.n . x»... ‘ .~ x. a.. firm”. a «my: .8 9 . . A 3 xsh‘dth IAIJ 1-,. :l. 2 : fimfiaénax? 9, r ,... ....,...U. C , . '— LIBRARY 3 Michigan State ' L University This is to certify that the dissertation entitled OVERHEAD COST POOL CLASSIFICATION AND JUDGMENT PERFORMANCE presented by MATTHEW CHRISTIAN MASTILAK has been accepted towards fulfillment of the requirements for the DOCTORAL degree in ACCOUNTING 95.9% (ll/w (y f L/ Major Professor‘ ignature Swapé . (9" 6100 7 Date MSU is an aflinnative—action, equal-Opportunity employer .1—.-....—. _ —-I-.-n-t~c-a PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 6/07 p:/C|RC/DateDue.indd-p.1 OVERHEAD COST POOL CLASSIFICATION AND JUDGMENT PERFORMANCE By Matthew Christian Mastilak A DISSERTATION Submitted to Michigan State University in partial fulfillment Of the requirements for the degree Of DOCTOR OF PHILOSOPHY Department Of Accounting 2007 ABSTRACT OVERHEAD COST POOL CLASSIFICATION AND JUDGMENT PERFORMANCE By Matthew Christian Mastilak This dissertation addresses the research question: does an organization’s choice Of overhead cost pOOl classification affect individuals’ predictive judgments about overhead costs? I use psychology research tO develop hypotheses about how overhead cost pOOl classification affects the expected accuracy Of participants’ predictions of a target overhead cost based on their estimation Of the coefficients relating the target overhead cost to four predictor overhead costs. The predictor costs are located either within the same overhead pOOl as the target cost or in another pool. I vary relationships among the costs such that the larger predictor-target coefficient is either within a single cost pool or across different cost pools, and the relationships are positive or negative. I hypothesize that implicit coefficients on predictors within the same pool as the target cost will be estimated more accurately than implicit coefficients on predictors in a different pool. I also hypothesize that predictions of the target cost are affected by an interaction of the location of the stronger predictor and the sign Of the relationships. As hypothesized, participants estimate implicit coefficients within a pm] more accurately than coefficients across pools. Further, the location Of the stronger predictor and the sign Of the relations interact, such that the effect Of the location Of the stronger predictor on judgment accuracy is greater when relationships among costs are negative than when they are positive. I thus provide theory-consistent evidence Of a cognitive effect Of overhead cost pOOl classification in addition tO informational effects modeled by the analytic literature. Copyright by Matthew Christian Mastilak 2007 DEDICATION This dissertation is dedicated tO my wife Tina, who supported me emotionally and financially throughout my graduate work, and without whom I would never have been able tO finish this work, my daughter Elizabeth, who has made my life immeasurably more joyous, and any and all children with whom God blesses us in the future. It is for my family that I chose this path. iv ACKNOWLEDGMENTS I thank the Lord God Almighty, maker Of Heaven and Earth, from Whom and through Whom and to Whom are all things. TO Him be the glory forever! I thank His Son Jesus Christ, my merciful and precious Savior and Redeemer, in Whom I hope. I thank the Holy Spirit, the Counselor and Spirit of Truth, and the Guarantee of what is to come. I thank my committee chair, Dr. Joan Luft, for her interminable patience, refreshing wit, piercing insight and indispensable judgment. She was with this idea from its inception; her gracious help made this work possible. I also thank my committee members, Dr. Mike Shields, Dr. Cheri Speier and Dr. Rick Deshon. Their assistance made this work much better, and I am grateful for their help. I thank wife Tina and daughter Elizabeth and my parents, Sandy and Ken Mastilak, for always loving and supporting me and making my success possible. I also thank my in-laws, Birdie and Ron Black, for being wonderful to me always, and for providing a home away from home for Tina and Elizabeth when it was needed. I thank my fellow doctoral students at Michigan State, especially my classmates Fabienne Miller and Alex Woods, and my friends Freddy Coronado, Roger Stace, Chang Song and Han Yi, for their encouragement, support, friendship, practical help in innumerable ways, and overall gOOd company. My time at MSU would have been much poorer without them. I also thank the faculty in the Department Of Accounting and Information Systems at Michigan State University, particularly Dr. K. Ramesh. Their willingness to provide all manner Of help, encouragement and support is much appreciated. I also thank the staff in the Department, especially Deb North and Patty Geller, for their friendship and help. TABLE OF CONTENTS LIST OF TABLES ....................................................................................................... viii LIST OF FIGURES ....................................................................................................... ix 1. Introduction ................................................................................................................. 1 11. Literature review and Hypothesis Development ........................................................ 5 Overhead cost pool classification ............................................................................... 5 Example of a prediction task ....................................................................................... 7 Variation in overhead cost pool classification ............................................................ 8 Overhead cost pool classification and beliefs about costs ........................................ 15 Beliefs about costs, attention allocation and judgment performance ........................ 17 Effect of cost behavior on judgment performance .................................................... 19 Hypothesis development ........................................................................................... 22 Hypotheses l and 2: Coefficient estimation errors ................................................... 24 Hypothesis 3: Prediction accuracy of overhead costs ............................................... 27 III. Research design ...................................................................................................... 32 Experimental design .................................................................................................. 32 Participants ................................................................................................................ 34 Procedure and task .................................................................................................... 34 Measurement of variables ......................................................................................... 37 IV. Results .................................................................................................................... 4O Hypotheses 1 and 2 ................................................................................................... 42 Hypothesis 3 .............................................................................................................. 45 V. Discussion and conclusion ....................................................................................... 59 Appendix ....................................................................................................................... 65 vi References ................................................................................................................... 106 vii LIST OF TABLES Table 1: Experimental design ........................................................................................... 33 Table 2: Experimental parameters .................................................................................... 37 Table 3: Manipulation checks ........................................................................................... 41 Table 4: Hypotheses 1 and 2 ............................................................................................. 43 Table 5: Sensitivity analysis: across-pool minus within-pool implicit coefficient errors by cost behavior condition: Each cost behavior condition contrasted with the mean of the other three cost behavior conditions ................................................................................................................... 46 Table 6: Hypothesis 3 ....................................................................................................... 47 Table 7: Supplementary analysis ...................................................................................... 52 viii LIST OF FIGURES Figure 1: Two-stage overhead cost allocation .................................................................... 6 Figure 2: Overhead cost pool classifications .................................................................... 13 Figure 3: Conceptual depiction of lens model method ..................................................... 21 Figure 4: Overhead cost pool classification and cost behavior ......................................... 23 Figure 5: Graphical presentation of Hypothesis 3 ............................................................ 30 ix I. INTRODUCTION A fundamental function of accounting is the combinatiOn of individual accounts into groups Of accounts. For example, external financial reporting and tax return preparation involve the reporting Of hundreds or thousands Of individual accounts in several dozen lines on financial statements or tax returns. Similarly, organizations classify individual overhead cost accounts into overhead cost pools. In this dissertation I examine how different overhead cost pool classification decisions affect judgment performance of individuals who use accounting reports. Throughout this dissertation, I use the label “overhead cost pool classification” to describe the assignment Of individual overhead cost accounts tO specific overhead cost pools. Pooled overhead costs are the numerator Of the ratio used to allocate overhead costs tO cost Objects such as products, services or customers. Prior literature on overhead cost pools has focused mainly on the denominator Of the cost allocation rate calculation (the allocation base or cost driver), with little consideration Of questions about the numerator alone. The literature has focused mainly on how cost driver choices and the degree Of aggregation Of costs into cost pools affect errors in cost allocations and whether non-volume cost drivers exist (Noreen 1991; Datar and Gupta 1994; Anderson 1995; Christensen and Demski 1995, I997; Ittner et a]. 1997). The literature has not fully investigated either the causes or the effects Of organizations’ decisions about cost pool classification; thus, there is a lack Of evidence on how classifications might affect individuals’ judgment. Individuals within organizations Often require information to support decisions about individual cost accounts within cost pools, and thus research on effects Of overhead cost pool classification on such decisions is valuable. I partially address this gap in the literature by asking the research question: how dOes an organization’s choice of overhead cost pool classification affect individuals’ predictive judgments about overhead costs? In this dissertation I examine a setting in which individuals receive reports about both overhead cost pool totals and individual overhead cost accounts within the pools. That is, the individual overhead costs are reported within their overhead cost pools. Individuals use these reports in a budgeting task that requires the prediction of the variable portion of an overhead cost (the “target cost”). Consistent with empirical evidence on managerial decisions (e.g., Datar et al. 1993), I focus on judgments about relationships among individual overhead costs within and across overhead pools, rather than judgments about relationships between overhead costs and cost allocation bases (i.e., cost drivers). I use psychology research tO develop hypotheses about how overhead cost pOOl classification affects the accuracy of participants’ predictions of a target overhead cost based on other overhead costs that may be predictive Of the target (the “predictors”). I use an experimental research design in which participants make predictions about the target based on information about the predictors. The predictor costs are located either within the same overhead pOOl as the target cost or in another pool. Participants’ predictions of the target cost are based on their estimate Of the implicit coefficients relating the target to the various predictors.I I vary the relationships among costs in two ways. First, the location Of the stronger predictor—target relationship is varied such that it is either within a single cost pool or across different cost pools. Second, the relationships 1 model partrcrpants judgments usrng regressron and generate estimates of the Intercept and rmplrcrt coefficients in their prediction models. Participants do not directly estimate the coefficients relating the target cost to the predictors. among costs are either positive or negative. These manipulations allow me to examine the effects on judgment of different overhead cost pool classification decisions in settings with differing cost behaviors (e. g., costs that are substitutes vs. complements).2 Based on psychology research, I predict that overhead cost pool classification affects judgment performance by directing individuals’ attention toward within—pool relationships and away from across-pool relationships. Thus, I hypothesize that individuals will estimate within-pool implicit coefficients more accurately than across- pOOl implicit coefficients. I further expect that, consistent with evidence from the psychology literature, relationships to which less attention is allocated are underestimated rather than randomly misestimated (e. g., Broniarczyk and Alba 1994). Thus, individuals will tend to underestimate across-pool predictors more than within-pool predictors. I also predict that accuracy Of individuals’ predictions Of the target overhead cost is an interactive function of the sign Of the relationships among costs and the location Of the stronger predictor, such that the effect of the location of the stronger predictor on judgment accuracy is greater when relationships among costs are negative than when they are positive. The experimental results are generally consistent with hypotheses. Participants’ predictions and their implicit coefficients exhibit patterns consistent with overhead cost pool classification affecting their judgments about costs. First, participants’ estimates of the implicit coefficients relating predictor and target overhead costs are more accurate when predictor and target are within the same pool than when they are across pools. In this dissertation, I use the term cost behavror to describe whether the variable portron of target and predictor overhead costs are positively or negatively correlated with each other and whether the relation between target and predictor is weak or strong. Second, participants underweight across-pool predictors more than within-pool predictors. Further, the sign and strength of the relationships among predictors and the target interact to affect the accuracy Of participants’ predictions Of the target. The effect Of overhead cost pool classification on judgment accuracy is greater when relationships among costs are negative than when they are positive. Finally, supplemental analysis provides evidence that overhead cost pool classification affects the allocation of attention across different predictors, and the allocation of attention affects prediction accuracy. I thus provide theory-consistent evidence of a cognitive effect of overhead cost pool classification in addition to informational effects modeled by the analytic literature. This study considers subjective judgments about overhead costs. Subjective judgment is used often in organizations (MacKinnon and Bruns 1992; Kaplan and Norton 1996; Ittner et al. 2003), and thus it is an important topic for research. Even a formal statistical analysis such as a multiple regression analysis will only increase judgment accuracy if the appropriate relationships are examined (i.e., if appropriate variables are included in the regression). When performing formal statistical analyses, individuals Often use subjective judgment to choose relationships to examine and the functional form of those relationships. My results imply that overhead cost pool classification can affect iwhich relationships are examined as well as how accurately they are estimated. Thus, overhead cost pool classification may influence formal statistical analyses Of overhead costs as well as subjective judgments. The rest of this dissertation is organized as follows. Chapter 2 presents the literature review and hypothesis development. Chapter 3 describes the research design. Chapter 4 presents results, and Chapter 5 concludes. II. LITERATURE REVIEW AND HY POTHESIS DEVELOPMENT First, I will introduce accounting literature on overhead cost pool classification. Then, I will provide an example of a judgment task that is affected by organizations’ overhead cost pool classification decisions. Next, I explain why these overhead cost pool classification decisions vary across organizations. Then, I use findings from psychology literature to describe (1) how variation in overhead cost pool classification will affect individuals’ beliefs about costs, (2) how those beliefs will affect individuals’ allocation Of their attention when they perform the task, and (3) how attention allocation affects judgment performance in the task. Further, I make predictions about how different cost behaviors will affect judgment performance in the task. Finally, I state my hypotheses. Overhead cost pool classification Organizations’ decisions regarding the allocation of overhead costs to cost Objects can be modeled as a two-stage process, as shown in Figure 1 (Cooper and Kaplan 1999). Prior academic literature on overhead cost pool classification has focused primarin on the second stage. Analytical literature has considered the effect of interactions between organizations’ cost functions and overhead cost pool classification on the cost allocation error reported by various cost allocation systems (e. g., Noreen 1991; Datar and Gupta 1994; Christensen and Demski 1995, 1997; Bromwich and Hong 1999). Empirical literature has typically examined the validity of claims that important cost drivers other than production volume exist (e.g., Foster and Gupta 1990; Banker et al. 1995; Anderson 1995; Ittner et al. 1997). Experimental literature has examined the effects of cost system accuracy, economic complexity and different types Of feedback on cost forecasting and product pricing/output decisions (Gupta and King 1997; Briers et al. 1999), and the interaction of cost system and incentives on innovation, efficiency and profitability in teams (Drake et a1. 1999). To my knowledge, there is no literature on the effects of overhead cost pool classification on individuals’ judgment performance about individual overhead costs. Figure l: Two-stage overhead cost allocation Individual OH cost accounts First stage: Assign costs to cost pools OH Cost pools g Second stage: \ / Assign costs to cost Objects —-> . via bases Cost Objects - This figure depicts the two-stage process for assigning overhead cost accounts from the chart of accounts to cost objects. In the first stage, overhead costs are assigned to overhead cost pools. In the second stage, overhead costs from the overhead cost pools are assigned to cost Objects via allocation bases. The potential for an effect of overhead cost pool classification on individuals’ judgment performance is an important topic for research. There is evidence that overhead costs can be related to costs in other overhead cost pools in statistically and economically significant ways. For example, Anderson (1995) documents strong negative correlations among setup costs, accounted for in a setup cost pool, and power costs, accounted for in an administrative cost pool. Ittner et al. (1997) document significant correlations among costs in different pools and at different levels of the activity-based costing hierarchy. There is also evidence that relationships among costs in different cost pools are difficult for individuals to estimate. Joshi et al. (2001) report that managers grossly underestimate the magnitude of environmental regulation compliance costs that are “hidden” in non- regulatory cost pools. Example of a prediction task Consider the following example of a task an individual in an organization might face. As part of an organization’s operating budgeting process, the individual must predict the next month’s expected spoilage costs for each of several similar manufacturing plants. Unlike such costs as direct labor, spoilage costs are not the direct result of a spending decision, but rather an indirect result of a variety Of other spending and operating decisions. The individual’s task is to predict spoilage cost from levels of spending on Other overhead costs based on relationships among overhead costs observed in, and learned from, a prior month’s report. Specifically, the individual uses spending on inspection and supervision as predictors of spoilage costs. Assuming a linear, additive relationship between spoilage, inspection and supervision, the model is Spoilage = Constant + ,BIInspection + ,BzSupervision + error. To successfully predict spoilage, the individual must estimate the constant (i.e., the fixed cost component of spoilage cost) and B] and [32, the coefficients on inspection and supervision (i.e., the components of spoilage cost that vary with inspection and supervision). The sign and magnitudes of the coefficients will depend on the organization’s cost function.3 That is, [31 and [32 may be positive or negative (or zero), and [31 may be larger or smaller than [32. Further, the organization’s overhead cost pool classification will affect whether inspection and/or supervision costs are in the same pool as spoilage costs. The next section discusses reasons why there can be variation in organizations’ overhead cost pool classifications. Variation in overhead cost pool classification Relationships among overhead costs within and across cost pools will vary across organizations because organizations have different purposes for classifying overhead costs into pools. Two possible purposes are considered here: product costing and responsibility accounting.4 Product costing One important purpose for classifying overhead costs into pools is the allocation of costs to products or services for such purposes as pricing, deciding whether to drop a product from the product line, determining optimal input mixes, and designing products to minimize cost (hereafter “product costing”). If product costing is the primary purpose of an organization’s cost accounting system, then the organization will likely choose to assign to the same cost pools those overhead costs that are strongly positively correlated with each other (Garrison et al. 2006; Horngren et al. 2002; Kaplan and Cooper 1996). Overhead cost pool classification decisions that result in pools containing costs with 3 The magnitude of the constant (the fixed portion of the overhead cost) will also depend on the cost function. However, in this dissertation I focus only on coefficients (the variable portion of the overhead cost). 4 . . . . Other reasons exrst, such as inventory valuation for external reporting purposes, that may also affect overhead cost pool classification decisions (Kaplan and Cooper 1999). stronger positive within-pool correlations will minimize product costing errors (Noreen 1991; Christensen and Demski 1995; Bromwich and Hong 1999). Noreen (1991, 163), discussing the requirements for a costing system to provide costs that are appropriate for product-drop and product-design decisions, writes, [A] well-specified ABC system exists in which product costs are avoidable costs and activity costs are incremental costs if and only if: 1) the underlying cost function. . .can be partitioned into cost pools, each of which depends only upon a single activity; 2) the cost in each pool is strictly proportional to its activity, and 3) each activity can be divided among products in such a way that the portion attributed to each product depends only upon that product.” If costs within a pool are all proportional to a single activity, then it follows that the costs will be highly correlated with each other. Bromwich and Hong (1999) extend Noreen’s (1991) analysis; consistent with Noreen (1991) they report that for a pooled accounting system to provide incremental costs for pricing, product portfolio and make-or-buy decisions, the technology must be locally homothetic and the overhead cost pool classification decisions must reflect that homotheticity: A necessary condition for the construction of a cost pool is therefore that there must exist a specific aggregate input (in the accounting context, an aggregate cost driver) reflecting a constant mix of elementary inputs for the cost pool irrespective of volume which implies local homotheticity of cost pool technology. Thus, local homotheticity allows a single cost driver to be used for a cost pool irrespective of volume. (50) Again, the existence of a cost driver as a “specific aggregate input” and a “constant mix of elementary inputs” mean that for a cost driver to summarize the costs within a pool, the costs within that pool must be correlated with that activity and with each other. Similarly, Christensen and Demski (1995) model a firm’s cost-minimizing input- mix decision and consider the effect of cost pools on errors in costs reported by a cost system. They report, In particular, we are accustomed to grouping factors in cost pools. Classically, this makes sense of the factor quantities respond in like fashion to the underlying technology and economic forces. Intuitively, this means we use the factors in a pool in roughly the same manner. Interactions with other factors are common across factors in the pool. (p. 16; emphasis added) Thus, in comparing the “classical” model of cost to the “modem” or ABC model, Christensen and Demski (1995) state that in order for the “modern” model to report costs within pools without error, the use of inputs (and the resulting costs of those inputs) within a pool must be proportional to each other (that is, highly correlated). In addition to the proper specification of inputs and costs within pools, Christensen and Demski (1995) also identify a requirement of the relationships between different pools. “Cross pool elasticities must be zero, however, if the factor usage in a particular pool is to be independent of prices for factors outside the pool” (17). In the classical model, input prices determine input quantities for a specified technology and output mix. Thus, the requirement of zero cross-pool elasticities means the effect of a change in prices Of factors in one pool must not affect the use of factors in other pools. In effect, then, costs in one pool must be uncorrelated with costs in other pools for a pooled cost accounting system to report costs appropriate for the input-mix decision. In summary, if product costing is the primary determinant of overhead cost pool classification decisions, it is likely that costs within pools will be strongly correlated, and costs within a pool will be weakly correlated or uncorrelated with those in other pools. Responsibility accounting Another important purpose for classifying overhead costs into pools is responsibility accounting (that is, a system under which costs for which an individual or responsibility center is responsible are pooled to provide a summary statistic for that individual’s or responsibility center’s performance). If responsibility accounting is a 10 primary purpose for the cost system, within-pool relationships can be strong or weak. Responsibility accounting purposes for cost system design may result in weakly correlated overhead costs being classified into a single pool. This may occur when a single individual or responsibility center has responsibility for weakly correlated costs, and when measurement error creates difficulty in dividing those costs into different pools (e.g., Datar and Gupta 1994). For example, an employee may spend time on three different activities, and it is difficult to measure the time spent on each of two of those activities. The portions of that employee’s compensation related to time spent on those two activities might then be recorded within a single pool even if they are weakly correlated, to prevent negative consequences that might result from allocating the time more finely but with error. Note also that in this example, there will likely be negative correlations among hours the employee spends on various activities. (Given a fixed amount of time spent in total, more hours spent on one activity results in fewer hours for other activities.) However, responsibility accounting does not necessarily create negative correlations among costs recorded within a pool. On the other hand, overhead costs in one pool may be strongly correlated with overhead costs in a different pool, because different individuals or responsibility centers are responsible for the two costs, and accounting for them in separate pools facilitates calculations such as variances and responsibility center-level reports (Garrison et al. 2006; Horngren et al. 2002; Cooper and Kaplan 1999). Thus, in an organization whose overhead cost pool classification decisions are based primarily on responsibility accounting, it is possible (but not necessary) that weakly correlated overhead costs will I] be classified into a single overhead cost pool and strongly correlated costs may be classified into different overhead cost pools. Example The example of the prediction of spoilage costs from inspection and supervision costs can be used to illustrate overhead cost pool classification by product costing vs. responsibility accounting purposes. There are several overhead cost pool classification decisions the organization could make. Assume that inspection and supervision costs are uncorrelated and are in separate responsibility centers, so the organization has decided to account for these costs in separate pools. Further, assume the organization has decided to assign the three costs to two pools, and so must now decide whether to assign spoilage to the same pool as inspection or the same pool as supervision. Also assume that spoilage and inspection have a stronger relationship ([31 = 1.01) and spoilage and supervision have a weaker relationship ([32 = 0.46).5 Finally, assume that responsibility for and some influence over both spending on supervisors and reducing spoilage exist within a common responsibility center. Figure 2 presents the example graphically. For simplicity, differences between larger and smaller coefficients are assumed to be statistically significant. Further, costs are assumed to have similar scales, so larger coefficients imply stronger correlations and more predictive validity. 12 Figure 2: Overhead cost pool classifications Panel A: Cost pools based on cost allocation Pool A Pool B Inspection Spoilage Supervision A A AA A A Panel B: Cost pools based on responsibility accounting Pool A Pool B Inspection Spoilage Supervision A A A A 3:101 B=OA6 This figure depicts two potential overhead cost pool classification decisions. In Panel A, costs are classified such that the stronger relationships (inspection and spoilage) is within a pool, and the weaker relationship (supervision and spoilage) is across pools. This classification might result if the organization decided that allocation of costs to cost objects is the most important use of the cost system. In Panel B, costs are classified such that the stronger relationship (inspection and spoilage) is across cost pools, and the weaker relationship (supervision and spoilage) is within a pool. This classification might result if the organization decided that responsibility accounting use of the cost system is most important, and spoilage and supervision are the responsibility of the same manager. 13 Given the assumptions and the discussion above, there are two possible overhead cost pool classification decisions the organization can make. If allocating spoilage, inspection and supervision costs to cost Objects is of primary importance, then the organization would likely include spoilage in the same cost pool as inspection. Figure 2, Panel A depicts this overhead cost pool classification. Spoilage has a stronger relationship with inspection than does supervision; as noted above, assigning spoilage and inspection to the same pool will allow for a better choice of allocation base than will assigning spoilage and supervision to the same pool. Note that in this overhead cost pool classification, there is a stronger relationship within a cost pool (spoilage and inspection) and a weaker relationship across cost pools (spoilage and supervision). Alternatively, if responsibility accounting is of primary importance, then the organization would likely assign spoilage and supervision to the same cost pool, to allow the cost system to support clear responsibility accounting. Figure 2, Panel B depicts this overhead cost pool classification. Note that in this overhead cost pool classification, there is a weaker relationship within a cost pool (spoilage and supervision) and a stronger relationship across cost pools (spoilage and inspection). In summary, there is likely variation in the overhead cost pool classification decisions made by organizations, based on different purposes for the overhead cost pool classifications. Whether and how these overhead cost pool classification decisions matter depend on how overhead cost pool classification decisions affect individuals’ judgment performance. The following subsection reviews the accounting and psychology literatures and develops hypotheses about how overhead cost pool classification decisions are expected to affect judgment performance. 14 Overhead cost pool classification and beliefs about costs The classification of costs into cost pools results in individual cost accounts being grouped together and given a common label (such as “quality control costs”). The grouping of individual items under a label has been investigated by the psychology literature on categorization (e. g., Markman and Gentner 2001 ). Categorization has been found to be a fundamental part of human information processing (Murphy and Medin 1985). The categories in which individual items reside, and the labels on those categories, have been found to influence the inferences individuals make about those individual items. Specifically, individuals infer that items given a category label are similar to other items given the same category label (Yamauchi and Markman 2000; Markman and Gentner 2001). Thus, when individuals make inferences about relationships among costs classified together in cost pools, they are likely to infer that the costs are similar to each other. Conversely, items categorized into different categories are assumed to be dissimilar (Yamauchi and Markman 2000; Markman and Gentner 2001). Thus, when individuals make inferences about relationships among costs classified into different pools, they will likely assume that costs classified into different cost pools are dissimilar. Discussions Of cost pool classification in accounting textbooks and literature typically focus on the correlations among costs within pools. Correlation among costs is the feature on which overhead cost pool classification is based (e.g., Christensen and Demski 1995; Kaplan and Cooper 1998; Horngren et al. 2002).6 Horngren et al. (2002, p. 142) write: 6 These examples refer to activity-based costing systems, and much of the discussion regarding overhead cost pool classification has been in the context of ABC systems. However, the choice of activities as the driver or allocation base is not required for the principle of overhead cost pool classification based on correlation to hold. 15 Each of the activity-related cost pools [is] homogeneous. Why? Because each activity cost pool includes only a narrow and focused set of costs (for example, setup or distribution). Over time, the costs in each activity-cost pool have a cause-and-effect relationship with the cost-allocation base for that activity (for example, setup-hours in the case of setup costs and cubic feet of packages moved in the case of distribution costs). Similarly, Kaplan and Cooper (1998, p. 86) cite an internal training manual saying, “A resource comprises a distinct and homogeneous grouping of existing costs fulfilling a similar function...” (emphasis added). Also, Roth and Borthick (1991, p. 39) write that the assumption of homogeneity in activity—based costing models “means that the costs in each pool are driven by a single activity or by highly correlated activities. Highly correlated means that changes in the level of one activity are accompanied by proportional changes in the other activities” (emphasis in original). These passages strongly imply that because costs within each pool move with a single activity driver, they move together with each other. Further, because each pool’s costs are “narrow and focused” or, alternatively, “distinct and homogeneous,” costs that are not highly correlated are classified into separate pools. Thus, correlation among costs is an important feature Of individual costs that individuals will likely infer is similar among costs classified within a cost pool, and dissimilar among costs classified into different cost pools. Thus, in the absence of strong evidence to the contrary, students and practitioners of accounting are likely to believe that overhead costs within a pool (across pools) are positively and relatively strongly correlated (not correlated). Despite the assumption in the literature about strong relationships being classified within pools, organizations’ overhead cost pool classification decisions vary, as described above. Therefore, individuals within organizations will sometimes use accounting information in which strong correlations are across—pool rather than within-pool, for 16 example, when overhead cost pool classification is based on responsibility accounting. In these circumstances, individuals will have knowledge of some, but not all, of the relationships among costs. Individuals have limited cognitive resources (i.e., they are boundedly rational), and they will likely estimate accurately and retrieve from memory only the more important relationships or those they encounter more frequently (Anderson 1990). Individuals may not have learned about certain relationships because those relationships may have been relatively unimportant in the past, or because information about those relationships was not available. Moreover, relationships that were and remain important may change in magnitude over time. For those relationships for which individuals do not have current knowledge, individuals may rely on overhead cost pool classification as a source of beliefs about relationships among overhead costs. The accuracy of individuals’ estimates of relationships among overhead costs may be affected by organizations’ overhead cost pOOl classification decisions and the relationships among overhead costs. Individuals’ implicit estimations of relationships among overhead costs, and their resulting predictions of the target overhead cost, will be affected by both the allocation of attention among potential relationships and the cognitive difficulty of estimating those relationships that receive attention. The following analysis describes how overhead cost pool classification and cost behavior can affect attention allocation and estimation accuracy. Beliefs about costs, attention allocation and judgment performance Beliefs about relationships in data can affect the allocation of individuals’ attention among the potential relationships (Brehmer 1974; Broniarczyk and Alba 1994; Luft and Shields 2001). Research in accounting and psychology has documented the 17 pervasive effects of individuals’ beliefs on how they cognitively examine data and their resulting judgments in a variety of task settings, such as credit scoring, GPA and temperature prediction, consumer judgments of a price-quality relationship, and profit prediction (Muchinsky and Dudycha 1974, 1975; Sniezek 1986; Broniarczyk and Alba 1994; Luft and Shields 2001). Psychology research indicates that individuals examine potential relationships between a target and multiple predictors one predictor at a time, rather than simultaneously (Brehmer 1974). The order in which individuals examine relationships is influenced by their beliefs. Individuals examine potential relationships they believe to be more important earlier than relationships they believe to be less important (Brehmer 1974). Relationships examined earlier are estimated more accurately because individuals’ limited cognitive capacity leads to fatigue, reduced cognitive effort and inconsistency in estimating relationships examined later (Brehmer 1974; Luft and Shields 2001). Thus, if individuals believe that overhead costs within (across) pools are likely (unlikely) to be related to each other, then overhead cost pool classification is expected to affect judgment accuracy. However, if individuals hold strong beliefs about relationships, but those relationships are not present in the data, those beliefs can overwhelm inspection of the data, individuals’ judgments can be biased in the direction of their beliefs (Broniarczyk and Alba 1994). Evidence from Joshi et al. (2001) is consistent with overhead cost pool classification affecting judgment performance among experienced managers. The authors found that large portions of the costs steel firms incur in complying with environmental regulations are accounted for in non-regulatory cost pools. These costs are misclassified because they are not easily identifiable as regulatory costs. For example, if a mill 18 switches from a cheaper, high-polluting coal to a more-expensive, cleaner-burning coal, the entire cost of the coal is counted as a raw material cost, even though the difference in coal prices is due to regulation and would not be incurred save for the regulation. Managers at the firms studied are aware of the existence of across-pool relationships and the fact that some regulatory cost burden is “hidden.” Further, managers consider environmental regulation costs to be an important factor in their firms’ ability to compete. The managers thus have both the opportunity and the motivation to understand the magnitude of the relationships. However, the managers grossly underestimated the magnitude of regulatory costs — they estimated that about half the cost was hidden, while the authors estimate thatup to 90% of the cost is hidden. In interviews, the managers said they believed their underestimation of the regulatory costs affected decisions such as performance evaluation and plant closure or continuance decisions. However, whether the overhead cost pool classification is the cause or the effect of managers’ beliefs about the magnitude of “hidden” regulatory costs is uncertain.7 Effect of cost behavior on judgment performance The sign and strengths of relationships can affect the accuracy of individuals’ estimations of the relationships independently of attention—allocation effects (Brehmer 1974; Sniezek 1986; Farrell et al. 2006). Even when individuals have correct beliefs about the sign of relationships, positive relationships are estimated more accurately than are negative relationships, indicating an inherent difference in the cognitive difficulty of estimating positive vs. negative relationships, all else equal. Thus, when individuals 7 That is, managers may have had beliefs that the amount of potentially hidden regulatory cost was small, and designed their regulatory cost pools based on those beliefs. The design of Joshi et al. (2001) does not support inferences about causality between overhead cost pool classification and managers’ beliefs. l9 attempt to revise beliefs about relationships among overhead costs, they are likely to do so more accurately when the costs are positively correlated than when they are negatively correlated. Further, psychology research indicates that relatively weak relationships (i.e., those with correlations below approximately 0.6) are generally difficult for individuals to estimate subjectively, while stronger relationships (i.e., those with correlations above approximately 0.6) are generally estimated more accurately (e.g., Broniarczyk and Alba 1994). Thus, the sign and strength of a relationship affect the accuracy with which it is estimated. There is also evidence that beliefs about relationships and the sign of relationships interact to affect judgment performance. Sniezek (1986, experiment 3) found that the effect of incorrect beliefs on judgment performance was greater when the relationships in the data were negative than when they were positive. Consistent with psychology and accounting research, I use a lens model to analyze how and how well individuals learn and predict relationships from data.8 A conceptual depiction of the lens model is shown in Figure 3. The predictors in the center are the predictors individuals can use to predict the target. The environmental or true model is depicted on the left side of the figure; this model is derived from the data (Ye and Xi) that are presented to individuals and that they could use to estimate the environmental coefficients ([39,, where ei refers to the environmental coefficient on predictor i). The subjective model on the right side of the figure represents the model of individuals’ subjective predictions of the target (Y5) based on levels of the predictors. For a thorough and recent treatment of the lens model in accounting research, see Luft and Shields (2001). 20 Lens model studies regress individuals’ subjective predictions of the target on the predictors and estimate each participant’s implicit coefficients (BM-j, where sij refers to participant j’s implicit coefficient on predictor 1'). Figure 3: Conceptual depiction of lens model method Target Predictors Target X1 e2 X2 \K BS3" X 3 / [3541' X4 / L j k / Y Environmental model Subjective model Ys o/;\=P 5b») .— 4 This figure presents a conceptual depiction of the lens model method. The predictors in the center are the four predictors individuals can use to predict the target. The environmental or true model is depicted on the left side of the figure; this model is derived from the data that is presented to individuals and that they could use to estimate the environmental coefficients ([36,, where ei refers to the environmental coefficient on predictor i). The subjective model on the right side of the figure represents the model of individuals’ subjective predictions of the target based on levels of the predictors. Lens model studies regress individuals’ predictions of the target on the predictors and estimate each participant’s implicit coefficients (Bsij, where sij refers to participant j’s implicit coefficient on predictor i). 2] Hypothesis development To aid the exposition of the hypotheses, I first briefly describe the experimental task. I examine an operating budgeting task in which participants predict the amount of the next month’s expected spoilage costs in each of 20 similar manufacturing plants. The prediction is based on relationships among overhead costs that participants learn by examining the prior month’s cost reports. The amount of reported spoilage costs is affected by two predictors (the “non-zero predictors”).9 Two additional predictors are not actually predictive of spoilage (the “zero predictors”). The five overhead costs are assigned to two pools. Each pool contains one cost that is predictive of spoilage (a non- zero predictor) and one cost that is not predictive of spoilage (a zero predictor). To examine whether the effect of overhead cost pool classification varies with variable cost behavior, I use two manipulations of cost behavior. First, I vary the pool location of the stronger predictor — the predictor with the larger coefficient is either within the same pool as spoilage or in the other pool (that is, across pools). Second, I vary the sign of the nonzero coefficients on the predictors — they are either positive or negative. Figure 4 depicts the overhead cost pool classification and cost behavior conceptually. Panel A presents the condition in which the stronger predictor relationship is within the same pool as spoilage; Panel B presents the condition in which the stronger predictor relationship is across pools. 9 Throughout this dissertation, I use the label “predictors” to indicate the costs that participants could use to predict the target cost regardless of whether they are statistically significant predictors. The inclusion of zero-coefficient predictors in a setting in which participants are likely to expect positive correlations allows for an empirical test of this departure from participants’ likely expectations, and allows for tests to determine the effect of overhead cost pool classification on participants’ estimation of coefficients. 22 Figure 4: Overhead cost pool classification and cost behavior Panel A: stronger relationship is within-pool Pool A Pool B Zero Stronger Tar artge Weaker Zero A 1 TA AA A A IE: ll.01| l j} = I0.46| B = 0.00 B = 0.00 Panel B: stronger relationship is across-pool Pool A Pool B Zero Stronger Target Weaker Zero A T TA AA A A I j3=|l.01| I [3: |0.46| p = 0.00 l3 = 0-00 This figure presents a conceptual depiction of the design of the experimental task. “Target” describes the target cost that the individual is to predict. “Stronger,” “weaker” and “zero” describe the relationships between the predictors and the target. The B coefficients on the arrows specify the strength of the relationships between the targets and the predictors. The boxes represent overhead cost pool classification boundaries. Note that in this example, all costs have similar scales, so larger coefficients imply stronger correlations and more predictive power. All differences between coefficients are statistically significant at p<0.05. Participants are told the expected signs but not the magnitudes of the relationships between spoilage and the predictors. To perform well on the task, participants must use the cost reports to estimate the relationships between spoilage and the other overhead 23 costs.10 They must then apply those estimates of the relationships to the next month’s budget of the four overhead costs and predict spoilage based on budgeted levels of the four other costs. An example of the experimental materials is shown in Appendix A. Hypotheses l and 2 make predictions about errors in participants’ implicit coefficients on individual predictors. Hypothesis 3 predicts how variation in overhead cost pool classification and cost behavior affect the overall prediction accuracy of overhead costs. Hypotheses l and 2: Coefficient estimation errors As described earlier, individuals generally examine potential relationships sequentially,‘with relationships believed to be more important examined earlier and more accurately than those believed to be less important (Brehmer I974; Luft and Shields 2001). Consistent with the accounting literature reviewed above, I expect stronger (weaker) within-pool and weaker (stronger) across-pool relationships to be consistent (inconsistent) with individuals’ beliefs about the relationships among overhead costs assigned to cost pools when the individuals have no other strong sources of belief about the relationships among the costs. Because individuals likely believe that overhead costs classified within the same pool are strongly related and overhead costs classified across different pools are not strongly related, they will devote more attention to within-pool relationships and estimate them more accurately than across-pool relationships. I expect this effect even though in my research design, all participants examine statistically identical data sets, with detailed cost information available in all conditions, and my 10 . . . . . . . . . . . Partrcrpants must also rmplrcrtly estimate the constant In addition to the coefficrents on the predictors. However, large errors in estimating the constant are unlikely, as they would result in predicted levels of spoilage that are outside the range of those in the prior month’s data. 24 research design includes relatively few potential predictors and offers a strong and direct financial incentive for optimal performance. Beliefs can affect judgment accuracy in at least two ways. First, beliefs can affect the bias of individuals’ judgments (Sniezek 1986; Billman et al. 1992; Broniarczyk and Alba I994; Luft and Shields 2001). Because, in the absence of specific knowledge to the contrary, individuals likely believe that within-pool (across-pool) relationships exist or are more important (do not exist or are less important), they will exhibit a pattern of error biases consistent with these beliefs. For the non-zero predictors’ coefficients, individuals will pay more attention to within-pool coefficients than to across-pool coefficients. Thus, individuals will estimate within—pool coefficients with relatively little bias. Because individuals will pay less attention to the across-pool coefficients, they tend to treat the across-pool coefficients as closer to zero than the data indicate. That is, individuals will underestimate across-pool non-zero predictors’ coefficients more than within-pool non- zero predictors’ coefficients. For the zero predictors’ coefficients, individuals will likely believe that the within-pool coefficients are non-zero. Broniarczyk and Alba (1994) present evidence that in the presence of strong beliefs about a relationship, individuals will make judgments as though such a relationship exists when in fact there is no relationship in the data. Thus, if individuals strongly believe that costs within pools are correlated, these beliefs may overwhelm the data, and the individuals may overestimate within-pool zero predictors’ coefficients. However, individuals might not overestimate the within-pool coefficients for several reasons. First, individuals will pay significant attention to the within-pool relationships, and therefore individuals will estimate the within-pool relationships 25 accurately. Second, individuals have financial incentives to estimate coefficients accurately. And third, statistical relationships among the data in my setting are strong, which will improve the accuracy of individuals’ coefficient estimates. Thus, I conclude that individuals are unlikely to overestimate within-pool zero coefficients. The second way in which beliefs can affect judgment accuracy is by affecting the variation (that is, the random error) with which individuals estimate coefficients. Estimates of relationships that are inconsistent with individuals’ beliefs tend to have more random error than do estimates of relationships that are consistent with individuals’ beliefs (Sniezek 1986; Luft and Shields 2001). Thus, because individuals without specific knowledge to the contrary will likely believe within-pool relationships are stronger or more likely to exist than across-pool relationships, within-pool coefficients will likely be estimated with less variation than across-pool coefficients. However, if individuals pay very little attention to across-pool coefficients, then the across-pool coefficients may be treated as though they are zero, and thus will exhibit little random error. The above analysis leads to the following hypotheses in alternative form: H1: Judgment accuracy will be higher (bias will be smaller and variation will not be larger) for within-pool non-zero predictors’ coefficients than for across-pool non-zero predictors’ coefficients. H2: Judgment accuracy will be higher (bias will be smaller and variation will not be larger) for within-pool zero predictors’ coefficients than for across-pool zero predictors’ coefficients. The existence of four types of coefficients (within- and across-pool; non-zero and zero correlations) means there are six possible pairwise comparisons. l have chosen to focus on the two pairwise comparisons for several reasons. First, Hypothesis 3 presents an omnibus test of prediction accuracy, including all coefficients. Second, there is not a well-developed theoretical prediction for the four coefficient pairs that I do not test. 26 Third, the pairs I test in hypotheses l and 2 allow for a clean test of whether differences in implicit coefficient estimation accuracy are due to differences in the overhead cost pool classification. That is, the untested pairs include a manipulation of both whether a coefficient is within-pool or across-pool and whether the environmental correlation is non-zero or zero. Thus, any test of those pairs is inherently a test of an interaction between overhead cost pool classification and cost behavior. Because of the lack of theoretical development and the fact that the emphasis in this portion of the paper is on overhead cost pool classification alone and not on its interaction with cost behavior, such a test would not be meaningful. Hypothesis 3: Prediction accuracy of overhead costs Hypotheses l and 2 make predictions about within-individual effects of overhead cost pool classification on implicit coefficient errors on specific predictors. Such evidence is important to understanding how overhead cost pool classification affects the accuracy of judgment in certain tasks (for example, estimating how much spoilage will change as a result of changing spending on one of the predictors). However, understanding how overhead cost pool classification affects performance on the overall cost prediction task requires understanding how cost behavior and all the individual coefficient estimates jointly affect cost prediction accuracy. Hypothesis 3 predicts how the interaction of overhead cost pool classification and between-individual manipulations of variable cost behavior affects a set of implicit coefficient errors and thus the overall accuracy of participants’ predictions of overhead costs. In the following subsections I describe how I expect the location of the stronger predictor and the sign of relationships to affect individual implicit coefficient errors and thus the overall prediction of overhead costs. 27 Location of stronger predictor Individuals likely have beliefs that costs within pools behave similarly, and costs across pools behave dissimilarly, in the absence of strong cues to the contrary. HOwever, empirical evidence shows that costs in different pools can be statistically significantly correlated (Anderson 1995; Ittner et al. 1997; Joshi et al. 2001). When the stronger predictor is across pools, individuals will devote less attention to the stronger across-pool relationship because it is inconsistent with their beliefs, and thus will underweight it. Further, individuals will have difficulty estimating the within-pool relationship because of its relatively low magnitude (Broniarczyk and Alba 1994). Because all of the implicit coefficients will be estimated less accurately, I expect that ceteris paribus, accuracy of individuals’ predictions of overhead costs will be higher when the stronger predictor is within-pool than when it is across-pool. Sign of relationships Overhead costs can be either positively or negatively correlated. As noted above, typical recommendations for cost system design assume that costs within pools are positively correlated. However, empirical evidence demonstrates that overhead costs can exhibit either positive or negative correlations (e.g., Datar et al. 1993; Anderson 1995; Joshi et al. 2001). Prior research in psychology and accounting finds that estimation of negative relationships is inherently more cognitively difficult, and thus less accurate, than estimation of positive relationships (e.g., Brehmer 1974). Thus, even if individuals are aware of the existence of a negative coefficient relating predictor costs to the target cost, they may be less able to estimate accurately its magnitude than they would if the coefficient were positive. Therefore, I expect that ceteris paribus, the accuracy of 28 individuals’ predictions of overhead costs will be higher when relationships between overhead costs are positive than when they are negative. Note that in this study, all the non-zero predictors are either positively or negatively correlated with the target; that is, there are no cases in which signs are mixed. While empirical evidence indicates that both positive and negative correlations occur simultaneously in practice (e.g., Datar et al. 1993), I chose to include only single-sign cases in this study for simplicity. Interaction of relative strength and sign I expect that the location of the stronger predictor and sign of the relationships among costs will ordinallyinteract to affect judgment accuracy. The effect of the location of the stronger predictor will be larger when relationships among costs are negative than when the relationships are positive. The expected interaction is presented graphically in Figure 5. 29 Figure 5: Graphical presentation of Hypothesis 3 Location of >, — 1- 3 stronger g ------------------- predictor: i-i :1 E); 2 """""""" Within pool i':’ ID ED 4 -o Across pools 5’. Positive Negative Sign of coefficients on predictors This figure presents the hypothesized interaction between the location of the stronger . predictor (either within the same pool as the target cost or across pools) and the sign of the coefficients relating the predictors to the target (positive or negative). When relationships among costs are positive, the reduction in prediction accuracy (1 — 2 in Figure 5) that results from the stronger predictor being across-pool rather than within-pool will be relatively small. Individuals will allocate less attention to the stronger predictor when it is in a different pool from the target cost compared to when it is in the same pool. However, because the relationships among costs are positive, estimation is not as cognitively difficult, and so individuals will be able to estimate the implicit coefficient on the stronger predictor with relatively high accuracy. Further, the cognitive difficulty of estimating the implicit coefficient on the weaker, within-pool predictor will be relatively small because of the positive sign and the relatively high attention it receives because it is within-pool. Thus, the effect on prediction accuracy of the stronger predictor being 30 across-pool rather than within-pool will be relatively small when relationships among costs are positive. When relationships among costs are negative, the reduction in prediction accuracy (3 — 4 in Figure 5) that results from the stronger predictor being across-pool rather than within-pool will be relatively large. Individuals will allocate less attention to the stronger predictor when it is in a different overhead cost pool from the target cost compared to when it is in the same pool. The reduced attention individuals allocate to the stronger predictor will be insufficient to overcome the cognitive difficulty of estimating negative relationships, and individuals’ estimates of the implicit coefficient on the stronger predictor will be less accurate. Further, the difficulty of estimating the implicit coefficient on the weaker, within-pool predictor will be exacerbated by the negative sign, despite the relatively high attention it receives because it is within-pool. Thus, neither implicit coefficient will be estimated accurately, and the effect on prediction accuracy of the stronger predictor being across-pool rather than within—pool will be larger when the relationships among costs are negative than when they are positive. That is, I predict the interaction will be of the form 3 — 4 > 1 — 2. This prediction is consistent with evidence from Sniezek (1986), who found that the effect on judgment accuracy of inconsistency between individuals’ beliefs and actual data is larger when relationships are negative than when they are positive. H3: Overhead cost prediction accuracy is an interactive function of the location of the stronger predictor and the sign of relationship between the predictors and the target. 31 III. RESEARCH DESIGN Experimental design The hypotheses are tested using an experiment with a 2 x 2 x 2 x 2 x 2 mixed design. There are two within-participant factors. The first within-participant factor is whether a predictor is within the same pool as spoilage or across pools (that is, in a different pool from spoilage). The second within—participant factor is whether a predictor has a non—zero coefficient or a zero coefficient relating the predictor to spoilage. There are three between-participant factors. The first between-participant factor is the location of the stronger predictor-target relationship, either within the same pool as spoilage or across pools. The second between-participant factor is the sign of the predictor-target relationships; they were all positive in one condition and all negative in the other. The third between-participants factor is the label of the costs. For half the participants, the predictors within the same pool as spoilage were labeled “inspection” and “testing,” while the predictors in the other pool were labeled “supervision” and “training.” For the other half of the participants, the labels were reversed, such that the predictors within the same pool as spoilage were labeled “supervision” and “training,” while the predictors in the other pool were labeled “inspection” and “testing.” This last factor was manipulated between participants to control for beliefs about relationships among costs with specific labels. There was not a main effect of label or a significant interaction between labels and other variables (all p values > 0.10); thus, this factor is omitted in all analyses reported below. The experimental design is shown in Table l.H ll . . , . . . . . . Partrcrpants responses were collected In five sessrons. A control variable for sessron did not result in a significant main effect or interaction with other variables; thus, session is omitted in all analyses. 32 Table 1: Experimental design uali control I Location of stronger relationship and Quality control Dool Supervision pool sign Within-pool Smilage Inspection Testing Supervisors Training (1) Positive -- l .01 0.0 0.46 0.0 (2) Negative -- -l .01 0.0 -0.46 0.0 Across-me (3) Positive -- 0.46 0.0 1.01 0.0 (4) Negative -- -0.46 0.0 -l .01 0.0 Sumrvision mol Supervision 0001 Quality control pool Within-pool Spoilage Supervisors Training Inspection Testing (5) Positive -- l .01 0.0 0.46 0.0 (6) Negative -- -l .01 0.0 -0.46 0.0 Across-pool (7) Positive -- 0.46 0.0 1.01 0.0 (8) Negative -- -0.46 0.0 - l .01 0.0 The entries in the cells are the coefficients relating each predictor cost to spoilage. Each row, representing a between-participant cost behavior condition, can be read as the specific environmental model for that cost behavior condition, with the addition of a constant. The constant varied between positive and negative sign conditions (-$24,145 and $104,125, respectively). For example, for participants receiving the cost behavior described in row (1), Spoilage = -$24,145 + 1.01 x Inspection cost + 0.46 x Supervisors cost. For participants receiving the cost behavior described in row (6), Spoilage = $104,125 - 1.01 x Supervisors cost — 0.46 x Inspection cost. The first between-participant factor was the location of the stronger predictor-target relationship (correlation = l0.9l; coefficient = 11.01 I), either within the same pool as spoilage or across pools; the other relationship was the weaker one (correlation =I0.4l; coefficient = 10.46l). The second between-participant factor was the sign of the predictor-target relationships; they were all either positive or negative. The third between-participants factor was the label of the costs. For half the participants, the predictors within the same pool as spoilage were labeled “inspection” and “testing,” while the predictors in the other pool were labeled “supervision” and “training.” For the other half of the participants, the labels were reversed, such that the predictors within the same pool as spoilage were labeled “supervision” and “training,” while the predictors in the other pool were labeled “inspection” and “testing.” The first within-participant factor is whether a predictor is within the same pool as spoilage or across pools. The second within-participant factor is whether a predictor has a non-zero coefficient or a zero coefficient. 33 Participants Eighty-one undergraduate business students enrolled in a cost-accounting class at a large public university volunteered to participate in the research. They were given course credit and $5 for participating, and incentive-based pay from $1 to $10 based on their judgment accuracy. Three participants’ responses were eliminated because their accuracy measures were influential outliers]2 The final sample thus included 78 participants. Procedure and task Participants were given a paper-and-pencil task in which they were asked to assume the role of a controller in a manufacturing firm. As part of their firm’s operating budgeting process, they were to predict spoilage costs for 20 similar manufacturing plants for the next month. Participants were asked to predict spoilage cost from the four other overhead costs (inspection, testing, supervisors, and training) in two overhead cost pools (quality control and supervision). Either inspection or supervisors (depending on the label condition) was strongly predictive of spoilage, and the other was weakly predictive. The other two costs (testing and training) were not predictive of spoilage. Each cost pool contained either the stronger or the weaker predictor and one of the zero-correlation predictors. The coefficient relating the stronger (weaker) predictor to spoilage was I1 .0“ (IO.46|). The relationship between spoilage and the two predictive costs is: Spoilage = Constant i 1.01 x Stronger :t 0.46 x Weaker + error, where Stronger (Weaker) refers to 2 . . . Because mean squared error of predictions encompasses all potential causes of error, I use mean squared error to identify outliers. l regressed mean squared error on sign, location of the stronger predictor, and the interaction of sign and location to determine influential outliers using Cook’s D. Two observations had values of Cook’s D greater than (4/n), a reasonable cutoff per Fox (1991), and were thus eliminated. 34 the cost with the stronger (weaker) relationship, and Constant = -$24,l45 and $104,125 in the positive and negative conditions, respectively, and error is zero-mean noise. The rows in Table 1 describe the specific model for each condition individually. Participants were told spoilage costs were not the direct result of a single spending decision but resulted from, and were predictable from, other spending decisions made during the month. Participants were given a learning data set that provided the prior month’s cost data by plant, reported in the two cost pools (see Appendix). To reduce noise, I attempted to set participants’ beliefs about the sign of relationships with the following wording for the positive and negative conditions, respectively: “Higher spending on one (or more) Of these activities could lead to higher levels of spoilage costs being recorded, because more care is taken to identify and discard defective units before they are sent to customers,” or “[ h jigher spending on one (or more) of these activities could lead to lower spoilage costs by improving production quality.” Setting participant’s beliefs about the sign of relationships reduces noise by reducing the likelihood that participants will attempt to estimate the relationship with an incorrect sign. Further, practicing managers may know the sign or direction of relationships among overhead costs without knowing the exact magnitude of those relationships. Participants also received a judgment data set, in which next month’s budgeted levels of the predictor costs were given, with blanks to fill in predictions of expected spoilage costs (see Appendix). Participants were allowed to use calculators.l3 A series of One participant obviously used a statistical calculator, as he generated a very accurate model and reported its R2 on his experimental materials. Results are not affected by the inclusion or exclusion of this participant’s responses. In this task setting, the data set was small enough that complete testing of a multiple regression model was relatively easy, and the loss of efficiency caused by including all possible predictors was minimal. As noted above, the use of calculation aids will help individuals examine relations 35 follow-up questions included manipulation checks, GPA in accounting and statistics courses, questions about beliefs about the relationships among costs, and questions about participants’ reasoning for examining particular relationships among costs. Experimental parameters Within each sign condition, all data sets were statistically identical. I first generated standardized data that fit the data requirements for the research design. I then linearly transformed the standardized data to generate plausible levels of costs. The transformations were such that the means, variances and ranges. of the two predictive costs (inspection and supervision) were identical to each other, thus eliminating any concerns about scale. The four columns of predictors were identical across experimental conditions; they differed only in their placement among cost pools.14 The only differences across any conditions were the values of spoilage, which differed between the negative and positive conditions by the use of a different linear transformation. Further, there were no statistically significant differences for the four predictor costs between the learning and judgment data sets. Thus, there is no concern about nonlinear relationships between spoilage and other costs due to mean differences between learning and judgment data.l5 Parameters are described in Table 2. accurately and apply their decision models consistently. However, an important step in assuring accurate estimation is the choice of relationships to examine. The use of a calculation aid will not necessarily help in this part of the task. 14 . . . The predictive costs were always two columns away from sporlage, to control for any effects of proximity on judgment. 15 . . . . . . . If means were different across learning and judgment data, partrcrpants might have assumed and applied nonlinear models, which would not be adequately captured by my linear modeling technique. This would result in inaccurate measures of their judgments. 36 Table 2: Experimental parameters P value of P value of Host for Levene’s mean Std. test for 3d. Predictor Dataset Mean difference Deviation difference Stronger Learning 33595 0.719 15,457 0.482 Judgment 36,915 13,756 Weaker Learning 38,480 0.702 1 1,345 0.177 Judgment 36,880 14,713 Zero Coefficient Learning 26,860 0.25] 3,170 0.307 Within-pool Judgment 28,195 4,022 Zero Coefficient Learning 15,475 0.202 2,763 0.189 Across-pool Judgment 1 4,460 2,] 34 Environmental model: Spoilage = Constant :I: 1.01*Stronger :I: 0.46*Weaker where Constant = -$24,125 in the positive condition and $104,125 in the negative condition Note: I use Levene’s test because the underlying data are known to be non—normal, and the F—test is sensitive to departures from normality. This table demonstrates that the data used for the learning data sets are not statistically significantly different from those in the judgment data set. Measurement of variables By regressing participants’ predictions of the target on the levels of the predictors, I estimate their implicit coefficients on each predictor.16 I then estimate participants’ implicit coefficient errors by subtracting the environmental coefficients from the implicit coefficients. Each participant’s judgment model has a constant and coefficients for each of the four costs. I chose to treat an implicit coefficient as non-zero if it was significant at l . . . . . . 6I assume partrcrpants use a linear, additive model because the model used to generate the experimental data is linear and additive. Further, evidence from psychology research (Brehmer 1974; Dawes and Corrigan 1974) indicates that individuals typically use linear, additive models in subjective judgment. 37 p<0.10; otherwise I treated the implicit coefficient as zero.17 Thus, my models of participants’ judgments include up to four coefficients. I also estimate the optimal or environmental model from the learning data set and generate the environmental coefficients, which represent the optimal coefficients participants could estimate. I compare participants’ implicit coefficients to the environmental coefficients as described below. To measure for an effect of overhead cost pool classification on bias (that is, differences between participants’ within-pool and across-pool estimation errors) for Hypotheses l and 2, I compute a directional error difference measure equaling the signed error (negative signed error) on each coefficient when relationships are positive (negative). Thus, coefficients whose magnitudes are overestimated (underestimated) will have positive (negative) directional error, regardless of the sign of the relationships in the data. I then subtract the within-pool directional error from the across-pool directional error to create the directional error difference variables. I do this for the costs with non- zero coefficients for H1 and the costs with zero coefficients for H2 (DirCE and DirZeroCE, respectively). If either variable is negative (positive) and statistically significant, participants have on average underestimated (overestimated) across-pool coefficients more than within-pool coefficients. To test for an effect of overhead cost pool classification on variation for Hypotheses l and 2, I use the variance of the within- pool and across-pool coefficient errors. I also estimate repeated-measure ANOVAS, treating the difference between within-pool and across-pool coefficients as a within-participant effect and the two cost '7 My results are unaffected by alternative cutoff choices of 0.05 and 0.15. 38 behavior manipulations (location of stronger predictor and sign) as between-participant effects. I then use the AN OVA to generate statistics describing whether the within- participant effect varies across the cost behavior conditions. For judgment accuracy in Hypothesis 3, I use several measures to test for differences in accuracy and explore reasons for those differences in accuracy. The main measure I use is the mean squared error of participants’ predictions compared to optimal predictions made by the environmental model.18 I also use three measures derived from the lens model to investigate how different sources of judgment error affect overall prediction accuracy.l9 Matching is the correlation between the predictions made by the environmental model and predictions made by each participant’s model. Matching measures how accurately participants estimated the relative values of the coefficients for the four predictors. Consistency is the correlation between participants’ predictions and the predictions made by their model. Consistency measures how consistently participants applied their implicit models. A final measure I use is consensus, or the averaged correlations of participants’ predictions with predictions of other participants in the same experimental condition. Consensus is not a measure of individual accuracy but rather a measure of the similarity of participants’ predictions to each other. However, this measure is important because low levels of consensus imply costly disagreement among individuals and reduced likelihood of reaching agreement about judgments and their effects on organization performance. 18 Use of mean absolute error does not affect results. ‘9 Lens model studies typically include achievement as well, which is the correlation between participants’ predictions and actual outcomes. Achievement is the product of three factors: matching, consistency and environmental predictability. In my study, environmental predictability is constant across conditions and is very close to its maximum value of 1.0; therefore, achievement is a redundant measure of judgment accuracy. For parsimony, I omit achievement from my study. 39 IV. RESULTS Six participants from the final sample of 78 participants failed the post-task manipulation check question that asked them to identify which cost pool contained each of the five costs (the four predictors and spoilage). The failure of a post-task manipulation check may be due to failure of memory rather than one of attention or understanding. Further, the failure to recall the label on the overhead cost pool does not necessarily mean that the participant did not understand which individual cost accounts were classified together within each pool. Therefore, I retain the responses of the participants who failed the manipulation check; results are unchanged by the exclusion of these participants’ data. Self-reported accounting and statistics GPA did not significantly affect any dependent variable or interact with any independent variable to affect any dependent variable (p < 0.10), and so are excluded from all analyses. 1 also perform manipulation checks to ensure that participants attended to the cost behavior manipulations. I examine whether participants’ implicit coefficients on the stronger predictor were larger than those on the weaker predictor, and whether participants’ implicit coefficients carried the appropriate sign. Table 3 shows results of these manipulation checks. Panel A presents subjective coefficients for the non-zero predictors across the two sign conditions.20 Data are consistent with participants placing more weight on stronger predictors than on weaker predictors and implicitly identifying the signs of coefficients accurately. Evrdence on rmplrcrt coefficrents on zero-correlation predictors IS found in test of Hypotheses l and 2. 40 Table 3: Manipulation checks Panel A: Participants ’ coefficients Stronger Weaker Sign predictor predictor Positive Mean 0.65 0.14 N 41 41 Std. Deviation 0,41 0.33 Negative Mean -0.65 -0.06 N 37 37 Std. Deviation 0,81 0.36 This panel shows the mean implicit coefficients on non-zero predictors in participants’ subjective prediction models by sign condition. Panel B: T-tests Std. Error Sig. (2- Sign Comparison Mean Mean t (if tailed) Positive Stronger predictor - Negative Stronger predictor - -O.59 0'09 _ 6.00 36 0.000 Weaker redictor ms to test whether participants attended to data. Panel B of Table 3 presents t-tests for mean differences between implicit coefficients on stronger and weaker predictors. Results in Panel B confirm that participants placed significantly more weight on stronger predictors than weaker predictors in both positive and negative conditions (p-values for nonzero < 0.001 for both signs). Thus, the results indicate that participants did not simply weight within-pool predictors and ignore across-pool predictors; rather, their pattern of weights indicates they paid attention to the relationships in the learning data set. 41 Hypotheses l and 2 Hypotheses l and 2 predicted that participants will estimate within-pool implicit coefficients more accurately than across-pool implicit coefficients for non-zero and zero predictors, respectively. Panel A of Table 4 present tests for bias. The first row presents results for Hypothesis 1. Participants underweighted across-pool non-zero predictors more than within-pool non-zero predictors, consistent with beliefs that costs assigned within pools are more likely to be strongly related than are costs assigned into different pools. DirCE is significantly negative (mean = -0.18, p = 0.005). Thus, Hypothesis 1 is supported for bias. Results for Hypothesis 2 are shown in the second row of Panel A of Table 4. Participants did not implicitly overweight within-pool zero coefficients more than across-pool coefficients. DirZeroCE is not significantly different from zero (mean = 0.01 , p = 0.917). Hypothesis 2 is not supported for bias. Panel B of Table 4 presents tests of Hypotheses l and 2 for variation. The variances of across-pool and within-pool non-zero predictors’ coefficient errors are not statistically significantly different (F = 1.01; p = 0.484). Thus, Hypothesis 1 is supported for variation. For zero predictors’ coefficients, the variance of the within-pool coefficient errors is significantly greater than that Of the across-pool coefficient errors (F = 1.61 ; p = 0.019). Thus, Hypothesis 2 is not supported for variation. 42 Table 4: Hypotheses l and 2 Panel A: Bias Dependent variable (A) - (B) = (A) (B) (C) Mean Mean across-pool within-pool Mean implicit implicit difference , coefficient coefficient (across-pool Paired COEffiC'em estimation estimation minus within- samples t- Sig. (2- H type n error error pool) value tailed) 1 Non-zero 78 -0.45 -0.27 -0.18 2.867 0.005 2 Zero 78 0.12 0.11 0.01 -0.104 0.917 This panel presents descriptive detail and differences for the measures used to test Hypotheses 1 and 2. Columns A and B present mean across-pool and within—pool implicit coefficient errors, respectively. Column C presents the difference between across-pool and within-pool implicit coefficient errors. The differences presented in column C are the dependent variables of interest used in testing Hypotheses l and 2. Panel B: Variation = (A) (B) Across-pool Within-pool implicit implicit coefficient coefficient estimation estimation Coefficient error error Sig. (2- Mothesis type II variance variance Variance ratio tailed) 1 Non-zero 78 0.172 0.171 1.01 0.484 2 Zero 78 0.362 0.584 1.61 0.019 This panel presents descriptive detail and differences for the measures used to test Hypotheses l and 2. Columns A and B present mean across-pool and within-pool implicit coefficient error variances, respectively. Column C presents the difference between across-pool and within—pool implicit coefficient error variances. The differences presented in column C are the dependent variables of interest used in testing Hypotheses l and 2. Directional error is the signed error (opposite of signed error) for participants in the positive (negative) condition. Positive (negative) directional error indicates overweighting (underweighting). Non-zero (zero) coefficient indicates that the errors represented in the row are implicit coefficient errors on predictors with non-zero (zero) environmental coefficients relating the predictor to the target cost. 43 Taken together, the results for Hypothesis 1 support the hypothesis for both bias and variation. Thus, I conclude that for non-zero predictors, judgment accuracy is higher for within-pool coefficients than for across-pool coefficients. That is, within-pool coefficients are estimated with less bias and no more variation than across—pool coefficients. Taken together, the results for Hypothesis 2 do not support the hypothesis for either bias or variation. Thus, I conclude that for zero-correlation predictors, judgment accuracy is not higher for within-pool coefficients than for across-pool coefficients. Specifically, bias is not statistically significantly different, and variation is higher for within-pool coefficients than for across-pool coefficients. Tests in Table 4 average results across the cost behavior conditions. It is possible that support for Hypothesis 1 is not invariant to the cost behavior condition; that is, it is possible that the effect of overhead cost pool classification on judgment is significantly different in one cost behavior condition than in the other conditions. To examine whether this is the case, I estimate a mixed—design ANOVA. The within—participant effects are the differences between the across-pool implicit coefficient estimation error and the within- pool implicit coefficient estimation error for both the nonzero and the zero-correlation predictors. The between-participant effects are the two manipulations of the relationships in the data (location of stronger predictor and sign). I use planned contrasts to generate statistics describing whether the within- participant effect in a particular cost behavior condition varies from the mean within- . . . . .. 2122 . partrcrpant effect in the other three cost behavror conditions. ’ Note that a finding that 21 . . . . . . . An alternative test would be to split my sample by cost behavror condition and test for a statistically significant effect of overhead cost pool classification within each condition. The sample is too small to allow reasonable power for such a test. 44 the within-participant effect in one condition does not vary from the mean effect in other conditions is the confirmation of a null hypothesis. Given my relatively small sample size, this test is not a reliable indicator of a lack of a difference across conditions. Further, I have no theoretical basis for hypothesizing a difference among cost behavior conditions. Thus, I treat these contrast tests as exploratory. Untabulated results indicate the within-participant effects from the mixed ANOVA are identical to the t-tests presented in Table 4, as expected. Table 5 presents the results of the planned contrasts for non-zero predictors. None of the planned contrast results was significant, and partial eta-squared values for the contrasts are small (all values < 0.01). Taken together, the above contrast test results provide no evidence that the effect of overhead cost pool classification varies across the four data conditions I examined. Hypothesis 3 Hypothesis 3 states an expectation that participants’ prediction accuracy is a result of the interaction between the two manipulations of cost behavior (location of stronger relationship and sign), such that the stronger predictor being across pools will reduce overhead cost prediction accuracy more when relationships between the predictors and the target are negative than when they are positive. As noted above, I use four measures of accuracy. Table 6, Panel A presents descriptive statistics for the measures of accuracy. Cell-by—cell examination indicates accuracy is affected by the interaction of the cost behavior conditions as hypothesized. The decrease in accuracy between the across-pool 2 Because there are four data conditions, there are three degrees Of freedom, and one of the four contrasts is redundant. However, because I am not making inferences from the contrast tests, the potential for inflated Type I error is not problematic. Also, 1 have no theoretical basis for excluding of any of the four contrasts. Therefore, I include all four contrasts for completeness of presentation. 45 and within-pool measures of accuracy is greater for the negative condition than for the positive condition. Table 5: Sensitivity analysis: across-pool minus within-pool implicit coefficient errors by cost behavior condition: Each cost behavior condition contrasted with the mean of the other three cost behavior conditions Between-participant condition (location of stronger predictor, sign) Partial compared to the mean of the other Contrast Standard eta three conditions estimate error Sig. squared Within-pool, positive -0.24 0.45 0.596 0.004 Across-pool, positive 0.13 0.45 0.779 0.001 Within-pool, negative 0.16 0.46 0.728 0.002 Across-pool, negative 005 0.47 0.913 0.000 This table shows the result of specified contrast tests for a mixed ANOVA. The within- participants effect is the directional implicit coefficient estimation error for the across—pool predictor minus the directional implicit coefficient estimation error for the within-pool predictor. The between-participants effects are the two manipulations of location of stronger predictor and sign of relationships. Directional error is the signed error (opposite of signed error) for participants in the positive (negative) condition. Positive (negative) directional error indicates implicit overweighting (underweighting) of a predictor. The contrast estimate is the result for a contrast that specifies the difference between the mean of the within-participant effect for the cost behavior condition and the mean of the within—participant effects for all other data conditions. 46 Table 6: Hypothesis 3 Panel A: Descriptive statistics Stronger Sign Predictor Mean Std. Deviation N Mean Squared Positive Within pool 223,234,087 212,028,568 21 Error Across pools 169,395,071 136,904,061 20 Total 196,971,152 179,229,418 41 Negative Within pool 226,601,499 277,416,666 19 Across pools 367,097,536 296,452,145 18 Total 294,950,922 291 ,631 ,408 37 Total Within pool 224,833,607 242,027,520 40 Across pools 263,043,607 244,972,613 38 Total 243,448,735 242,642,256 78 Matching Positive Within pool .75 .36 21 Across pools .84 .18 20 Total .79 .29 41 Negative Within pool .76 .40 19 Across pools .36 .67 18 Total .56 .58 37 Total Within pool .75 .38 4O Across pools .61 .53 38 Total .68 .46 78 Consistency Positive Within pool .85 .22 21 Across pools .85 .14 20 Total .85 .18 41 Negative Within pool .82 .25 19 Across pools .65 .42 18 Total .74 .35 37 Total Within pool .84 .23 40 Across pools .75 .32 38 Total .80 .28 78 Consensus Positive Within pool .48 .32 21 Across pools .52 .16 20 Total .50 .25 41 Negative Within pool .49 .20 I9 Across pools .12 .28 18 Total .31 .30 37 Total Within pool .48 .26 40 Across pools .33 .30 38 Total .41 .29 78 Sign is the sign of relationships among costs. StrongPredictor refers to the location of the stronger relationship. Mean squared error is the mean squared error of participants’ predictions compared to optimal predictions. Matching is the correlation between the predictions made by the environmental model and predictions made by the participants’ model. Consistency is the correlation between participants’ predictions and the predictions made by their model. Consensus is the averaged correlations of participants’ predictions with predictions of other participants in the same experimental condition. Table 6: Hypothesis 3, continued Panel B: Multivariate ANOVA Wilks' Hypothesis Error Effect Lambda F(a) df df Sig. Intercept 0.03 552.23 4 71 0.000 Sign 0.84 3.281 4 71 0.016 StrongPredictor 0.86 2.87 4 71 0.029 ' * 5‘“ 0.85 3.17 4 71 0.019 Stron Predictor a Exact statistic This panel presents the results of a MANOVA, with the following variables: Independent variables: Sign is the sign of relationships among costs. StrongPredictor refers to the location of the stronger relationship. Dependent variables: Mean squared error is the MSE of participants’ predictions compared to predictions derived from the environmental model. Matching is the correlation between the predictions made by the environmental model and predictions made by the participants’ model. Consistency is the correlation between participants’ predictions and the predictions made by their model. Consensus is the averaged correlations of participants’ predictions with predictions of other participants in the same experimental condition. 48 Table 6: Hypothesis 3, continued Panel C: Individual ANO VAs Source Dependent Variable Mean Square F Sig. Corrected Model Mean Squared Error 1.33E+l7 2.380 .076 Matching .856 4.581 .005 Consistency .177 2.415 .073 Consensus .637 10.477 .000 Intercept Mean Squared Error 4.73E+l 7 84.604 .000 Matching 35 .3 88 189.326 .000 Consistency 48.81 1 664.833 .000 Consensus 12.603 207.276 .000 Sign Mean Squared Error 1.96E+17 3.516 .065 Matching l .072 5 .736 .019 Consistency .268 3.652 .060 Consensus .737 12.1 15 .001 StrongPredictor Mean Squared Error 3.65E+16 .653 .422 Matching .458 2.452 .122 Consistency . .154 2.096 .152 Consensus .520 8.553 .005 Sign * Mean Squared Error 1.84E+l7 3.284 .074 StrongPredictor Matching 1.167 6.243 .015 Consistency .134 1 .826 .181 Consensus .755 12.409 .001 Error Mean Squared Error 5.59E+16 Matching .187 Consistency .073 Consensus .061 See Panel B for variable descriptions. Panel B shows multivariate results of a MANOVA on the four measures of accuracy. As a group, they are significantly affected by the interaction of the location of stronger predictor and sign conditions (p = 0.019). Panel C shows the individual ANOVAs for each accuracy measure; all except consistency are significantly affected by the interaction (p values 5 0.075; p value for consistency = 0.181). Thus, H3 is partially supported. Note that Panel C of Table 6 indicates significant main effects of location of stronger predictor and sign on several measures of accuracy; however, due to the presence of the significant interaction effect, I do not interpret the main effects. 49 Interpretation of the results of Hypotheses 3 is as follows. For participants for whom the stronger predictor was across pools, those in the negative condition exhibited lower prediction accuracy than those in the positive condition. As revealed by their lower matching score, participants in this condition implicitly estimate the relative coefficients less accurately, indicating that their predictions result from less well-specified models. Thus, their predictions vary more from the optimal predictions, as indicated by higher mean squared error.23 Finally, the participants in this cell use models that differ from each other more than those in other cells, as indicated by their lower levels of consensus. Supplemental analysis The above analysis demonstrates that overhead cost pool classification affects participants’ estimation of implicit coefficients relating costs within and across pools, and the accuracy of their judgments of costs that result from their estimations. However, the results above do not specifically address why the effect occurs. Thus, I include analysis of experimental follow-up questions that provide some evidence on the question of why overhead cost pool classification affects coefficient estimation performance and judgment accuracy. Because these results rely on retrospective self—reports of cognitive processes, which have come under scrutiny and are commonly viewed as suspect (e.g., Nisbett and Wilson 1977), these results are considered supplementary. Table 7, Panel A presents the follow-up questions, which are intended to elicit responses from participants as to both how and why the overhead cost pool classification affected their estimation and prediction accuracy. Participants were asked to think about 23 The interaction is not significant for consistency (p = 0.181) but is directionally consistent with predictions. Because participants were allowed to use calculators, they should have been able to apply their implicit coefficients equally consistently across cells. High levels of consistency are therefore to be expected across all experimental conditions if participants explicitly identified relationships and applied them using calculators. 50 how they estimated the relationships, and then to allocate 100 points among six possible descriptions of their estimation strategies. Item C is intended to capture the likely beliefs that costs within pools are strongly correlated. Other items are intended to capture likely alternative beliefs and other foci of attention that may have resulted from the various conditions. Mean points allocated to each item are shown in Panel A. Item C received more points than did any other item, indicating that on average, participants considered overhead cost pool classification an important factor in their cost prediction strategy because overhead cost pool classification likely describes overhead cost behavior. The difference between points allocated to Item C points allocated to all other items except Item A is statistically significant (Item A p = 0.917, all other p’s < 0.001). Note that item B, which offers responsibility accounting as an alternative reason for attending to overhead cost pool classification, received fewer points than did item C (p = 0.001, untabulated result). It is not necessary that Item C receive significantly more points than any other item (and that the within-pool relationship receive strictly more attention than any other relationship) in order to support the hypothesis development in Chapter 2. It is sufficient that some participants allocate a non-trivial amount of attention to the relationship and that that attention affect the accuracy Of those participants’ judgments. By construction, the objectively optimal strategy is to ignore the overhead cost pool classification and to use all the data independently of its pool location. Thus, any deviation from that strategy in the direction suggested by the overhead cost pool classification that reduces participants’ judgment accuracy supports the hypothesis. 51 Table 7: Supplementary analysis Panel A: Questions Points Statement 3. 26.30 I looked equally for relationships between spoilage and other costs regardless of the cost pool (that is, cost pools had no effect on my learning strategy) because I did not expect to learn anything about the behavior of the costs from the cost pools. b. 16.35 I first looked for a relationship between spoilage and the costs in the same cost pool because one person or group of people has responsibility for managing the costs within the pool. c. 26.78 I first looked for a relationship between spoilage and the costs in the same cost pool because costs within the same pool are likely to have strong relationships. d. 12.57 I first looked for a relationship between spoilage and the costs in the other cost pool because an important cause of spoilage costs is likely to be included in a different cost pool. e. 14.62 I ignored all the text descriptions and cost pools and only used the numerical values to learn about relationships between costs. f. 3.38 Other. 100.00 Total points This panel presents follow-up questions. The entries in the points column represent mean responses (this column was blank when presented to participants). 52 Table 7: Supplementary analysis, continued Panel B: Comparison of points allocation across cost behavior conditions Location of strong Std. Sig. (2- predictor N Mean deviation tailed) Item A Within pool 40 23.68 25.56 0.34 Across pools 38 29.05 23.41 Item B Within pool 40 17.51 13.95 0.46 Across pools 38 15.13 14.45 Item C Within pool 40 29.31 21.68 0.28 Across mols 38 24.11 19.78 Item D Within pool 40 10.50 14.13 0.17 Across pools 38 14.74 12.78 Item E Within pool 40 15.00 15.67 0.85 Across pools 38 14.21 21.36 Item F Within pool 40 3.75 7.99 0.48 Across pools 38 2.50 7.51 fign Item A Positive 41 28.44 25.31 0.42 Negative 37 23.92 23.75 Item B Positive 41 17.94 15.56 0.30 Negative 37 14.59 12.38 Item C Positive 41 26.30 17.74 0.84 Negative 37 27.30 24.00 Item D Positive 41 13.05 14.99 0.74 Negative 37 12.03 1 1.99 Item E Positive 41 13.17 18.31 0.47 Negative 37 16.22 18.91 Item F Positive 41 1.10 4.11 0.01 Negative 37 5.41 9.96 See Panel A for the post-experimental questions (Items A through F) 53 Table 7: Supplementary analysis, continued Panel C: Correlations among points allocations and prediction errors Strong Predictor = Within pool Mean (11 = 40) Squared Item A Item B Item C Item D Item E Item F Error Item B -0.34 Item C -0.63 0.07 Item D -0.26 -0.15 -0.13 Item E —0.15 -0.34 -0.21 -0.14 Item F -0.17 0.07 -0.14 -0.02 -0.08 Mean Squared Error 003 0.11 -0.15 0.19 0.01 -0.05 Mean Absolute Error 003 0.09 012 0.26 -0.04 -0.12 0.97 Strong Predictor = Across pools Mean (11 = 38) Squared Item A Item B Item C Item D Item E Item F Error Item B .044 Item C -0.55 0.24 Item D -0.17 -0.24 0.00 Item E -0.14 -0.24 -0.48 -0.28 Item F —0.17 -0.07 0.04 0.10 -0.21 Mean.Squared Error -0.30 -0.02 0.32 -0.01 0.08 -0.10 Mean Absolute Error 026 -0.07 0.28 -0.02 0.12 -0.11 0.98 Note: highlighted correlations are significant at p < 0.05, one-tailed See Panel A for the post-experimental questions (Items A through F). Mean squared error is the mean squared error of participants’ predictions compared to predictions derived from the environmental model. Mean absolute error is the mean absolute error of participants’ predictions compared to predictions derived from the environmental model. It is possible that participants’ ex-post self-reported attention allocation was affected by their cost behavior condition. If this is the case, then the post-experimental 54 questions would not be cleanly measuring the strategies participants used initially, which should be independent of the data to which they applied the strategies. Once a participant begins to examine the data, his strategy may change as a result of the data he sees. However, my purpose with these questions is to ascertain what strategy participants initially used — that is, to which relationships they initially directed their attention. To provide evidence on whether cost behavior condition affected self-reports of attention allocation, I examine participants’ responses to the questions and test for differences across data conditions. If results are not statistically significant, I will conclude that the cost behavior conditions did not affect the allocation of points, and that the responses to questions do not reflect how strategies may have been affected by the cost behavior conditions. Table 7, Panel B shows the results of t-tests on point allocations by location of strong predictor and sign. Location of strong predictor did not affect the allocation of points to any item (all p values > 0.17). Sign only affected the allocation of points to item F (p = 0.01); given that the mean number of points allocated to item F in the positive (negative) condition was only 1.10 (5.40), this result is considered inconsequential to the overall analysis. Further, an untabulated MAN OVA indicates that there was no interactive effect of the location of the strong predictor and sign on the points allocated to any item (all p values > 0.14). Thus, I conclude that the cost behavior conditions did not materially affect the allocation of points in the post-experimental questions. If the overhead cost pool classification affected the allocation of participants’ attention to relationships in the data, and the allocation of attention affected the accuracy of predictions, then there should be an effect of the participants’ prediction strategies (as 55 measured by the self-reported points allocation) on the error of participants’ predictions.24 This effect should be conditional on the location of the stronger predictor. When the stronger predictor is within the same pool as the target, then the prediction task is relatively easy. Given the magnitude of the stronger relationship and the lack of noise in the data, even moderate attention paid to the within-pool predictor should enable participants to estimate the coefficient on the strong predictor fairly accurately. Thus, strategies that involve either looking first at the within—pool relationship (items B and C) or looking at all relationships regardless of pool location (items A and E) should be associated with lower error. A strategy of first looking at the across—pool relationship (item D), however, would lead to higher error, as participants would allocate the majority of their attention toward the weaker predictor. Thus, points allocated to items A, B, C and B should be negatively correlated with prediction error, while points allocated to item D should be positively correlated with prediction error. Conversely, when the stronger predictor is in the opposite pool as the target, a strategy of allocation attention first to predictors within the same pool as the target and later to predictors in the other pool will result in higher error. Thus, for across-pool participants, prediction error should be positively correlated with points allocated to those items (B and C) that indicate prediction strategies that allocate attention first to predictors in the same pool as the target. A strategy of allocating attention to across-pool relationships (item D) or to all relationships regardless of pool location (items A and E) will allow participants to identify the stronger predictor in the opposite pool without first 24 . . . . . . Untabulated results indicate that correlations among pornts allocated to the post-experimental questions and DirCE are directionally consistent with the discussion below relating strategies to overall prediction error (MSE and MAE), but are not statistically significant. 56 having allocated much attention to the within-pool relationship, and so points allocated to these items should be negative correlated with error. Table 7, Panel C presents correlations among the points allocations and prediction error. For the within-pool participants, prediction error measures do not significantly correlate with points allocations. However, several of the correlations are quantitatively in the predicted direction. Correlations between both error measures and Item C are negative.25 Correlations between error measures and item D are positive (for MAE p = 0.055; for MSE p = 0.11; one-tailed). Thus, those subjects whose prediction strategies included early allocation of attention to predictors with the same pool as the target (which includes the strong predictor) made predictions with less error, and those who allocated initial attention to the predictors in the other pool (including the weak predictor) made predictions with more error. For across-pool participants, the results are somewhat stronger. Points allocated to item C are positively correlated with both error measures (p values for MSE and MAE both < 0.05) and points allocated to Item A are negatively correlated with both error measures (p value for MSE < 0.035; p value for MAE = 0.059). Thus, those subjects whose prediction strategies included early allocation of attention to predictors with the same pool as the target (which includes the weak predictor) made predictions with more error, and those who allocated initial attention to the predictors in the other pool (including the strong predictor) made predictions with less error. 25 It is possible that the restricted range of points allocated to items affects the correlations. The ranges for items are: A: 100; B: 60; C: 100; D: 75; E: 80; and F: 40. Thus, items A and C, which have the greatest ranges, also have statistically significant correlations with error measures for the across-pool participants, while other items have no significant correlations. 57 Though the results are not consistently statistically significant, the supplemental analysis overall is consistent with the theoretical development in Chapter 2 above. The evidence is consistent with prediction error resulting in part from participants’ allocation of attention among various predictors, and with participants allocating attention in part based on the classification of costs within pools. 58 V. DISCUSSION AND CONCLUSION Discussion This study provides evidence that overhead cost pool classification affects the accuracy of judgments about cost behavior by users of accounting systems. I demonstrate that while individuals do pay significant attention to the data they are given, they estimate coefficients within overhead cost pools more accurately than they do coefficients across overhead cost pools, and there is evidence consistent with the effect being robust to differences in the location of a stronger vs. a weaker predictive relationship and the sign of the coefficients. I also show that individuals underestimate the magnitude of across- pool coefficients more than they do for within-pool coefficients. This pattern is consistent with empirical evidence found in Joshi et al. (2001), in which the magnitude of relationships among costs assigned into different cost pools was underestimated by managers. While there are likely other causes for the estimation errors exhibited by managers at those firms, I have demonstrated that overhead cost pool classification is a plausible cause of the errors. I have also shown that the accuracy of participants’ cost predictions depends on an interaction between the pool location of the stronger predictive relationship and the sign of coefficients. Participants facing negative relationships between costs and stronger predictors located across cost pools exhibited greater error and lower consensus than did other participants. More broadly, my results provide evidence on tradeoffs organizations face when designing cost accounting systems. Organizations’ overhead cost pool classification decisions should be informed by judgment-related effects as well as the potential for product costing errors as described by prior literature. 59 Contribution I contribute to the academic and practice literature on cost system design by positing and finding evidence for a cognitive effect of cost system design in addition to the information effect previously considered (e. g., Noreen 1991; Christensen and Demski 1995, 1997). I present evidence that classification decisions affect the use of information provided by cost systems as well as the provision of the information itself. My results suggest that cost systems that separate into different cost pools those costs whose relationships are important and are estimated subjectively may hinder effective judgments and decisions by managers. Decisions such as those faced by managers of firms examined in Anderson (1995) (i.e., determining optimal production flexibility, including setup and production schedules) and Joshi et al. (2001) (i.e., determining the costs of regulation, and making various decisions not obviously affected by regulation but in a highly regulated context) may be affected by cost system design, even if managers have all the information they need. Future research can investigate these findings further; for example, researchers may endeavor to determine when classifications of accounting information are likely to have larger or smaller effects on judgment, or whether experience or instruction affects classification effects. Further, I present evidence that the effect of overhead cost pool classification on managers’ judgment performance is dependent on the sign of the relationships among costs. The effect of separating a relationship into different cost pools is greater when costs are negatively correlated (that is, when they are substitutes) than when costs are positively correlated (that is, when they are complements). This suggests that managers’ judgment performance will vary depending on whether the economic environment presents positive or negative correlations. For example, managers attempting to predict 60 the effects of investments in training on the cost of waste (assuming these two costs are negatively correlated) may experience significant judgment failures if the costs are in different pools. Future research can investigate the boundary conditions on this interaction (e. g., how environments with mixed signs affect judgment performance, whether similar effects are noted between cost and revenue items classified separately; whether transforming negative correlations into positive correlations improves judgments). I also contribute to practical and instructional literature on cost system design. As noted, typical textbook instruction of cost pool construction features heavy discussion of correlation-based pool construction for accurate product costing. Discussions of responsibility accounting, performance evaluation and other uses of pooled cost accounting systems are typically separated from discussions of cost system design. Practitioners and students might benefit from a more integrated approach to discussions of cost system design, including discussions of the assumptions underlying cost pool design. For example, students might be presented with “classroom experiment” in which they are given a single cost system and asked to make multiple decisions, including external reporting, product costing and performance evaluation decisions. Other students might be asked to make the same decisions for the same firm, but with a different set of cost pools. Such a task might illuminate the compromises inherent in cost systems in practice, and help mitigate the effects of the product-costing focus of cost system design instruction in classrooms. Similarly, practitioner-oriented recommendations about designing cost pools around highly correlated costs (e.g., Roth and Borthick 1991; 61 Kaplan and Cooper 1998) might provide balance by advising users that such cost system designs have the potential to affect users’ judgments about relationships among costs. Limitations This study is subject to several limitations. This study used student participants in a laboratory setting. My participants generally had little work experience, and'none of my participants had any experience working in the hypothetical firm or the given accounting system. Thus, there is a possibility that individuals with experience working with their firms’ accounting systems would produce different results. To the extent that the results of my experiment are not caused by cognitive processes common to most individuals educated in basic cost accounting, or may be affected by specific experience or knowledge, my results may not be representative of managers performing familiar tasks. Several factors provide evidence in favor of my results’ generalizability. One is the fact that important across-pool relationships were grossly underestimated by experienced managers in J oshi et al. (2001). This finding provides corroborating evidence that the effects noted here exist outside the laboratory. Also, the pervasive nature of categorization noted in the psychology literature (Murphy and Medin 1985; Markman and Gentner 2001) lends support to the notion that categorization effects will occur and affect reasoning in a broad variety of circumstances. In this vein, Hopkins (1996) provides evidence that experienced financial analysts’ predictions are affected by the balance-sheet classification of hybrid securities, thus corroborating the effects of classification on judgment of experienced users. Finally, Vera-Munoz et al. (2001) provide evidence that both task-specific accounting experience, not general accounting experience, and appropriate analysis formats are required for maximum task performance. If managers do not have task-specific experience making across-pool 62 judgments of relationships among overhead costs, general experience with accounting systems may not aid judgment. Second, it is possible that the results are largely driven by the particular reporting format I used. That is, the reporting format presented details of pooled items separated into columns on the page, thus potentially reinforcing the pooling and adding a perceptual (that is, visual) element of classification to the conceptual element of the accounting system design. Thus, it is possible that if systems used in practice report costs in a format that does not visually reinforce the classification, the effect on managers’ judgments will be reduced or will not occur at all. However, it is unlikely that many systems in practice do not visually segregate costs within classifications into portions of reports, or even report them on separate pages or reports. A primary purpose of cost pools is the segregation of costs for product costing purposes, so it is likely that reports of pooled costs will likely feature some degree of visual segregation. Future research can examine effects of various types of accounting information display of classified accounting information on judgment performance. Finally, the mediating variable, the attention-allocation effect of overhead cost pool classification, is not well supported by my measures. It is possible the retrospective reports of strategy were ill-constructed. It is also possible that participants’ attention allocation was not the only mediating variable, but there was another mediator, such as strong beliefs that certain costs have strong relationships or that cost pools are actually classified according to meaningful statistical relationships. My manipulation of the labels on costs was intended to eliminate the alternative explanation that participants had strong beliefs about relationships among certain types of costs. Given that there were no main or 63 interactive effects of the cost label variable, it does not seem likely that my results are driven by this alternative explanation. Further, manipulation checks reported in Table 3 indicate that participants generally attended to the data in all pools, and did not ignore across-pool relationships. The pattern of participants’ implicit coefficients indicated that the participants put more weight on stronger relationships both within- and across-pools, and generally identified the signs of the relationships. This appears inconsistent with participants having strong beliefs that across-pool correlations are necessarily low. However, future research can allow for more direct measures of attention allocation as a mediator, such as process measures or eye- and gaze-tracking devices, to test this mediator more fully. 64 APPENDIX 65 [Note: this is the instrument for the following conditions: positive correlations, stronger predictor is within-pool, spoilage is in the supervisory costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-terrn budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to higher levels of spoilage costs being recorded, because more care is taken to identify and discard defective units before they are sent to customers. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpfirl in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 1. The Quality Control pool includes inspection and testing costs. This pool captures the costs of ensuring that no defective products go out to customers. 2. The Supervision pool includes wages and benefits for supervisors and the costs of training programs, and also spoilage costs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 66 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s firture profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised. It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 67 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend an inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 68 555.:~ 555.5... 5515*. 555.N5 555.5 55545 55.; 555.55 55N.N5 555. _ m 52.55 555.5 555. _ 5 555155 55.3 555.3 555.55 55 155 555.5 55.5w 555.5N 555.5 555.5N 55 — .3 555.5 555.3 555.m~ 555.5 555. _ N 55 _ .wN 55v.vm 55— .mN 55.5.5. 555.3 555.5 55. _ m 55N.5N 555.9 555.5— 55_ .5 555.3 555.3 55.3 555.5 555.5 555.? 555. _ N 85.5 555.55 555.3 555.5. 555. _ n 55.5 555.5 5515 555.8 555.5 555.55 555.55 555.3 55.55 55.5..” _ 555.55 555.NN_ 555.5 555.55 55v.55 55m.m5_ 55min _ _ 555.5 55~.~w 555.5 _ 555. _5 555.55 555.55— 555.55 555.2 555.2 55.5— 55 _ .5N 55.55 555.3 555.5 523% 555.2 555.5 _ 555.5... 555; 5 555.3 555.2,. 555.55 555. _ ~ 555; N 55m. n 5 55 _ .mm 53.2 55 _ .3 55N.5 _ 55.5 _ 555.2 55N.v _ 555.5 _ 555.m _ 555.5 _ 555.5 _ 555.5— 555.5 _ 555.5— 55N.5_ 555.~ _ 555.5 _ 55.5— 555.5 _ 55m.” — 555.3 555.2 555.5 555.3. 555.55 55.3 555.55 555.5 55m.5m 555.5 555.5 55155 555.5— 555.9.” 555.55 555.2,. 555.5 555.5 555.3. NMVW©hw05 69 555.: a 555.5 a 555.5. a 555.55 a 555.2 a 555.: a 555.5 a _ .35. 5.58.5 3639.: .85. ems—.25 9:525 tear—2.55 :3..— .35 £80: .2..— «=8»: .3.— 33 .2230 3:25 .3.— §o 830 roar—99.5 E89? 95.539- :8 9:5 5... 33 .35..— "— «3.5m 55 _ ._ m 555.55 555.: 555.3 .5 555.3 555.3 55—.5— 555.5 m 555.5N 555.3. 551m _ 555.2” m 85.2 8%: 8a.: 83.. 0 553 m 55.3 555.2 555.9. .— 555.5N 555.5 555; _ 55 _ .3 0 555.3 53.3 53.5. 55.»— Z 53....” 5513. 555.m _ 53.5 .2 55~.5m 555.5 555.3 555.5— A 555.3 555.3 555.5— 555.5~ v— 555.5~ 555.5». 551: 555.9” _. 55 _ .5~ 555.2 555.5— 55v.5m _ 55mg 555.9. 555.3 555.3 I 55.5..” 555.2 25...” _ 555.- O 52.5..“ 555.5”. 555.5 _ 55.2” n. 555.3 55.5w 555.2 55.55. m 555.5N 555.5m 555.m _ 555.3 0 55m.- 555.5N 55.5.: 55~.vm U 55m.~m 555.3 55.2 555.3 m 555.2... a 555.55 a 555.2 555.55 a < Immunoh. .3335»:— 335 dififlh 9.320535 «an: «:30: £85 .2:— «39 .9..an 5:25 .2333 330453.555 a... :8 30.3.. 3...... .35... a 23$ 70 [Note: this is the instrument for the following conditions: positive correlations, stronger predictor is across-pool, spoilage is in the supervisory costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost - that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to higher levels of spoilage costs being recorded, because more care is taken to identify and discard defective units before they are sent to customers. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 1. The Quality Control pool includes inspection and testing costs. This pool captures the costs of ensuring that no defective products go out to customers. 2. The Supervision pool includes wages and benefits for supervisors and the costs of training programs, and also spoilage costs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 7] Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised. It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 72 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about$ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 73 @365 25.5 8a.: 8a.? 85.N— 8N6. @86— 8 ccmdv 85.3 §.MN co _ .5 23.2 81: 8 — .cN a— 8 _ .8 80.3 23.5w 25.9. 8&2 8Nd— com.NN @— 8n.55 25.3 8%: 813 813 93.2 85.mv 5— 85.3 2: .aN 89mm 806: ocmde $8.2 8&3 @— oow.5n 8n.Nm 8m.mN 8N.Nw 893 ASN.3 8m£m m— ocn€5 gméN coodn 2.de— 81m n 08.»— 8w6m 3 823 85.9 $3.0m 818— 8&3. 8&2 25.9 m— 86.50 806m oom.5m 81: 68.9 25% _ coo. _ N N _ Sndn com. _ N 895m 8‘8 81m.— coo.n_ 256m _ _ 813 cc— .wN 256m and. _ 85.5w 80.2 Sade @— oond5 $3.: 81mm 2.va 8v. _ v 25.5— ccch a 25.3. 8— .MN 256— 8m€¢ 8&3 coed. condv a $8.8 oanN con.Nm 8v. :: 23.2” 8N6. 85. _ n 5 coma» 8¢6N Smfim 85. _ m- 85.8 23.2 gnaw o comdn 895.5. SQNM c865 con. _ N 8°.»— coofim n comdm 8m. _ m 25.: @865 85. _ N 8N6. Stem v SNdm 8N.5N 8ch 8m.N: 8N. _ e 8e.» _ 25.3 m Snfio 25.9 2%.: $3.. _ 5 8 — .m N 83: coc.wN N 74 8a.: a 8o.Nm a 25.3. a 8%? a 85.2 a 25.: a Goud— a _ .55. E585 3392—2: .53. own—.25 mum-:I:. fleatunaw «an..— .2: £55 .8.— as... .2:— 33 .95an 3:30 .02— 63 830 bear—2.5 83a? name-330a .39 Se...— Sae :3 3.....— un «3.5m 2:.— m 25.2 85...... 28.3. 25.3 §.mm 2.5.2” 2%.? 2.»..— m 25.3 25.3 2:.3 25.3 25.»— 23... m 23.3 8N2” 25.2 2.x.mN 2.5.3 con.5N coflmm 2:.5N 23.5..” 8N6.“ con._N caricm 25.3 2:2” 25.2” 25.3 2%.: 25.5 256m 8m.NN ASN.3” ccm.Nm SSH 25.3 a 2%.? a ”.585 .3333..— 3.8.5 .25 2.3 .9550 .3225 23.3 25.3 .5 2:6— 2.23 m 2:.2 25.:. a 25.2 Swen o 2.3: 2.ch .— 25.: 256m 0 8N6. 259. Z §.m _ 2:.3 E 22.6— 2.23 A 23.5. 25.3 v. 2:..— ocadm .. 8n.5_ 25.3 _ 2%.: 82mm I 2&4: 8n .5 _ O 25% _ 25.3 m 8...: 23.3 m 25.2 cocdm o 2.3.— SndN U 25...: 25.3. m 25.3 8m.5n a < lounEX—m “553,—. 28.32.; «an—m 2.80: .can «39 330 E33055 5.... =8 33...... it... .35.... a .32.... 75 [Note: this is the instrument for the following conditions: negative correlations, stronger predictor is within-pool, spoilage is in the supervisory costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to lower spoilage costs by improving production quality. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 1. The Quality Control pool includes inspection and testing costs. This pool captures the costs of ensuring that no defective products go out to customers. 2. The Supervision pool includes wages and benefits for supervisors and the costs of training programs, and also spoilage costs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. . Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 76 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 77 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two Options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend an inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 78. 79 $$m. _ v $$$.VN $$0.0~ $$$.oN— $$m.50 $$N.0— 80.9. $N $$w.mv $$5.mN $$ _ .$N $$m.v$— $$ — .50 $$— .2 $$0.MN $— $$ — .wv $$0.nN $$m.NN $$v.N_ _ $$5.v0 $$N.$— $$m.5m @— $$5.N5 $$$.aN $$5.mv $$5.5_ _ $$$.mn $$m.m— $$m.wv 5— 8¢.mm $$ _ .aN $$M.v0 $$m.55 $$5.$_ $$$.M— $$0.mm 0— $$$. — 5 $$m.Nm $$n.wm $$$.? $$n.$n $$N.v_ $$m.m N n— $$m. _ a $$n.vN $$w.0m $$N.nm $$0.0N 80.2 $860 2 $$5.n 0 $$5.MN $$$.Nv $$m.0w $$5.nm $$w.m— $$w.0m m— $$N.Nm 80.9.... 80. _ N 80.2 _ $$v.00 $$$.n— $$m.5m N— 80. _ 0 Sn. _ N $$$.om $$0.0~ _ $$0.¢0 $$$.m— $$$.5m — — $$ — .55 8 — .wN $$$.ov $$o.5w $$m.NM $$0.o— $$$.0m $— $$v.mm $$V.VN $$$.aN $$5.$— — $$0.wm $$$.: $$ — .nm a $$0. _ 5 $$ — .MN $200.3. 81am $$N.mn $$$.a— $86— a $$N.$5 80.5.5. $$5. _ n $$N.mw $$n.$v $$N.$_ $$0.Nm 5 80¢” $$$.0N $$n.mm $$m. _ a $$m.¢— $$n.N— $$m.an 0 $$n.N0 $$m.5N $$$.nm $$n.w$_ $$m.wm $$$.w— $$$.NM m $$5.50 $$m. _ m $$v.0m $$n.N$— $$m.wm $$N.0— $$$.wN v 8 — .8 $$N.5N $$$.N0 $$N.00 $$w.w_ $$v.w_ $$$.aN m $$$. — n $$$.MN $$$.wN $$5.@— _ $$$.vn $$m.w— $$n.0v N 20”.”? $$$.Nm m 25.0— % $$m.$N_ a $$m.50 m $$$.: $$$.?” a _ Imaolfirll Iago“. 5.82—2.— ligflhl lama—Bum «£39.... 9.9.3.395 2:...— .8.. 3.3.5 .2... :35 .3: 33 .9555 3:30 .3..— uneu 830 fear—2.5m 38...? nut-=33- 68 SP... 8... «v.3 .92....— 3 «3an 2.3m 25.0.». 255.3 25.3 25.3 25.; 255.9” 25.3.. 25.— m 25.3 25.3 25.0». 25.3 25.3. 23.: A5109 Smdm 25.3 25.3 25.3 25.5 256m 813 25.3 25.3” 25.9 256m 25.: 2:.cm 25.3. 25.3 25.3 25.3 25.2.. 25.3 25.3 25.2” 256m 25.3 a 25.5w a Jack. 5.39%..— 3.32— .92. 33 .9530 3:30 8a.: $$$.MN .5 $$ — .0— $$$.wm m 8 — .2 $8.05. a $8.2 $$n.Nv O $$¢.0— $$$.av .— $$$._ _ $$ — .v0 0 $$N.$— $8.3 2 85.2 Siam 2 $$w.0— $$$.a— 4 $$$.: $$5.$N v— $$—.— — 80.3.. a $$n.5— $$v.5m — 80.: gm. — N I $$N.m— $$$.NN 0 $$$.n— $$$.om m 8a.: 80.5? m— $$$.m— $$$.nm G $$N.v— $$N.vm U $$m.0~ $$m.MN m $$$.: a $$n.mn m < louwaadwl. 455...; Ecatogm «an; 2.3.... .2... $3 830 3.5....an a... :8 33...... 3...... 3...... .N .22.... 8O [Note: this is the instrument for the following conditions: negative correlations, stronger predictor is across-pool, spoilage is in the supervisory costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to lower spoilage costs by improving production quality. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 11 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 3. The Quality Control pool includes inspection and testing costs. This pool captures the costs of ensuring that no defective products go out to customers. 4. The Supervision pool includes wages and benefits for supervisors and the costs of training programs, and also spoilage costs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 81 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage costjpredictions now. Once you are finished with your predictions, please turn to the next page. 82 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend an inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10, 000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 83 @235 23€N 2%.? 22.2: oomfio 8N6— 806— ON comic 85.3 23.3 2:33 2:.5 2:.2 81cm 2 8 _ .8 23.3 23.2” 23.3 295.3 8N6— oon.NN w— canfih 25.3 2%.: 25.2. 25mm 2:...“ _ 2.5.9. 5 oofig 2:.mN 82mm. 25.: 2.5.2 25.n— comsé @— comfin 8n.~m 8m.n~ $3.8. 83% 2E...— ccndm 2 Sn.: 893” 25.8 25.2: 25.3 822 oowdn v— ocmde 2:..mN 8wdm 2%.; $2.?” 23.n— 25.N¢ m _ 25.5 236m 2.me 25.8— ccvde 2&6 _ 22...: «- 2fi.wn 898 25.3“ 25.2: 25.3 256 _ 256m 2 cc _ .3 81mm 256m 25.2: 8m.~m 26.2 25.3. o— comfih 23.3.. 2.1mm 25.3w 2%.: 25.: 25.3 m 25.3. 2:.MN 22.no— condfl 8~.mm 25.0— 2%.3 m 25.8 2%.: 8n.~m 23.2: 2.3:. 2&6— 2K. _ n b 23.6w 256w Smdn 2233 25.3 892 oom.wm o 83% SnfiN 25.Nm 2K.— : 23.3 25.”— coofim n camaan 8m. _ m coo£~ 25.2 _ 8min ASN.0— covdm v ASN.3 2&5 25.3 8 _ .2: 23.n— covd— 25.3 m @350 25.8 236v 8N. _ 2 256m 22m.»— occdm N 84 25.2. m 25.2” a 25.2” a 2:.wo a ooQS a 25.3 a 25.0— m _ .55. «5.8% 5392—2.— .3:.—. oua=am Inez—7.5. region; :35 .2..— Eaeon .8.— £3:— 32— «39 .2230 3:30 .8233 £30 ffitasm 889? 25.533 .39 59: «use .39 .35..— : «Siam 2.3m 25.3 25.: 25.5 H 25.3 25.5 52.5- 25.vo m 25.5w 25.2“ 2:.2 25.; a 85mm 83.. 8a.: 8~...~ o 25.— m 25.2~ 25.5— 25.5N .— 25.5~ 22.3 25.: 256m O 25.3 25.5— 5~.2 25.9. 2 25.; 25.5w 25.2 52.3 2 535 25.5— 25.0— 25.0N A 25.3 25.3 25.2 25.2. M 525 25.2.” 2:..— 255». _. 52.3 25va 25.: 25.N~ _ 25.3“ 5n._~ 25.3 25.mn : 25.2.” 25.- 25.2 25.: O 2: .cm 25.5..” 25.2 252. m 25.v~ 25.2. 25.: 25.3 m 25.5 25.2 25.m _ 25.8 a 5m.- 25.3” 25.: 25.8 U 5m.~m 25.9 25.5— 25.nm m 25.5 a 25.3 n 25.: 25.3 a < Imam; 3.30%.: Imaml lug—:85. floatomsm Ea..— u=80= «:52— ..xx— 33 .2230 3:25 .25 63 830 Caner—2.5 a... 38 93.3.... it... .35... a .335 85 [Note.' this is the instrument for the following conditions: positive correlations, stronger predictor is within-pool, spoilage is in the quality control costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to higher levels of spoilage costs being recorded, because more care is taken to identifiz and discard defective units before they are sent to customers. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firrn’s cost accounting system reports all these costs in two cost pools for each plant: 1. The Quality Control pool includes impaction, testing and spoilage costs. This pool captures the costs of ensuring that no defective products go out to customers. 2. The Supervision pool includes wages and benefits for supervisors and the costs of training programs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 86 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage cosm for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised. It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 87 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $1 0, 000 per month to spend an inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10, 000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 88 89 con. _ v 8o.v~ 8e.e— 23...; 2K..N_ 8N.e— confiv on 25.9. 85.3 8 _ 6N Sedn 25.n— 8 1: 82mm @— 2: .wv 26.3 SQNN 25.Me 8m.n— 8Nd— 8a.?” a— 2K.Nh 25.3. 22...? 25d» 2: .eN 23.n— 8e£v h _ 233$ oo _ .aN 8m€e 25.3: gmfie 25.n— coe.mn e _ 25. :. 8n.~m 8min Sade comdm ASN.3 8m.e~ n— ccm. _ w 2K€N Swen 25.NN_ 23.9 25.”— 25.2u. v— 2K..ne 2K.MN coo.Nv 256$ 22%: 2am...”— ocw.em m— 8N.Ne coedm 8e. — N 25.8 25.n— §.e— 2.me N— eon; n com. _ N 256m 223% 23.2 25.2 25.5” _ — co _ .2. 8 _ .3. 25.3. 23.2: ocfihv coed— 25.em o— covfie SYVN 255N 23.n- _ cov. — v 25.: co _ .mn @ ace. — h 8 _ .MN comdv 2K..Ne ccw.eN 25.2 25.3 a comdb SeKN 2K. _ n com.~w confim 8N.o_ 8e.Nm h coflvw 256m Bede 2K..Nm_ 2:..oe 8n.~— 22min e 2..m.~e ocmfim 25.nm 2%.: 2%.: 25.n— 25.Nm m cchfie cem. — m 2:..em coofie 2K. _ N 8N.e_ 25ch v 8 _ .8 comKN 25.ne 25.2: 8N. _ e 23..»— 25.¢N m 25. — n 25.2 25.n~ 25$» cc — .nm 233: 23.3. N 25.5. a 25.2” a 256— % 2K.ne a 2K.N— a 25.: a Scfim a _ .83. «£595. 9.8.3035 and Ioua=eam “canon. 5592.2: 2:...— _8._ as... .25 . 2.5.5 3% .39 330 boner—2:5 32— 3.3 3.5.30 5:25 889? new—=33. 63 Se...— 8... 3.3 .83..— u— .32—um co . ._ m 25.3 25.: 25.nm H 25.: 25...... 2:6. 25.3 m 25.8 25.3. 2:...” . 25.2.. a 8.... 83.. 8o... 8.... o 22...— m 25.3 25.3 25.3 m 25.3 256m 25.. _ 2:46 O 25.3 25.9. 25.2 25.5— 2 2:..— m 2513. 25...” _ 23.3 2 25.2.. 25.3 255.3 25.2 .._ 25.2 25.2. 25.5 25.5w M 25.2 25.8 2:.— _ 8n.mm .. 52.2 25.- 25.: 25va _ 25.3” 25.9. 25.2 5n._~ I 25.2” 25.: 25.2 25.N~ O 8 _ .2” 25.2. 256 . 25.3 m 25.3 25.2” 25.: 25.5 m 25.5 25.2” 25.m . 25.3 a 8m.- 25.3 25.: 254% 0 25.2” 25.2 25.2 25.3 m 25.3 25.3 a 25.3 a 25.3 m < lmfififlh region; Ilmmu=emm 358,—. 9382.2.— .5...— n=50: 3.5.5 .3.— «39 830 Pangaea .8.— .39 .9..an 322.0 a... :8 .532... 3...... 2...... a .22.... 9O [Note: this is the instrument for the following conditions: positive correlations, stronger predictor is across-pool, spoilage is in the quality control costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to higher levels of spoilage costs being recorded, because more care is taken to identifiz and discard defective units before they are sent to customers. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 3. The Quality Control pool includes inspection, testing and spoilage costs. This pool captures the costs of ensuring that no defective products go out to customers. 4. The Supervision pool includes wages and benefits for supervisors and the costs of training programs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 91 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 92 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10, 000 per month to spend on inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $1 0, 000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 93 94 23.2. §.VN 8min? 8n.nv 2K.N— SNd— 25.0— ON 2.on 2K.MN SQMN 815v 2.5.2 2: .3 2: .ON a— 8 _ .mo 25qu 23.2” 25.3. 8m.n— SNd— 8m.NN @— oonfib 25.5N 8nd? 81mm 8 — .oN 8m.n— 2.6.9. 2 2K.Nw cc _ .aN §.mn 23.0: gmdo 25.n— 8m.$ @— ccwfin 2%.Nm 8m.nN 8N.Nw 8n.mN 8N: cerium n— 823. 2.an 22.5w 8w.wN_ 23.4% 80.2 swdm v— oondc 2K.MN cowdm 8 _ .2: 25W: coax: c2..Nv .m— gafie godm oomfim cc _ . — m 2.5.: Safi— coo. _ N N — c¢m£m con. _ N 2.5.2“ oovdo coin— 256 — 256m _ — 251% cc _ .wN 25.0w comd— _ 255.5% 25.2 255v c— oomfih SYVN 251mm 25:.» Gov. — v 25.: 2.5ch o cocdv cc _ .MN 25.2 8m.vo 25de 25.2 825. w 2560 SwahN cam.Nm 8v. _ o— 8min ch.o_ 2K. — n b coma” ooodN gnfim 2K.— m— 22.60 8m.N_ 8n.wn o oom.om confiN 25.Nm 23.3. can. _ N 25.»— 25.nm m com.on 8m. — m occ.wN 23.3. 2K. _ N 8N.e_ 236m v 8N6n ASN.2” 25.5N 23.9; ocN. — 0 22V.”— 23.Ne m 8mdc 25.2.. 23.3. cow. — b co — .nN oon— ooo£N N 25. — h a 25.NM a 625.2” a 23.3. a 255.2 a 25.: a 2&6— ” — .83. “£59; Eextomsw Jal .Iouuzam acumen. 5592......— 2:...— _8._ as»: .2..— £55 .2:— uneo 830 bear—2.3m .2:— 33 .2230 5:25 80%.? macs—.39. 33 8P: 8... 33 .83..— 3 .235— 52.5 25.3 25.: 25.5 ._. 25.3 25.5 5—6. 2560 m 25.5 25.55 52.5— 25.:. M 25.55 25.9 25.5— 255 O 25.5 25.2. 25.3 25.5 .— 25.o~ 52.3 25.— _ 25.5 0 256m 25.»— 256— 25.9. 2 25.5 25.5 25.5 _ 52.3 2 25.5 25.2 25.0— 25.5N )— 25.mm 25.5” 25.: 25.9~ M 25.5 25.55 2:..— 2565 a 2:45 25.5 25.: 25.N~ _ 25.5 25.3 25.3 25.55 m 25.5 25.3 25.5 _ 25.2 0 2: .5 25.5 25.5— 523 “— 25...~ 25.5 25.: 25.5 m 25.5 25.55 25.5— 2565 O 25.- 25.5 25.: 25.5 0 25.5 25.m~ 25.2 25.55 m 25.5 25.55 a 25.3 a 25.5 a < M5595. Egtonam Imus—3am «Eamon. 3.32.»..— 2:...— n=Soa £80: .ccfl 63 830 Egtaam .3: «So .95an 3:25 5:. «30 Eon-=5 .1552— .5aa: "N «331m 95 [Note: this is the instrument for the following conditions: negative correlations, stronger predictor is within-pool, spoilage is in the quality control costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to lower spoilage costs by improving production quality. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 1. The Quality Control pool includes inspection, testing and spoilage costs. This pool captures the costs of ensuring that no defective products go out to customers. 2. The Supervision pool includes wages and benefits for supervisors and the costs of training programs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 96 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised. It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Please complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 97 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 98 99 80 ._ 9 88.9N 80.0— 8.8N— 80.80 8N.0— 80.09 8N 80.09 88.0N 8 — .8N 80.98— 8180 8 _ .9— 880.0N 8 — 8 _ .09 80.0 N 80.NN 89.N— — 88.90 8N.8_ 80.80 0 — 888.N8 8.8N 88.09 88.8— — 88.00 80.0— 880.09 8 — 889.08 88 _ .8N 80.90 880.88 88.8— 88.0— 880.0 0 0— 88. _ 8 880.N0 880.00 888.88 880.80 88.9. 880.0N 0— 80.; 880.9N 880.00 88.08 880.0N 80.0— 888.80 9— 888.00 88.0N 8.N9 80.00 88.0 0 80.0 — 80.00 0 _ 88N.N0 80.80 80. — N 880.8— _ 89.00 88.0 _ 80.80 N— 80. _ 0 80. — N 8.80 80.0— — 80.90 8.0 — 88.80 — _ 8 _ .88 8 _ .0N 8.89 888.80 880.N0 880.8— 8.00 8 — 89.00 89.9N 8.8N 888.8: 880.00 888.8— 8100 8 80. _ 8 8 _ .0N 80.09 8180 8N.00 88.8— 88.0— 0 88N.88 80.8N 88. _ 0 8N.0 0 880.89 8N.8— 80.,N0 8 80.90 88.0N 80.00 80. — 8 880.8— 80.N_ 80.80 0 80.N0 880.8N 88.00 880.08— 80.00 8.0 _ 8.N0 0 88.80 880. _ 0 89.00 80.N8— 80.00 8N.0_ 8.0N 9 88 _ .88 88N.8~ 888.N0 88.00 880.0— 889.3 888.8N m 888. _ 0 888.0N 88.0N 888.8— _ 88.90 880.0 — 80.09 N 880.09 0 88.N0 880.0— 0 880.8N— 80.80 0 888.9— 0 888.80 0 _ .598 Mat—22.8 region—.0 '3' Iowa—3&0 "£308. 5.308.»...— 2:...— _8._ as»: .2... :35 .can 33 380 8.0330850 32:30 .8230 5530 E39? anus-S3- 030 89: 8.0 «So .83..— u— .32—um 8 _ ._ 0 88.00 80.9 — 8.0N .8 88. _ N 88.90 810— 8.00 0 888.8 880.; 82.0— 8.00 x 888.00 83.3 88.0 _ 880.N9 O 80. — 0 80.8N 880.0— 8.89 .— 888.80 888.00 88.— _ 82.90 0 8.8m 8N.09 8N8. 888.0. 2 89. _ 0 8— .09 888.0— 89.80 2 88N.80 88.0N 80.0— 88.8- x— 80.0N 8.89 888.8— 88.8N v— 80.8N 888.80 81: 80.00 a 8 _ .8N 88.NN 80.8 — 89.80 — 88N.90 8.00 80.9. 80. _ N I 80.80 80.8 _ 8N.0_ 88.NN O 8180 888.89 888.0 _ 88.80 m 80.9N 880.8N 880.9— 80.89 m 888.8N 888.80 888...: 888.00 a 880.NN 80.8N 8N.9— 8N.90 U 880.N0 888.00 880.0— 80.0N m 8.0N 880.80 a 888.9— 0 80.00 0 < 055.... .8 9.33.3850 '00an 0:23 .8 .3238»..— 2:...— aix. as... .can 33 330 8.3.2230 .3.— 63 .8230 3:25 5.... :8 3.3.... it... 3...... a .33.... 100 [Note: this is the instrument for the following conditions: negative correlations, stronger predictor is across-pool, spoilage is in the quality control costs pool. ] Background Assume you are the controller of a manufacturing firm. One of the important parts of your job is to predict future costs as part of your firm’s short-term budgeting process. One cost that you must predict for the plants in your firm is the monthly reported spoilage cost — that is, the cost of products that are discarded because they do not pass inspection. This cost must be predicted because it is not an amount that plant managers directly decide to spend, but is the result of things that happen during the month. Some (not necessarily all) of the other activities at the plant influence the amount of spoilage each month. The main activities in the plant other than direct production are inspection, testing, supervision, and training. Higher spending on one (or more) of these activities could lead to lower spoilage costs by improving production quality. No formal statistical analysis of the relationships among these costs has been performed, so it is not yet known which, if any, of the other activity costs are good predictors of spoilage costs. The plants in your region are all comparable in terms of workforce, equipment and product mix, so the relationships, if any, between inspection, supervision, training and spoilage are very similar across plants. Also, wages and input prices are similar across plants, so differences in spending levels represent differences in activity levels, not differences in prices or wages. In order to help you predict next month’s spoilage costs, your plants’ cost accountants have provided the information on Exhibit 1 from the plants’ cost accounting systems. The information includes the last month’s spoilage costs. Also included in Exhibit 1 are the other costs that could be helpful in predicting spoilage costs. Last month’s costs are typical for your firm, and should offer a good basis for predicting next month’s costs. Note that the firm’s cost accounting system reports all these costs in two cost pools for each plant: 3. The Quality Control pool includes inspection, testing and spoilage costs. This pool captures the costs of ensuring that no defective products go out to customers. 4. The Supervision pool includes wages and benefits for supervisors and the costs of training programs. This pool captures the costs of preventing poor quality work by ensuring the line employees are well trained and have adequate supervision. Costs within each pool are totaled, and only the pool total is regularly reported to most managers. The pool total is used to allocate costs to individual products and customers. Please examine Exhibit 1 and learn as much as you can about predicting spoilage costs from the other costs provided. When you are finished, please continue on to the next page. 101 Cost prediction When you are finished examining Exhibit 1, please turn to Exhibit 2. Exhibit 2 contains partial budgets for the region’s plants, as prepared by the plants’ managers. The amounts included on Exhibit 2 include the monthly quality control and supervision costs the plant managers have budgeted for next month. Based on your conversations with the cost accountants and plant managers and your review of your plants’ production plans, you believe the budgeted amounts on Exhibit 2 are reasonable. You also believe that the relationships between these costs and spoilage costs are likely to be the same in the next month as they were last month. Please use the budgeted costs on Exhibit 2 to predict monthly spoilage costs for next month based on what you have learned from Exhibit 1. Feel free to use a calculator. It is important that you try to predict costs as accurately as possible. If your predictions are inaccurate, the firm’s future profitability and your performance evaluations will suffer. Recall also that your pay is based on your predictions. Your predictions will be compared to the best predictions that can be made from the data provided. The closer your predictions to the best possible predictions, the higher your pay will be. The plant names on Exhibit 2 are disguised. It is not a good strategy to try to “match” one plant on Exhibit 2 with one or two plants on Exhibit 1. It is a better strategy to try to learn relationships among the costs and use these relationships to predict the spoilage costs. Plecgse complete the 20 spoilage cost predictions now. Once you are finished with your predictions, please turn to the next page. 102 Advice to top management Now that you have learned about the relationships among the various costs and have made your predictions, top management has asked you for your advice on a decision. Because of recent concerns about product quality, top management would like to increase its spending on quality-improvement initiatives. However, top managers are unsure about how to make the greatest impact on quality-related costs. Thus, they are considering two different options. Note that these two options are mutually exclusive — that is, management will pursue only one of them, but not both. For help in deciding which option to pursue, management has asked you for an estimate of the effect of each of the two options. Your estimate should be based on the relationships you have learned from Exhibit 1. Please feel free to refer to Exhibit 1 when making these estimates. Option 1: What would be the change in spoilage costs if management gave each plant an additional $10, 000 per month to spend an inspection? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Option 2: What would be the change in spoilage costs if management gave each plant an additional $10,000 per month to spend on supervisors? There would be no change in any other spending levels. Each plant’s monthly spoilage costs would be expected to (circle one) increase or decrease by about $ Thank you very much. Now, please return this packet to the researcher, and pick up your second packet. 103 104 889.88 888.9N 880.09 88 _ .88— 880.88 88N.8— 888.8— 8N 880.89 888.0N 888.0N 880. ~8— 88 _ .88 8819— 88 — .8N 8— 88 _ .08 888.0N 880.80 889.88 888.98 88N.8— 880.NN 8 _ 880.88 888.8N 880.89 888.N— ~ 88.00 880.0— 888.09 8 — 888.N8 88 — .8N 888.00 888.88 888.8— 888.0— 880.98 8 _ 888.80 880.N0 880.0N 88N.08— 880.80 88N.9— 880.80 0— 880 .98 880.9N 888.80 888.N8_ 888.8N 888.8— 888.80 9 _ 880.88 888.0N 888.80 880. ~ 8 888.0 0 888.0— 888.N9 0 _ 888.88 888.80 880.80 888.08— 889.88 888.0— 888. —N N _ 880.80 880. _ N 88.80 888.88— 888.98 888.0— 888.80 _ — 88 ~ .98 88 _ .8N 888.80 888.88— 880.N0 888.8— 888.89 8 _ 880.88 889.9N 88— .00 888.98 888.80 888.8— 888.8N 8 888.89 88~ .0N 888.8— 888.8N— 88N.00 888.8— 880.89 8 888.88 880.8N 880.N0 889.N8— 880.89 88N.8_ 888. — 0 8 880 .0 8 888.8N 880.80 880.88 880.8— 880.N_ 880.80 8 880.80 880.8N 888.N0 880._ : 880.80 888.8— 888.00 0 880.80 880. — 0 888.8N 888.8 — _ 880.80 88N.8— 889.80 9 88N.80 88N.8N 888.8N 88188— 888.8— 889.8— 888.N8 0 880.88 888.0N 880.89 88N. _ 8 _ 888.90 88 0.8 — 888.8N N 888. _8 8 88.N0 8 888.80 8 88 _ .88 8 880.88 8 888.9— 8 888.8— 8 _ .898 “£538. £83395 Jflmbl low—3.5 8.53.8 .8239...— 2:...— _8.. 2...... .2... as... .3.— .39 smoUfleaEém .2:— «39 .9...an 5:30 8390» 8.55.39. «39 :3...— Sne «so 3.....— .— .32—um 88...0 888.0N 888.9. 888.80 ... 888..N 888.80 88.6. 888.98 8 888.8N 888.00 88..0. 888.—9 8 888.0 0 880.N9 888.0. 88N.9N O 880..0 888.89 888.8. 888.8N .. 888.8N 88. .98 888.. . 888.80 0 888.8N 888.8. 88N.8. 88N.09 Z 889..0 889.80 888.0 . 88.69 2 88N.80 888.8. 888.8. 888.8N .. 880.00 888.8N 888.8. 88.89 V. 880.8N 880.00 88.... 888.80 .. 88..8N 889.80 880.8. 888.N~ . 880.90 880..N 880.9. 888.00 .. 880.80 888.NN 88.0. 880.8. 0 88.80 888.80 888.0. 888.89 .. 888.9.8. 880.89 888.9. 880.88. m 888.8N 888.0 0 888.0. 88.80 D 880.NN 880.90 880.9. 880.8N U 880.N0 880.0N 880.8. 888.00 0. 888.8N 880.00 a 888.9. a 880.80 a 9. «£59.... 9.83.2.5 own—.888 8.58... 5.82.2.— 2...... 3.80: 3.308 .92. 33 830 0.3.»..038 .3.— 33 .9530 03.2.0 3... =8 3.3.... it... .35.... a .32.... 105 REFERENCES Anderson (1990). Cognitive Psychology and its Implications, Third Edition. New York: W. H. Freeman and Company. Anderson, S. W. (1995). Measuring the impact of product mix heterogeneity on manufacturing overhead cost. The Accounting Review, 70(3), 363. Anonymous (2006). The data is the data. Business Performance Management Magazine. Vol. 4, Iss. 1, p. 36 Banker, R. D. and S. M. Datar. (1989). Sensitivity, precision, and linear aggregation of signals for performance evaluation. Journal of Accounting Research (Spring): 21- 39. Banker, R., and H. Johnston (1993). An empirical study of cost drivers in the US. airline industry. The Accounting Review, 68(3), 576. Banker, Rajiv D., G. Potter, and R. Schroeder. (1995). An empirical analysis of manufacturing overhead cost drivers. Journal of A ccounting & Economics 19 (February):l 15-137. Billman, D., B. Bomstein, and J. Richards. (1992). Effects of expectancy on assessing covariation in data: “Prior belief” versus “meaning.” Organizational Behavior and Human Decision Processes 53, 74-88. Brehmer, B. (1974). Hypotheses about relations between scaled variables in the learning of probabilistic inference tasks. Organizational Behavior and Human Decision Processes, 11(1), 1-27 Briers, M., C. Chow, N. Hwang and P. Luckett. (1999). The effects of alternative types of feedback on product-related decision performance: a research note. Journal of Management Accounting Research 11, 75-92 Bromwich, M., and C. Hong. (1999). Activity-based costing systems and incremental costs. Management Accounting Research 10 (1), pp. 39-60 Broniarczyk, S. M. and J. Alba. (1994), Theory versus data in prediction and correlation tasks. Organizational Behavior and Human Decision Processes, 57, 117-139. Christensen, J. and J. Demski. (1995). The classical foundations of "modem" costing. Management Accounting Research, 6(1), 13. 106 Christensen, J. and J. Demski. (1997). Product costing in the presence of endogenous subcost functions. Review of Accounting Studies 2 (1), 65-87. Cooper, R. and R. Kaplan (1999) Design of Cost Management Systems: 2nd Edition New Jersey: Prentice Hall Cooper, R., R. Kaplan, L. Maisel, E. Morrissey, and R. Oehm. (1992). Implementing Activity-Based Costing: Moving from Analysis to Action. Montvale, NJ: Institute of Management Accountants Datar, S., M. Gupta, (1994). Aggregation, specification and measurement errors in product costing. The Accounting Review, 69(4), 567. Datar, S. M., S. Kekre, T. Mukhopadhyay, and K. Srinivasan. (1993). Simultaneous estimation of cost drivers. The Accounting Review, 68(3), 602. Dawes, R. M. and B. Corrigan (1974). Linear models in decision making. Psychological Bulletin, 81, pp. 95-106 Drake, A., S. Haka, and S. Ravenscrofi (1999). Cost system and incentive structure effects on innovation, efficiency and profitability in teams. The Accounting Review 74 (3): 323-345. Farrell, A., J. Lufi, and M. Shields. (2006) Accuracy in judging linear and nonlinear effects of cost and profit drivers. Contemporary Accounting Research, forthcoming. Foster, G. and M. Gupta. (1990). Manufacturing overhead cost driver analysis. Journal Of Accounting & Economics, 12(1—3), 309. Fox, J. (1991). Regression Diagnostics. Thousand Oaks, CA: Sage Publications. Garrison, R., E. Noreen, and P. Brewer. (2006). Managerial Accounting. 11th ed. New York: McGraw-Hill. Gupta, M., and R. King (1997). An experimental investigation of the effect of cost information and feedback on product cost decisions. Contemporary Accounting Research, 14 (1) 99-127. Hopkins P. (1996). “The Effect of Financial Statement Classification of Hybrid Financial Instruments On Financial Analysts’ Stock Price Judgments” Journal of Accounting Research (Supplement): 33-50 Horngren, C., S. Datar, G. Foster. (2002) Cost Accounting: A Managerial Emphasis, 11th ed. New Jersey: Prentice Hall 107 Ittner, c., Larcker, D, and M. Meyer. (2003). Subjectivity and the weighting of performance measures: Evidence from a balanced scorecard. The Accounting Review, 78 (3), 725 Ittner, C., D. Larcker, and T. Randall. (1997). The activity-based cost hierarchy, production policies and firm profitability. Journal of Management Accounting Research 9: 143-162. Joshi, S., R. Krishnan, and L. Lave. (2001). Estimating the hidden costs of environmental regulation. Accounting Review 76(2), 171-198 Kaplan, R. S., and R. Cooper. (1998). Cost and Effect: Using Integrated Cost Systems to Drive Profitability and Performance. Boston: Harvard Business School Press. Kaplan, R. S. and D. P. Norton. (1996). The Balanced Scorecard: Translating Strategy into Action. Boston: Harvard Business School Press. Lufi, J. and M. Shields. (2001). Why does fixation persist? Experimental evidence on the judgment performance effects of expensing intangibles. The Accounting Review 76(3): 561-587. Markman, A.B., & Gentner, D. (2001). Thinking. Annual Review of Psychology, 52, 223- 247. McKinnon, S. and Bruns, W. (1992). The Information Mosaic. Boston: Harvard Business School Press Muchinsky, P. and A. Dudycha. (1974). The influence of a suppressor variable and labeled stimuli on multiple cue probability learning. Organizational Behavior and Human Decision Processes, 12(3), 429-444. Muchinsky, P. and A. Dudycha. (1975). Human inference behavior in abstract and meaningful environments. Organizational Behavior and Human Decision Processes, 1 3(3), 377-391 . Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289-316. Nisbett, R., and T. Wilson. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review 84, 231-259. Noreen, E. (1991). Conditions under which activity-based cost systems provide relevant costs. Journal of Management Accounting Research 3, p. 159-168. 108 Roth, H. and Borthick, A. F. (1991). Are you distorting costs by violating ABC assumptions? Management Accounting 73 (5) 39-42. Sharman, P. A. (2003). Bring on German cost accounting. Strategic Finance 85 (6) 30- 38. Sniezek, J .A. (1986). The role of labels in cue probability learning tasks. Organizational Behavior and Human Decision Processes 38, 141-161. Yamauchi, T., & Markman, AB. (2000). Inference using categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 776-795. Vera-Munoz, S. C., Kinney, W., and Bonner, S. (2001). The effects of domain experience and task presentation format on accountants information relevance assurance. The Accounting Review 76 (3), 405-429. 109 IIIIIIIIIIIIIIIIIIIIIIIIIIIIII Will] MlljllllillllHfllllflllflllll 1‘ il ill 93 295 7