed a course in cost accounting volunteered to

6 Although the self-reports of judgment processes that Nisbett and Wilson (1977) recount appear to be sincere,some self-reports probably also reflect an element of impression management. Individuals may report weightsthat they believe will make them appear knowledgeable or rational.

Bima C. Pratama

Highlight


participate in this experiment. Full-time managerial experience (not including other workexperience) ranged from zero to ten years (mean of 35 months for M.B.A. students and 4months for undergraduates). We asked subjects with managerial experience to estimate thepercentage of time they had spent on budgeting (mean � 12 percent), forecasting andplanning (32 percent), making spending decisions (15 percent), advising on or recom-mending spending decisions (12 percent), and quality programs (9 percent). Because t-testsrevealed no significant (p � 0.05), differences between these two subject groups on themeans of any of the variables collected (except for work experience and number of ac-counting and math courses completed), the tests described below pool data from bothsubject groups.7 We paid subjects performance-contingent compensation, as describedbelow.

TaskSubjects received information about 20 similar manufacturing plants. The plants made

the same product and were built to the same design, using similar technology and produc-tion scale. All plants participated in a quality-improvement initiative, which included em-ployee training, process and quality engineering, and preventive maintenance. Internal ac-counting reports classified spending on this program as either a quality-improvementexpense or an investment (the experimental treatment). In the investment condition, the firmcapitalized the expenditures and straight-line amortized them, deducting the amortizationexpenses in computing gross profits. In the expense condition, the firm subtracted the entireamount of the expenditures in calculating gross profits in the quarter during which thespending occurred.

An introductory page in the experimental materials told subjects that because the plantswere so similar, the effect on quarterly gross profits of a dollar of spending on the quality-improvement program was roughly the same across plants. However, because the programwas new, the firm was still learning how the program affected quarterly gross profits andwhat the optimal level of spending was. Local managers had some freedom to experimentwith different quality-improvement spending levels. There were no seasonal variations inthe data and no significant external shocks or unusual internal events that would have alteredor masked the effects of quarterly quality-improvement expenses on quarterly gross profits.Therefore, if the quality-improvement program had a significant effect on quarterly grossprofits, then it should be detectable in the data.

After reading this introduction, each subject received expenditure data for 20 plants forthe just-completed quarter (t), expenditure data for the preceding three quarters (t � 1,t � 2, and t � 3), and actual gross profit for quarter t. This is the learning data set. Table1 shows the learning data subjects received, which was identical across experimental con-ditions except for the word ‘‘expense’’ or ‘‘investment’’ in the column heading.

The advantage of providing identical expenditure and profit data to subjects in bothconditions is enhanced experimental control; but the disadvantage is that the underlying

7 The groups differed in one respect consistent with the difference in experience: Prior beliefs about the effect ofquality expenditures on profit were more diverse among undergraduate subjects. The variance in prior be-liefs was significantly greater for undergraduates than for M.B.A.s (Levene’s F � 32.82, p � 0.00 for priorbeliefs about current-period effects; F � 30.81, p � 0.00 for prior beliefs about the sum of the next three periods’lagged effects). Although more experienced subjects exhibited more consensus before they saw the learning data,there was no difference across subjects groups in post-learning consensus. Variance in the t � 3 lagged coefficientwas a principal driver of low consensus (see Table 4); but this variance was not significantly greater for under-graduates than for M.B.A.s (F � 1.17, p � 0.05).

Bima C. Pratama

Highlight

Bima C. Pratama

Highlight

Bima C. Pratama

Highlight

Bima C. Pratama

Highlight

Bima C. Pratama

Highlight


TABLE 1Experimental Materials: Learning Data

Subjects received these data as a basis for learning the effect of quality improvement expenditureson gross profits at 20 similar manufacturing plants.

Plant

Actual qualityimprovement

,a 3quarters ago


,a 2quarters ago


,a

previousquarter


,a

quarter justcompleted

Actual grossprofits,

quarter justcompleted

A $1.368 M $1.890 M $0.795 M $0.984 M $21.976 M

B $0.900 M $2.933 M $1.647 M $1.534 M $12.562 M

C $0.997 M $0.886 M $0.247 M $2.337 M $18.544 M

D $1.542 M $2.439 M $2.146 M $1.440 M $24.734 M

E $1.995 M $1.065 M $0.984 M $1.449 M $26.174 M

F $0.959 M $1.135 M $1.170 M $2.261 M $16.181 M

G $2.032 M $2.080 M $1.275 M $1.277 M $27.841 M

H $0.888 M $0.853 M $0.798 M $1.886 M $19.354 M

I $1.144 M $0.567 M $0.669 M $1.287 M $17.480 M

J $2.301 M $1.552 M $1.082 M $1.826 M $25.622 M

K $1.294 M $2.206 M $2.277 M $0.793 M $22.972 M

L $1.540 M $0.644 M $2.357 M $1.163 M $20.897 M

M $1.078 M $0.247 M $1.498 M $0.649 M $22.246 M

N $0.001 M $1.718 M $1.709 M $1.590 M $ 9.125 M

O $1.319 M $2.183 M $1.020 M $0.214 M $18.915 M

P $1.115 M $1.665 M $0.634 M $0.789 M $22.628 M

Q $0.733 M $2.018 M $1.611 M $0.973 M $13.750 M

R $2.255 M $1.517 M $1.444 M $0.022 M $27.757 M

S $1.398 M $0.281 M $1.140 M $0.749 M $22.065 M

T $2.580 M $0.631 M $1.339 M $1.488 M $27.332 M

a For subjects assigned to the investment condition, ‘‘investment’’ was inserted in the blank. For subjects assignedto the expense condition, ‘‘expense’’ was inserted in the blank.

revenue-generating and expense-calculation processes were not identical in the two con-ditions. Profit at t in the investment condition included amortization expense from theexpenditure at t � 3, but profit at t in the expense condition did not. Therefore, in orderfor profit to be equal in the two conditions, revenues at t must also be larger in the in-vestment condition. Net cash flow at t � 3 was also likely to be higher in the investmentcondition.


We chose to allow a difference across conditions in underlying (unobservable) revenueand expense processes, rather than in the observed data subjects used in learning andjudgment. The difference in unobservable processes, if it had any effect, worked againstfinding support for the study’s hypotheses. If subjects tried to disentangle the revenue andexpense effects, rather than simply to relate expenditures to total profits as required by theexperimental task, then profit prediction would be more difficult in the investment condition.(The multiple-period amortization requires more calculation steps and is complicated byuncertainty about the length of the intangible asset’s life.) This effect biases against findingthe predicted superior judgment in the investment condition.

After studying the learning data without a calculator, subjects received a set of expen-diture data from another 20 similar plants (the judgment data set) and predicted gross profitsfor these plants based on the quality-improvement expenditure data. As in the learning data,the expenditures were classified as either investments or expenses. Subjects could retainand refer to the learning data when they examined the new data set and made theirjudgments.

The presentation format of the judgment data was identical to that in Table 1, exceptthat the profit column was blank. The experimental materials told subjects that qualityexpenditure data were available in a timely fashion, but that the accounting departmenttook up to 15 business days to finalize profit calculations. The managers in the firm, how-ever, wanted an estimate of quarterly gross profits immediately at the end of the quarter,so each subject’s job was to predict quarterly gross profits using the quarterly qualityspending data.

A computer program created both learning and judgment data sets from a model withspecified population means, variances, and correlations.8 Table 2 shows statistical propertiesof the data subjects received. Parameters specified in the model were identical for bothlearning and judgment data sets, but as Table 2 shows, sampling variation resulted in slightdifferences between the learning and judgment data sets in the sample means and corre-lations. None of the differences between learning and judgment data was statistically sig-nificant (p � 0.05).

Gross profits at time t had (1) a negative but insignificant correlation with spending att, (2) no significant correlation with spending at t � 1 or t � 2, and (3) a strong positivecorrelation with spending at t � 3. The expense at t was small enough relative to grossprofits that its contemporaneous effect was swamped by other sources of variation ingross profits; but the effect on gross profits at t�3 was substantially larger and thus dom-inated the noise. The three-period lag this study employed was consistent with archival datafrom manufacturing. Ponemon et al. (1994), using data from 47 paper and pulp mills, showthe strength of the relation between prevention expenditures and failure-cost reductions(i.e., profit increases) peaking at about 8 months after the expenditure for internal failurecost reductions and 13 months after the expenditure for external failure cost reductions. Inan analysis of annual quality data from 12 plants of a Fortune 500 firm, Ittner et al. (2001)find a significant effect of prior years’ quality-improvement (prevention) expenditures oncurrent-year defects (which in turn affect current-year profits via nonconformance costs),

8 The process was similar to using a random number generator to create samples from a distribution of a singlerandom variable with a specified mean and variance; but it was more complex in that it created data for fivevariables (four quarters of expenditures and one quarter of gross profits) with specified correlations among them.


TABLE 2Parameters of Experimental Data

Learning Data

QuarterGross Profit

t

Quality Spending

t t � 1 t � 2 t � 3

Mean $20,907,800 $1,238,000 $1,292,100 $1,425,500 $1,372,000

S.D. $ 5,222,000 $ 609,600 $ 560,000 $ 766,300 $ 617,900

Pearson Correlations:

Gross Profit �0.27 �0.02 �0.17 0.90*Quality Spendingt �0.23 �0.12 �0.18Quality Spendingt�1 0.30 0.01Quality Spendingt�2 �0.12

Judgment Data

QuarterGross Profit

t

Quality Spending

t t � 1 t � 2 t � 3

Mean $20,688,800 $1,187,900 $1,133,800 $1,513,900 $1,503,300

S.D. $ 5,164,700 $ 547,600 $ 476,700 $ 626,500 $ 628,700

Pearson Correlations:

Gross Profit �0.36 0.17 0.07 0.95*Quality Spendingt �0.33 0.09 �0.28Quality Spendingt�1 �0.03 0.13Quality Spendingt�2 0.04

* Correlation differs significantly from zero (p � 0.05).

but no significant effect of current-year quality-improvement expenditure on current-yeardefects.9

Independent VariableThe accounting treatment was a between-subjects variable. In the learning and judgment

data, we classified quarterly quality-improvement expenditures as either an investment oran expense.

9 Other key characteristics of the experimental data set included the lack of serial correlation among the Xi’s andthe strength of the X–Y relation. See Ponemon et al. (1994) for an illustration of quality costs that fluctuatemarkedly, with no apparent pattern, from period to period. The relation between quality expenditures and grossprofits was stronger in the experimental materials than it was likely to be in the natural environment, to allowsubjects a reasonable chance to detect the quality-profit relation even in the expense condition. This choice didnot bias toward finding the expected results; if expense-condition subjects had difficulty identifying the laggedrelation even when it was relatively easy to see, then they were unlikely to perform better when it was difficultto see.


Dependent VariablesWe used each subject’s gross profit predictions to develop the dependent variables. We

measured judgment accuracy (H1a) for each subject as the mean absolute error of thesubject’s gross profit predictions:10

� �Y � Y � /20, j � 1,...20,j sj ej

where:

Ysj � subject s’s profit prediction for plant j, andYej � actual profit for plant j.

Appendix A shows the calculation for judgment achievement (ra), matching (G), and con-sistency (Rs), the dependent measures for H1b, H2a, and H2c, respectively. The measurefor H2b was the difference between standardized coefficients on intangibles expendituresat t � 3 in the environmental model and subjects’ policy-capturing models (standardizedvalues of bst�3 � bet�3 in the models shown in Appendix A).11

We used two measures of judgment consensus (H3). One, used frequently in lens-modelstudies (Ashton 1985; Ashton 1992; Stewart et al. 1997), was the mean of the Pearsoncorrelations between the judgments of each pair of subjects in an experimental condition.The second was the variance of the coefficients in subjects’ policy-capturing models (�s’sin Equation [2], Appendix A).

Also consistent with prior literature (Cook and Stewart 1975), we measured self-insight(H4) as the correlation between two sets of gross profit predictions: (1) predictions basedon the policy-capturing model estimated for each subject (Ys in Appendix A) and (2)predictions based on the weights that subjects supplied ex post when we asked how im-portant each of the four periods of expenditure data was in making their predictions. Ifthey correctly reported the relative impacts of the four periods of expenditure data on theirpredictions, the self-insight measure would equal 1.0. To the extent that the relative weightsthey reported differed from the relative weights in their policy-capturing models, the mea-sure would be smaller.

ProcedureWe randomly assigned each subject to one of the two treatment conditions (expense or

investment). Subjects self-paced their way through the experimental materials in a lab set-ting. The first section of the materials, which subjects completed and returned before seeingthe rest of materials, collected information on subjects’ prior beliefs about the effect ongross profits of spending on a quality-improvement program in a manufacturing facility.The description of this program (which included employee training, process improvement,etc.) was the same as the description that appeared subsequently in the main task of theexperiment. Subjects estimated the profit impact of a given level of spending on the qualityprogram in the quarter of the expenditure and in each of several subsequent quarters. Wetold subjects in the investment condition that the expenditures were capitalized, and subjectsin the expense condition that they were expensed.

10 We also calculated and tested mean squared prediction error (MSE), since the partitioning described in SectionII applies to MSE, not to mean absolute error. See footnote 13 for results. We used mean absolute predictionerror as the primary dependent variable because it was more nearly normally distributed and more interpretable.

11 The psychology literature often uses standardized coefficients because they facilitate comparisons across judg-ment tasks with differently scaled predictors (Cooksey 1996).


This estimate captured two factors that we needed to control in the hypothesis tests.One was subjects’ prior beliefs about the benefits of spending on quality programs. Theother was subjects’ beliefs about the signaling value of the accounting classification. If theybelieved that management was conveying credible information about the timing of theprogram’s benefits by classifying expenditures as expenses or investments, their initial es-timate would incorporate this belief.

The experimental materials then described the compensation system for the experiment.Subjects’ payment depended on the accuracy of their profit predictions and could rangefrom $6 to $20. We used a quadratic loss function because we assessed the accuracy ofsubjects’ judgments relative to the best-possible judgments, i.e., predictions from an OLSmodel based on a quadratic loss function. We calculated a squared error measure for eachsubject summed over his or her 20 judgments: � (your judgment � best possible judg-ment)2. Cash payment was inversely related to the magnitude of the error measure.

After learning about the compensation system, subjects examined the learning andjudgment data and accompanying instructions, and made their profit predictions. After theyturned in these materials, subjects allocated 100 points across the four periods of quality-expenditure information, indicating their relative importance to subjects’ judgments.12 Thenext section of the experimental materials asked questions related to the just-completedjudgment task (how difficult it was, how familiar, etc.). The final section asked the subjectsto identify, in retrospect, what they were thinking about quality-profit relations as theymade their predictions. Eight response alternatives (including an ‘‘other’’ category) identi-fied possible causal relations with different temporal (contemporaneous, lag) and directional(increase, decrease) properties and causal mechanisms (see Appendix B). Subjects allocated100 points across these alternatives based on how important each had been to their thinkingwhen they predicted profits. They also provided demographic data (work experience, edu-cation level, etc.).

IV. RESULTSHypothesis 1a posited that individuals’ profit predictions would be less accurate when

the intangibles expenditures were expensed than when they were capitalized. Table 3 showsthe mean absolute error in subjects’ profit predictions in the expense and investment con-ditions and the key components of the mean squared error partition. Consistent with H1a,subjects’ mean absolute error was about 25 percent larger in the expense condition ($4.81million vs. $3.94 million), and this difference was significant (t � 1.98, one-tailed p� 0.03).13

As described in Section II, there were three possible sources for difference in meansquared error (and thus in mean absolute error) across conditions: bias, variability, andachievement (ra). Bias occurred if the mean predicted profit differed systematically fromthe mean actual profit. Subjects’ mean profit predictions ($19.64 million in the expensecondition and $19.65 million in the investment condition) were not significantly differentfrom actual profits in the judgment data set ($20.7 million) in either condition (t � 0.63,

12 The psychology literature elicits subjects’ perceptions of the relative importance of different predictors in avariety of ways (e.g., allocating 100 points, asking for ratings on Likert scales). Results for self-insight hypoth-eses are not sensitive to elicitation methods, however; Cook and Stewart (1975) find that seven methods ofeliciting these perceptions yield qualitatively the same results in tests of self-insight.

13 Mean squared prediction error was also significantly larger in the expense condition. Because the distributionof MSE was strongly influenced by extreme errors and highly skewed, we performed a t-test on ln(MSE) anda Mann-Whitney test on MSE (one-tailed p � 0.06 for both tests). Mean absolute prediction error was alsosomewhat skewed, although less than MSE; we therefore also performed a Mann-Whitney test on mean absoluteerror, which yielded similar results (one-tailed p � 0.01).


TABLE 3Judgment Accuracy

Mean Absolute Error, Bias, Variability, and Achievement

Expense(n � 18)

Mean(Std. Dev.)

Investment(n � 19)

Mean(Std. Dev.)

t-statistic(two-tailed p)

Mean absolute error in gross profit predictions $4.81(1.22)

$3.94(1.43)

1.98(0.06)

Mean gross profit prediction $19.64(2.42)

$19.65(2.88)

0.01(0.99)

Standard deviation of predictions $4.59(2.13)

$4.59(1.89)

0.00(0.99)

ra (Achievement) 0.34(0.40)

0.68(0.21)

3.19(0.01)

The first row shows mean absolute errors in subjects’ predictions of gross profits when expenditures on intangibleswere classified as expenses and as investments. The three following rows show means of key elements in thepartition of mean squared error shown in Appendix A, Panel A: the mean and standard deviation of subjects’ profitpredictions and ra (achievement). Measures are in millions of dollars except for ra, which is a correlation measure.The t-test for ra is performed on Fisher Z-transformations of subjects’ ra.

p � 0.54, expense; t � 0.52, p � 0.61, investment). Mean profit predictions were almostidentical in the two experimental conditions (t � 0.01, p � 0.99) and thus cannot accountfor differences across conditions in prediction error. Classifying the intangibles expendituresas expenses rather than investments did not lead subjects to make more optimistic or pes-simistic predictions overall.

Variability errors would have occurred if the standard deviation of predicted profitsdiffered from the standard deviation of actual profits. The mean of the standard deviationof subjects’ predictions ($4.59 million in each condition, Table 3) was not significantlydifferent from the standard deviation of actual profits in the judgment data set ($5.16million, Table 2) in either condition (t � 1.13, p � 0.28, expense; t � 1.32, p � 0.21,investment). The mean standard deviations of subjects’ predictions were quite similar acrossconditions (t � 0.00, p � 0.99). Thus, differences in variability cannot account for differ-ences across conditions in prediction error. Classifying the expenditures as expenses ratherthan investments did not, in general, cause subjects to overreact or underreact to the ex-penditure data. Subjects in both conditions clearly learned from the data they received,because they generated profit predictions with means and standard deviations that closelymatched the learning data set.

Hypothesis 1b predicted that achievement would be lower in the expense than in theinvestment condition. This difference in achievement is the principal source of differencein prediction error across conditions, since bias and variability are nearly identical in ex-pense and investment conditions. Consistent with H1b, Table 3 shows that mean ra wassignificantly lower in the expense condition (0.34) than in the investment condition (0.68;t � 3.19, p � 0.01).

Hypothesis 2a predicted that matching would be lower when intangibles were expensedthan when they were capitalized. As the first row in Table 4 shows, the mean level of


TABLE 4Comparison of Subjects’ Policy-Capturing Models with: (1) Environmental Model (Matching)

and (2) Subjects’ Predictions (Consistency)

Expense(n � 18)

Mean(Std. Dev.)


Mean(Std. Dev.)

t-statistic(two-tailed p)

Matching (G) 0.52(0.57)

0.85(0.17)

2.33(0.03)

�st � �et 0.14(0.29)

0.19(0.28)

0.52(0.61)

�st�1 � �et�1 0.14(0.23)

0.15(0.20)

0.41(0.69)

�st�2 � �et�2 0.21(0.29)

0.24(0.24)

0.27(0.79)

�st�3 � �et�3 �0.48(0.46)

�0.14(0.22)

2.86(0.01)

Consistency (Rs) 0.69(0.26)

0.86(0.15)

2.40(0.02)

Matching (G) measures the similarity between subjects’ policy-capturing models and an environmental model (seeAppendix A for definitions of measures and models). �st � �et is the difference between the standardized coefficienton contemporaneous intangibles expenditure in subjects’ policy-capturing models and the corresponding coeffi-cient in the environmental model (and similarly for t�1, t�2, and t�3). Consistency (Rs) measures the degree towhich subjects use the same model without error in multiple predictions.

matching (G) was significantly lower in the expense (0.52) than in the investment condition(0.85; t � 2.33, p � 0.03), supporting H2a.

G measured the simultaneous effect of all four coefficients in each subject’s policy-capturing model and did not indicate what was specifically wrong with subjects’ models inthe expense condition. Rows two to five of Table 4 show mean differences between thestandardized coefficient (�e) on each quarterly spending variable in the environmental model(Equation [1], Appendix A) and the corresponding standardized coefficient in subjects’predictive models (�s) (Equation [2] in Appendix A) in each condition. The accountingclassification of expenditures on intangibles did not affect the accuracy of subject-modelcoefficients on contemporaneous intangibles expenditures (�st � �et) (t � 0.52, p � 0.61),nor on expenditures at t � 1 and t � 2 (t � 0.41, p � 0.69 and t � 0.27, p � 0.79,respectively). Consistent with H2b, however, mean errors in subject-model coefficients forthe three-period lag (�st�3 – �et�3) were significantly greater when intangibles expenditureswere classified as expenses (�0.48) than when they were classified as investments (�0.14;


t � 2.86, p � 0.01).14 Expensing the intangibles caused subjects to significantly underes-timate the strength of this important positive lagged effect.15

The last row in Table 4 shows results for consistency (Rs). As predicted in H2c, sub-jects’ mean level of consistency was lower in the expense condition (0.69) than in theinvestment condition (0.86; t � 2.40, p � 0.02).

Hypothesis 3 predicted lower judgment consensus in the expense condition than in theinvestment condition. As Table 5 shows, mean pairwise correlations between subjects’ pre-dictions were lower in the expense condition (r � 0.19) than in the investment condition(r � 0.58). We did not perform statistical tests of this difference, however, because thecorrelations were not independent (each subject’s predictions were correlated with the pre-dictions of all other subjects in his or her experimental condition). Individual subjects’policy-capturing models provided an alternative measure of consensus. Variation in sub-jects’ models (�s’s) created variation in predictions (lack of consensus). (Table 4 reportsstandard deviations of �s’s.) The variance of �st�3 was significantly greater in the expensecondition than in the investment condition, indicating less consensus about the key predictorof profits in the expense condition (0.21 vs. 0.05; F � 4.57, p � 0.01). Variances of �st,�st�1, and �st�2 did not differ significantly across conditions: F’s � 1.50, p’s � 0.10. Theseresults supported H3.

Hypothesis 4 predicted that self-insight would be lower in the expense condition thanin the investment condition. The last row in Table 5 shows the self-insight measure, whichcaptured the similarity between the relative weights on expenditure variables in subjects’policy-capturing models and the relative weights subjects provided ex post. Mean self-insight was significantly lower in the expense condition (0.47) than in the investment con-dition (0.85; t � 2.08, p � 0.05), consistent with H4.

Supplementary AnalysesThe distributions of most of the judgment performance measures were somewhat

skewed. We therefore performed nonparametric (Mann-Whitney U) tests of the hypotheses.Results did not differ qualitatively from the parametric results reported above. We alsotested all hypotheses with a reduced sample, omitting the six undergraduate subjects, usingboth parametric and nonparametric tests. Results were qualitatively similar except for thenonparametric test of self-insight, which did not reach conventional levels of significance(one-tailed p � 0.16).

Prior BeliefsWe have argued that subjects’ profit predictions differed in the expense and investment

conditions because subjects processed the learning data differently in the two conditions.One potential alternative explanation was that profit predictions differed because subjects

14 Analysis of unstandardized coefficients led to qualitatively similar inferences. We also obtained qualitativelysimilar results with a MANOVA on all four coefficients, which allows for the potential within-individual cor-relations of the four coefficients.

15 An alternative analysis pooled all subjects’ profit predictions and estimated the following regression: Ysj � a� b1Xjt � b2Xjt�1 � b3 Xjt�2 � b4Xjt�3 � Z � b5ZXjt � b6ZXjt�1 � b7ZXjt�2 � b8ZXjt�3, where Ysj � profitprediction by subject s for plant j, Xjt � expenditure at plant j at time t, and Z � 0 in expense condition, 1 ininvestment condition. The adjusted R2 of the pooled regression was 0.32 (F � 37.74, p � 0.001). The experi-mental condition significantly affected subjects’ use of the expenditure data at t � 3; b8 was significantly positive(t � 3.90, p � 0.001). Consistent with the analysis in Table 4, the experimental condition did not affect subjects’use of the expenditure data at t, t � 1, or t � 2; b5, b6, and b7 were not significantly different from zero (t’s� 0.60, p’s � 0.54).


TABLE 5Judgment Consensus and Self-Insight

Expense(n � 18)


Expense vs. InvestmentTests

Consensus: Mean pairwise correlations betweensubjects’ gross profit predictions 0.19 0.58 NA

Consensus: Variance of �st�3 0.21 0.05 F � 4.57 (p � 0.01)

Self-insight mean (std. dev.) 0.47(0.60)

0.85(0.18)

t � 2.08 (p � 0.05)

See Appendix A, Panel C for calculation of measures.

in the investment condition had prior beliefs more closely resembling the underlying rela-tions in the learning data than did subjects in the expense condition. A related alternativewas that subjects believed the accounting classification provided a credible signal about themagnitude and timing of the profit impacts of expenditures on quality improvement, andthat this belief directly affected their profit predictions, independent of the way they pro-cessed the learning data. For example, subjects in the expense condition might have ac-curately estimated the three-period lagged effect in the learning data, but discounted it injudgment because of a belief that the accounting classification signaled a minimal laggedeffect.

We controlled for both of these alternatives by performing analyses of covariance onthe six key dependent measures: mean absolute error, achievement (ra), matching (G), errorin subjects’ �st�3 coefficients, consistency (Rs), and self-insight. The covariates in theseanalyses were subjects’ pre-experiment estimates of the contemporaneous and future effectson gross profits of quality-program expenditures, classified as either investments or ex-penses.16 The covariates were at least marginally significant (one-tailed p � 0.10) in theanalyses of mean absolute error, achievement, matching, and the error in the t � 3 coeffi-cient, but not in the analyses of consistency and self-insight. The signs of the coefficientson the covariates were intuitively plausible. When subjects expected the lagged effect tobe large, the t � 3 coefficients in their policy-capturing models were larger, their achieve-ment and matching scores were higher, and their mean absolute error was smaller. Whenthey expected the current-period effect to be large, their t � 3 coefficients were smaller,their achievement and matching scores were lower, and their mean absolute error wasgreater.

After controlling for the effects of these prior beliefs, however, the effect of the ac-counting classification remained significant (two-tailed p � 0.05) for each of the dependentvariables examined. This result indicated that the judgment differences between expenseand investment conditions were not due solely to prior beliefs about quality expense andinvestment, independent of what subjects inferred from the learning data set.

16 The future-quarter prior-belief estimate was the sum of the profit effects subjects estimated for the first, second,and third quarters after the expenditure. We obtained the same qualitative results using the estimated effect forthe third quarter after the expenditure and the sum of all future quarters’ estimated effects. Subjects’ mean(standard deviation) estimate of the effect of $1 of expenditure on the quality program was $1.58 ($8.44) forthe current quarter and $1.61 ($6.71) for the next three quarters.


Attention DirectingThe self-insight results suggested the limitations of retrospective reports, but also in-

dicated that these reports were informative; the self-insight measure was significantly pos-itive (one-tailed p � 0.05) for 32 of the 37 subjects. Subjects’ retrospective reports of theirjudgment processes (see Appendix B) further supported our claim that the expense andinvestment classifications focused attention on different subsets of the learning data. Theinvestment classification focused individuals’ attention more on future-period effects (ex-planations 5–7 in Appendix B), whereas the expense classification focused their attentionmore on current-period effects (explanations 1–4 in Appendix B). On average, subjects inthe investment condition assigned 28.2 more points to future-period explanations than tocurrent-period explanations, while subjects in the expense condition assigned 12.5 fewerpoints to future-period than to current-period explanations. This difference in point assign-ment between investment and expense conditions was significant (t � 2.91, p � 0.01).

V. DISCUSSION AND FUTURE RESEARCHPrior research has questioned whether individuals will learn to see through accounting

when significant learning opportunities are available (Wilner and Birnberg 1986; Lipe1998), but evidence on learning has been limited. Our experiment shows that opportunityto learn is not necessarily a quick remedy for accounting fixation, because accounting canaffect the allocation of attention and thus can influence the learning process itself. We donot suggest that people are incapable of overcoming fixation and learning to make correctinferences from accounting data. With fewer competing demands on attention, additionalcues to direct attention to important relations, or incentives that increase total attention,individuals might allocate enough attention to a lagged expense-profit relation to predictprofits more accurately. However, all else equal, people can arrive at accurate predictionsmore quickly and easily when accounting is closer to economic reality. This means learningat lower cost in terms of data, attention, and effort (lower out-of-pocket compensation costsfor effort and lower opportunity costs of directing attention away from other issues). Insome settings there may also be competitive-advantage benefits to learning key economicrelations more quickly.

Our results also contribute to understanding what it means to be a sophisticated userof accounting information. Early fixation studies see a lack of accounting-calculation knowl-edge as the reason for judgment biases: ‘‘If outputs from different accounting methods arecalled by the same name, such as profit, cost, etc., people who do not understand accountingwill tend to neglect the fact that alternative methods may have been used to prepare theoutputs’’ (Ijiri et al. 1966, 194, emphasis added; see Ball 1972, 1 for a similar argument).In this view, individuals should not fixate if they have sufficient knowledge of relevantaccounting rules.

More recent studies have emphasized the importance of category knowledge aboutindirect economic-causal effects of accounting; for example, experienced analysts knowthat decreases in stock price typically accompany one category of financing (new equityissues) but not another (new debt issues) (Hopkins 1996). Individuals who possess thisknowledge may appear more sophisticated than those who do not; but they are also morelikely to display fixation when encountering an atypical instance of the category, such asa hybrid security classified as debt or equity (Hopkins 1996; Libby et al. forthcoming).Additional encounters with atypical instances could help individuals refine their categoryknowledge and deal appropriately with atypical members of a category, however. In thisview, the sophistication required to avoid fixation on accounting would consist of refined


category knowledge that clearly distinguishes atypical from typical instances and supportscorrect expectations about the behavior of atypical instances.

Our study suggests that the requirements of user sophistication can be even more de-manding; a clear understanding of atypical instances in principle is not always sufficient toeliminate fixation in practice. Intangibles are an atypical expense in that they affect future-period profits, whereas typical expenses do not. Our subjects understand this, as shown intheir responses to the pre-experiment question about the effect of an intangibles expenseon future profits. Subjects in the expense condition expect a lagged effect on profits—buttheir performance in estimating this effect and using it in judgments is still significantlyworse than that of subjects in the investment condition.

Subjects in the expense condition perform relatively poorly because they look lessclosely at the most important data, not because they draw erroneous conclusions from thedata at which they do look closely. (Table 4 showed that subjects in the expense conditiondo not overestimate the current-period effects of the expenditures, compared to subjects inthe investment condition.) This attention-allocation story may help to explain the findingsof prior research about the effectiveness of learning opportunities in reducing fixation. Forexample, in Waller et al. (1999), fixation diminishes rapidly with opportunities to learn,perhaps because the experimental setting poses no significant attention-allocation problem.Subjects in that experiment receive cost information on a single product, choose a pricefor the product, and then learn the profit they make, which is a deterministic function ofprice. Thus, subjects in Waller et al. (1999) must learn a deterministic relation between oneX (price) and one Y (profit), rather than a probabilistic relation between four Xi’s (fourperiods of expenditures) and one Y (profit), as in our experiment. The likelihood thatsubjects will not closely examine data on the right Xi–Y relation is presumably minimalwhen there is only one Xi.

17 Similarly, in Gupta and King (1997) subjects must learn toestimate costs for three products by observing the relation between their cost estimates (Xi)and aggregate profits (Y); the more accurate their estimates are, the higher their profits.Learning is slower when subjects must estimate three Xi � Y relations (because the ac-counting system provides inaccurate data on all three) than when they must estimate onlytwo (because the accounting system estimates one product cost correctly).

One potentially important difference between our experiment and Waller et al. (1999)and Gupta and King (1997) is the way we presented the learning data. Subjects in Walleret al. (1999) and Gupta and King (1997) decided on prices or cost estimates in each trialof the experiment before seeing the actual profit outcome. In the learning phase of ourexperiment, subjects learned profit outcomes of intangibles expenditures without having tomake predictions first. Some psychologists have argued that requiring predictions or deci-sions during the learning phase should promote learning because it creates greater involve-ment; others have argued that it should reduce learning because it inhibits integration oflearning data across cases or causes anchoring on initial judgments, which are likely to beinaccurate because they are made on the basis of very little data (Klayman 1988;Broniarczyk and Alba 1994; Well et al. 1988). Experiments have found, however, thatrequiring case-by-case predictions during the learning phase makes no difference to per-formance in the subsequent judgment phase, other things equal (Broniarczyk and Alba1994; Well et al. 1988). It seems unlikely, therefore, that our experimental results are drivenby the absence of prediction requirements during the learning phase.

17 Subjects actually choose a price-quantity combination in Waller et al. (1999), but quantity errors are rare; thelearning that takes place in the study is predominantly reduction in pricing error.


Broniarczyk and Alba (1994) also find, however, that simultaneous presentation of allthe learning cases (as in our experiment) leads to poorer learning than does a sequentialpresentation that requires subjects to examine each case for some time before they can seethe next. These two presentation formats correspond to different real-world tasks—for ex-ample, examining reports from a dozen similar business units for a given period simulta-neously vs. examining reports from a given unit over time. Our experimental setting cor-responds to the first situation, and the results may not be fully generalizable to the second.

Another limitation to the generalizability of this paper’s results is that our experimentalsetting does not include actions some firms take to aid potentially faulty judgment. Oursubjects made their judgments alone; the opportunity to discuss the data with others mightmitigate or exacerbate the performance deficits observed in the expense condition. Oursubjects also make their judgments without the aid of formal statistical methods that mighthelp overcome judgment biases. Task-properties feedback leads to better judgment perform-ance than outcome information alone (Bonner and Walker 1994); and in tasks like ours,task-properties feedback is defined as statistical information such as correlations and re-gression coefficients (Balzer et al. 1989). Some firms provide statistical models of thedrivers of financial performance to employees as substitutes for subjective estimation, butmany do not; the efficacy of statistical models is a contested issue (Kaplan and Norton1996b; Ittner and Larcker 1998), and individuals’ willingness to rely on such models insteadof their own subjective judgment is an open question.

When individuals make subjective estimates, other attention-directing devices cancounter accounting methods’ tendency to lead managers to misallocate attention. For ex-ample, additional nonfinancial data may prompt closer examination of lagged relations bydrawing attention to the link from expenditure at one point in time to increased revenuesor decreased operating costs at later times by way of intermediate-period improvements inquality, productivity, customer satisfaction, etc.

The usual limitations of laboratory research suggest caution in assuming that our resultswill replicate fully in natural settings. The task and data are simplified and stylized, andsome subjects have limited work experience—although many have relevant experience, andexperience differences did not drive results. Sufficient specialized experience can provideknowledge about intangibles that eliminates the judgment task represented in this experi-ment; that is, it can create settings in which there is no longer significant ex ante uncertaintyabout the timing of benefits from intangible expenditures. In many important settings, how-ever, individuals lack this specialized knowledge because they are facing new situationsand must learn from observing available data. For many expenditures on ‘‘softer’’ intan-gibles such as employee training and process improvement, it is far from clear how soonfirms should expect to reap benefits from a particular new initiative, or how long additionalbenefits will persist. Even experienced managers depend on accounting data to infer thetiming and magnitude of these effects.


APPENDIX APrediction Error and Lens Model Statistics

Panel A: Partition of Mean Squared Prediction Error

2 2MSE � (Y � Y ) � (S � S ) � 2(1 � r )S S (1)s e Ys Ye a Ys Ye

where:

MSE � �j(Ysj � Yej)2 /20, j � 1,...,20;Ysj � subject s’s gross profit prediction for plant j;Y �ej actual (environmental) gross profit for plant j;

s �Y �jYsj /20;e �Y �jYej /20;

SYs � standard deviation of Ys;SYe � standard deviation of Ye; and

ra � r(Ye ,Ys).

Panel B: Modeling the Environment and the Individual

Learning data subjects receivedYei � actual (environmental) gross profit at plant i in quarter t, i � 1,...,20.Xit, Xit�1, Xit�2, Xit�3 � quality expenditure at plant i in quarter t, t � 1, etc.

Judgment data subjects receivedXjt, Xjt�1, Xjt�2, Xjt�3 � quality expenditure at plant j in quarter t, t � 1, etc., j � 1,...,20.

Predictions and outcomesYsj � subject s’s prediction of gross profit for plant j in quarter t.Yej � actual (environmental) gross profit for plant j in quarter t.

Environmental modelEstimated by regressing Yei on Xit, Xit�1, Xit�2, and Xit�3:

Y � a � b X � b X � b X � b X . (2)e e 1e t 2e t�1 3e t�2 4e t�3

Individual policy-capturing modelEstimated separately for each subject by regressing Ysj on Xjt,a Xjt�1, Xjt�2, and Xjt�3

Y � a � b X � b X � b X � b X . (3)s s 1s t 2s t�1 3s t�2 4s t�3

Panel C: Lens Model Equation and Measures

2 1 / 2 2 1 / 2r � R GR � C(1 � R ) (1 � R ) (4)a e s e s

where:

ra (achievement) � r(Ye, Ys);Re (environmental predictability) � r(Ye, Ye);

G (matching) � r(Ye, Ys);Rs (consistency) � r(Ys, Ys);

C (residual achievement) � r(dYe, dYs);dYej � Yej � Ysj; anddYsj � Ysj � Ysj.


Additional measures

Self-insight � r(Ys, Yw),

where Yw � b1wXt � b2wXt�1 � b3wXt�2 � b4wXt�3 and bw is the weight provided by subject s in apost-experiment question.

Consensus (pairwise) � r(Ysm, Ysn),

where m and n are individual subjects, m � n. Pairwise correlations are averaged to provideconsensus measures for larger groups.

a Note that, consistent with prior literature (Libby 1981; Lee and Yates 1992; Cooksey 1996), we use the judgmentdata to estimate subjects’ policy-capturing models, but use the learning data to estimate the environmental model.Although subjects reveal their policy-capturing models through use of the judgment data, they estimate theirmodels from the learning data. The appropriate measure of the quality of their subjective estimation is thereforea statistical model estimated from the learning data.

APPENDIX BRetrospective Thoughts about How

Quality-Improvement Expenditures Affect Profits

Please try to remember how you thought quarterly quality-improvement expenses affect quarterlygross profits when you made your quarterly gross profit predictions. Allocate 100 points acrossthe following factors, based on how important you thought they were when you made yourpredictions. Enter zero for any explanations that did not occur to you when you were making yourpredictions. [Mean responses for expense (E) and investment (I) conditions are shown in the tworight-hand columns.]

Means

E I

1. Increased quality-improvement expenses in the current quarterdecrease profits in that quarter because they are subtractedfrom revenues in the profit calculation.

14.6 13.7

2. Increased quality-improvement expenses in the current quarterdecrease profits in that quarter because they cause tempo-rarily disruptive and costly changes in products and/or oper-ating processes.

14.6 9.5

3. Increased quality-improvement expenses in the current quarterincrease profits in that quarter because improved qualitymakes the product more attractive to customers and increasesthat quarter’s sales revenues.

12.8 1.8

4. Increased quality-improvement expenses in the current quarterincrease profits in that quarter because they make the oper-ating processes more efficient and reduce product costs thatquarter.

13.9 1.6

5. Increased quality-improvement expenses in the current quarterincrease profits in future quarters because the product grad-ually becomes more attractive to customers, as they becomeaware of the improvement.

20.0 20.3


6. Increased quality-improvement expenses in the current quarterincrease profits in future quarters as the full effect of in-creased operating efficiencies is gradually realized.

19.2 31.1

7. Increased quality-improvement expenses in the current quarterdecrease profits in future quarters because they take moneyaway from other, more profitable investments your firm couldmake.

4.2 3.4

8. Other (explain): 0.8 18.2

REFERENCESAboody, D., and B. Lev. 1998. The value relevance of intangibles: The case of software capitalization.

Journal of Accounting Research 36 (Supplement): 161–191.Ashton, A. H. 1985. Does consensus imply accuracy in accounting studies of decision making? The

Accounting Review 60 (2): 173–185.Ashton, R. H. 1992. Effects of justification and a mechanical aid on judgment performance. Orga-

nizational Behavior and Human Decision Processes 52: 292–306.Baiman, S. 1982. Agency research in managerial accounting: A survey. Journal of Accounting Lit-

erature 1: 154–213.———. 1990. Agency research in managerial accounting: A second look. Accounting, Organizations

and Society 15 (4): 341–371.Ball, R. 1972. Changes in accounting techniques and stock prices. Journal of Accounting Research

10 (Supplement): 1–38.Balzer, W. K., M. E. Doherty, and R. O’Connor. 1989. Effects of cognitive feedback on performance.

Psychological Bulletin 106 (November): 410–433.Birnberg, J. G., and M. D. Shields. 1984. The role of attention and memory in accounting decisions.

Accounting, Organizations and Society 9 (3/4): 365–382.Bonner, S. E., and P. L. Walker. 1994. The effects of instruction and experience on the acquisition of

accounting knowledge. The Accounting Review 69 (1): 157–178.Brehmer, B. 1974. Hypotheses about relations between scaled variables in the learning of probabilistic

inference tasks. Organizational Behavior and Human Performance 11: 1–27.———. 1979. Preliminaries to a psychology of inference. Scandinavian Journal of Psychology 20:

193–210.Broniarczyk, S. M., and J. W. Alba. 1994. Theory vs. data in prediction and correlation tasks. Or-

ganizational Behavior and Human Decision Processes 57 (1): 117–139.Chan, L. K. C., J. Lakonishok, and T. Sougiannis. 1999. The stock market valuation of research and

development expenditures. Working paper, National Bureau of Economic Research.Chen, K. C. W., and M. P. Schoderbek. 2000. The 1993 tax rate increase and deferred tax adjustments:

A test of functional fixation. Journal of Accounting Research 38 (1): 23–44.Cook, R. L., and T. R. Stewart. 1975. A comparison of seven methods for obtaining subjective

descriptions of judgmental policy. Organizational Behavior and Human Decision Processes 13(1): 31–45.

Cooksey, R. W. 1996. Judgment Analysis: Theory, Methods, and Applications. San Diego, CA: Ac-ademic Press.

Diehl, E., and Sterman, J. D. 1995. Effects of feedback complexity on dynamic decision making.Organizational Behavior and Human Decision Processes 62 (2): 198–215.

Gupta, M., and R. R. King. 1997. An experimental investigation of the effect of cost information andfeedback on product cost decisions. Contemporary Accounting Research 14 (1): 99–127.

Hand, J. R. M. 1990. A test of the extended functional fixation hypothesis. The Accounting Review65 (4): 740–763.

Hirst, D. E., and P. E. Hopkins. 1998. Comprehensive income reporting and analysts’ valuation judg-ments. Journal of Accounting Research 36 (Supplement): 47–75.


Hopkins, P. E. 1996. The effect of financial statement classification of hybrid financial instrumentson financial analysts’ stock price judgments. Journal of Accounting Research 34 (Supplement):33–50.

———, R. W. Houston, and M. F. Peters. 2000. Purchase, pooling, and equity analysts’ valuationjudgments. The Accounting Review 75 (3): 257–281.

Hutchinson, J. W., and J. W. Alba. 1997. Heuristics and biases in the ‘‘eyeballing’’ of data: The effectsof context on intuitive correlation assessment. Journal of Experimental Psychology: Learning,Memory, and Cognition 23 (3): 591–621.

Ijiri, Y., R. K. Jaedicke, and K. E. Knight. 1966. The effects of accounting alternatives on managementdecisions. In Research in Accounting Measurement, edited by R. K. Jaedicke, Y. Ijiri, and O.Nielsen. New York, NY: American Accounting Association.

Ittner, C. D., and D. F. Larcker. 1998. Innovations in performance measurement: Trends and researchimplications. Journal of Management Accounting Research 10: 205–238.

———, V. Nagar, and M. Rajan. 2001. An empirical examination of dynamic quality-based learningmodels. Management Science 47 (4): 563–578.

Kaplan, R. S., and D. P. Norton. 1996a. Linking the balanced scorecard to strategy. California Man-agement Review 39 (1): 53–79.

———, and ———. 1996b. The Balanced Scorecard: Translating Strategy into Action. Boston, MA:Harvard Business School Press.

Klayman, J. 1988. On the how and why (not) of learning from outcomes. In Human Judgment: TheSJT View, edited by B. Brehmer, and C. R. B. Joyce. New York, NY: Elsevier Science Publishers.

Kothari, S. P. 2001. Capital markets research in accounting. Journal of Accounting and Economics(forthcoming).

Lee, J-W., and J. F. Yates. 1992. How quantity judgment changes as the number of cues increases:An analytic framework and review. Psychological Bulletin 112 (2): 363–377.

Libby, R. 1981. Accounting and Human Information Processing. Englewood Cliffs, NJ: Prentice Hall.———, R. Bloomfield, and M. W. Nelson. Experimental research in financial accounting. Accounting,

Organizations and Society (forthcoming).Lipe, M. G. 1998. Discussion of comprehensive income reporting and analysts’ valuation judgments.

Journal of Accounting Research 36 (Supplement): 77–83.Maines, L. A., and L. S. McDaniel. 2000. Effects of comprehensive-income characteristics on non-

professional investors’ judgments: The role of financial-statement presentation format. The Ac-counting Review 75 (2): 179–208.

McKinnon, S. M., and W. J. Bruns. 1992. The Information Mosaic. Boston, MA: Harvard BusinessSchool Press.

Muchinsky, P. M., and A. L. Dudycha. 1975. Human inference behavior in abstract and meaningfulenvironments. Organizational Behavior and Human Performance 13: 377–391.

Nisbett, R. E., and T. D. Wilson. 1977. Telling more than we know: Verbal reports on mental pro-cesses. Psychological Review 84: 231–259.

Ponemon, L., L. Carr, and J. Wendell. 1994. The validity of quality costs: An empirical study of millsin the paper and pulp industry. Advances in Management Accounting 3: 213–236.

Simon, H. A. 1990. Information technologies and organizations. The Accounting Review 65 (3): 658–667.

Sniezek, J. 1986. The role of variable labels in cue probability learning tasks. Organizational Behaviorand Human Decision Processes 38 (2): 141–161.

Sterman, J. D. 1989a. Misperceptions of feedback in dynamic decision making. Organizational Be-havior and Human Decision Processes 43 (3): 301–335.

———. 1989b. Modeling managerial behavior: Misperceptions of feedback in a dynamic decisionmaking experiment. Management Science 35 (3): 321–339.

Stewart III, G. B. 1991. The Quest for Value. New York, NY: Harper.Stewart, T. R., P. J. Roebber, and L. F. Bosart. 1997. The importance of the task in analyzing expert

judgment. Organizational Behavior and Human Decision Processes 69 (3): 205–219.


Sunder, S. 1999. Common knowledge and accounting. Plenary address, Accounting, Behavior andOrganizations Research Conference. Available at http://www.som.yale.edu/faculty/sunder/ABOPresentation.

Theil, H. 1966. Applied Economic Forecasting. Amsterdam, The Netherlands: North-Holland.Tucker, L. R. 1964. A suggested alternative formulation in the developments by Hursch, Hammond,

and Hursch and by Hammond, Hursch and Todd. Psychological Review 71: 528–530.Tully, S. 1993. The real key to creating wealth. Fortune (September 20): 38–50.———. 1998. America’s greatest wealth creators. Fortune (November 9): 193–204.Waller, W. S., B. Shapiro, and G. Sevcik. 1999. Do cost-based pricing biases persist in laboratory

markets? Accounting, Organizations and Society 24 (8): 717–739.Well, A. D., S. J. Boyce, R. K. Morris, M. Shinjo, and J. I. Chumbley. 1988. Prediction and judgment

as indicators of sensitivity to covariation of continuous variables. Memory and Cognition 16 (3):271–280.

Wilner, N., and J. Birnberg. 1986. Methodological problems in functional fixation research: Criticismand suggestions. Accounting, Organizations and Society 11(1): 71–82.

Copyright of Accounting Review is the property of American Accounting Association and its content may not

be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.

akuntansi perilaku

Documents

intangibles expenditures

accounting choice intangibles

future profits

effect of intangibles

accounting reviewvol

intangibles conference

financial accounting

basisfor learning