اثرات قضاوتی در نقشه های استراتژی در ارزیابی عملکرد کارت امتیازی متوازن

We examine whether supplemental information displays affect decisions made using a common strategic performance measurement system, the balanced scorecard. A distinguishing feature of the balanced scorecard (BSC) is the number and diversity of its metrics. To effectively formulate a decision from such a complex information set, managers must view these measures within their strategic context (Kaplan and Norton, 1993 and Kaplan and Norton, 1996). However, academic studies indicate that problems in communication and comprehension of the strategic logic underlying the scorecard hinder its implementation and use (Lipe and Salterio, 2000, Malina and Selto, 2001, Ittner et al., 2003a and Ittner et al., 2003b). We investigate whether a supplemental information display, in the form of a strategy map, results in performance evaluation judgments consistent with the recognition of relations between performance metrics and strategy. Strategy maps are causal diagrams depicting temporally-separate and non-linear relations between scorecard performance measures and overriding strategic objectives. As predicted, we find that performance evaluation decisions are more consistent with the achievement of strategic objectives when participants are provided with strategy maps.

This study investigates whether supplemental information displays, in the form of strategy maps, affect decisions made using a common strategic performance measurement system, the balanced scorecard (BSC).3 A strategic performance measurement system is a set of nonfinancial and financial objectives and performance measures representing a causal chain of activities that articulates management's hypothesis of strategy (Epstein and Manzoni, 1997). A distinguishing feature of the BSC is the number and diversity of its metrics: BSCs contain sixteen or more leading and lagging measures that capture performance along multiple dimensions (customer relations, internal processes, organizational learning and growth, and financial). The BSC's inherent complexity creates difficulties in communication and comprehension of its underlying logic which hinder implementation and use (Lipe and Salterio, 2000, Malina and Selto, 2001, Ittner et al., 2003a, Ittner et al., 2003b, Ittner and Larcker, 2003, Banker et al., 2004 and Dilla and Steingbart, 2005). Research on causal modeling suggests that strategy maps can simplify and facilitate the transmission of complex systems, and thus implies that strategy maps have the potential to help decision makers overcome the cognitive challenges posed by the BSC (Fiol and Huff, 1992 and Vera-Muñoz et al., 2007). This approach is consistent with studies in accounting, information systems, and psychology that find that altering or supplementing the manner in which information is presented can improve performance on decisions requiring complex judgments (e.g., Vessey, 1991, Tuttle and Kershaw, 1998, Lipe and Salterio, 2002 and Dilla and Steingbart, 2005). Accordingly, the purpose of this study is to investigate the impact of supplemental strategy maps on BSC decisions using an experiment. Strategy maps are causal maps depicting relations between BSC performance measures and overriding strategic objectives. Strategy maps can aid managerial decisions if they enable managers to assess a measure's relative importance to the achievement of strategic goals (i.e., linkage to strategy) and thus provide cues for managers to weight and aggregate BSC measures in formulating an overall decision. Kaplan and Norton, 2000, Kaplan and Norton, 2004 and Kaplan and Norton, 2006 instruct managers to use strategy maps to communicate the BSC; and, several companies use strategy maps with the BSC ( Mair, 2002, Kaplan and Norton, 2004, Kaplan and Norton, 2006, Urrutia and Eriksen, 2005 and Veth, 2006). Commercial scorecard applications, such as Oracle Hyperion Performance Scorecard, even produce strategy maps.4 Research in information systems and management has long recognized the value of causal maps in structuring complex problems (e.g., Axelrod, 1976, Eden et al., 1992 and Fiol and Huff, 1992). Causal maps help individuals to construct more accurate mental models of complex systems (Johnson-Laird, 1983), and experimental research suggests that decision makers perform better when their mental models are more similar to the external systems they represent (e.g., Wyman and Randel, 1998, Davis and Yi, 2004 and Capelo and Dias, 2009). Our experiment, which is based upon Banker et al. (2004), requires participants to use the BSC to evaluate managers in two different business units of the same company. Each business unit has its own 16-measure BSC. The measures differ regarding their relationship to business unit strategy: some are causally linked to strategy (linked) and some are not (non-linked). Banker et al. (2004) investigated whether decision makers relied more on strategically linked measures than non-linked measures or measures that are common across scorecards.5 They manipulated detailed strategic information at two levels: (1) no strategic information (Benchmark Treatment), and (2) a narrative on strategy and a strategy map (Strategy Information Treatment). Our research question and experimental design differ from Banker et al. (2004) in that we investigate whether supplementing narrative strategic information with a strategy map leads to decisions that are more consistent with strategic objectives. Thus, we add a third treatment in addition to the two treatments used in Banker et al. (2004). Participants in this third treatment receive narrative strategy information, but not a strategy map (Narrative Treatment). This enables us to investigate whether a strategy map, which is informationally equivalent to the narrative regarding unique BSC measures, improves decision-making. Consistent with our predictions, we find that participants that receive both a strategy map and a narrative on strategy rely more on strategically linked measures than those that receive only a narrative. This result suggests that strategy maps help managers view performance measures within their strategic context, and thereby, reduce the cognitive difficulty of using the BSC for performance evaluation. This research makes three contributions to the literature. First, it extends prior studies that investigate methods to improve managers' ability to effectively use the BSC framework to make decisions (Banker et al., 2004, Libby et al., 2004, Webb, 2004 and Dilla and Steingbart, 2005). In particular, our findings suggest that Banker et al. (2004) would not have obtained the results they did without the inclusion of a strategy map. Accordingly, this study is of interest to the many managers who use scorecards and other cognitively demanding strategic performance measurement systems. Second, our results indicate that a causal map facilitates the transmission of complex ideas from individual to individual, and hence support a map's decision usefulness for individuals other than those who formulate it. Thus our results support causal mapping as a communication tool, in addition to a problem-structuring tool. Finally, our results are consistent with studies in the information systems literature that find decision makers must understand an information system to fully reap its benefits (e.g., Davis, 1989, Davis et al., 1989, Subramanian, 1994 and Venkatesh and Davis, 2000). The next section reviews related literature and presents the research hypothesis. We describe our experimental design in Section 3 and present results in Section 4. Conclusions are discussed in Section 5.

4.1. Manipulation checks After completing the performance evaluation exercise, we asked participants to rate, on an 11-point scale centered at zero, their agreement with seven statements assessing their understanding of the performance measures and the task (Table 2). Statements one through four address performance measure categorization and appropriateness. Manipulation checks show that participants believed that “measures were usefully categorized,” that the “emphasis on financial measures was appropriate,” that the two SBUs “used some different performance measures,” and that use of different performance measures was “appropriate” (p < 0.01). Agreement ratings for participants in the Strategy Map Treatment are significantly higher than ratings of participants in the Benchmark Treatment for these four statements (p < 0.05, p < 0.01). Agreement ratings for the Strategy Map Treatment significantly exceed ratings for the Narrative Treatment regarding the usefulness of performance measure categorization (p < 0.01) (Statement 1) and the appropriateness of SBU-specific measures (p < 0.01) (Statement 4), suggesting that participants who received strategy maps were better able to evaluate performance measure characteristics. Analysis of the last three statements indicates that all participants found the case easy to understand, not too difficult to complete, and realistic (p < 0.01). Additionally, responses to the last three statements do not differ across treatments4.2. Preliminary analysis Our hypothesis is predicated on the assumption that strategy maps enhance strategy comprehension; therefore, we tested this assumption by asking participants to rate their agreement with four statements designed to capture the essential elements of each SBU's strategy.7Table 3 contains the strategy statements and the average agreement rating for each statement by treatment. Agreement ratings for the Strategy Map Treatment significantly (p < 0.01 to p < 0.10, one-tailed) exceed those of the Benchmark Treatment for all four questions. For three of the four questions, agreement ratings from the Strategy Map Treatment also significantly (p < 0.01 to p < 0.10, one-tailed) exceed those of the Narrative Treatment. We factor analyzed the four statements to develop a single measure of strategy comprehension, denoted Strategy Comprehension.8 Consistent with our expectations, Strategy Comprehension for the Strategy Map Treatment (mean = 3.56) significantly exceeds (p < 0.01, one-tailed) Strategy Comprehension for the Narrative Treatment (mean = 3.03). Thus, univariate tests are consistent with the expectation that participants who received strategy maps had a better understanding of strategy than those who did not.Following Banker et al. (2004), we also compare strategy comprehension across treatments after controlling for demographic variables that may affect participants' understanding of strategy, such as age, gender, U.S. residency, and work experience. We perform this comparison by regressing Strategy Comprehension on indicator variables for the Benchmark and Narrative Treatments and a set of demographic control variables (Table 4). The omitted treatment is the Strategy Map Treatment, thus coefficients on the Benchmark and Narrative Treatment indicator variables (Treatment B, Treatment SM) estimate the effect that the information provided in these treatments has on strategy comprehension relative to information provided in the Strategy Map Treatment. The negative and significant (p < 0.01, one-sided) coefficients on both of the treatment variables indicate that students in the Narrative and Benchmark Treatments did not understand strategy as well as those in the Strategy Map Treatment. An F-test of the control variables indicates that we can reject the hypothesis that none of the control variables differs significantly from zero (p = 0.02); hence, inclusion of the control variables provides a more precise estimate of the treatment effects. Accordingly, we include control variables in tests of our hypothesis.4.3. Hypothesis tests We hypothesize that participants in the Strategy Map Treatment will place greater (less) weight on strategically linked (non-linked) measures in evaluating performance than those in the other treatments. Recall that the dependent variable is the difference in evaluation scores assigned to each manager by same participant, denoted (Wi − Fi). To investigate the hypothesis, we compare ratings differences for participants who received the same BSC combination, but were in different information treatments. The first set of tests provides simple, cross-sectional comparisons that are easy to interpret and have directional significance. In Table 5, we compare the mean impact of linked and non-linked measures across treatments. The mean impact of linked measures (Table 5, Row C) is calculated by subtracting the mean ratings differences for BSC combinations in which The Women's Store is not favored on any linked measures (Row B, BSC Combinations 3 and 4) from the mean ratings differences for combinations in which The Women's Store is favored on all linked measures (Row A, BSC Combinations 1 and 2).9 As shown in Column 4 of Table 5, the mean impact of linked measures in the Strategy Map Treatment (2.233) marginally exceeds (p = 0.086, one-tailed) the impact of linked measures in the Benchmark Treatment (1.500). The mean impact of linked measures in the Strategy Map Treatment also exceeds (p = 0.130, one-tailed) that of linked measures in the Narrative Treatment (1.633), but this difference is not statistically significant. The mean impact of non-linked measures is contained in Row F. Consistent with the hypothesis, the mean impact of non-linked measures in the Strategy Map Treatment (0.433) is less than that of non-linked measures in the Narrative (0.567) and Benchmark (1.100) Treatments; however, the difference is not statistically significant.We also use multivariate analysis to test our hypothesis because simple means tests do not measure interaction effects or control for factors outside of the single cross-section of interest. Repeated measures ANOVA allows for interactions among scorecard combinations and information treatments; however, ANOVA cannot accommodate continuous variables capturing important participant characteristics that preliminary tests have shown to be significant. Following Banker et al. (2004), we overcome this limitation by using regression analysis to compute the mean effects of linked and non-linked measures.10 We regress the dependent variable, Wi − Fi, on indicator variables that capture the four scorecard combinations, the three information treatments, all possible higher order interactions between measure type and treatment, and a set of control variables that capture participant-specific characteristics (Table 6, Eq. (2)). The regression model has moderate explanatory power (adjusted-R2 = 22.07) and is statistically significant (p < 0.01). An F-test of the model rejects the null hypothesis that all of the coefficients on the control variables equal zero (p < 0.05). In particular, a control variable for gender is positively and significantly (p < 0.01, two-tailed) associated with the ratings difference. Also, coefficients on control variables identifying participants with retail experience and those who have lived in the U.S. for at least five years are both statistically significant (p < 0.10, two-tailed). Hence, including the control variables in the analysis yields more precise estimates of the impact of measure type and treatment on performance evaluation.To test the hypothesis, we first compute the mean impact of linked and non-linked measures in each treatment using coefficients estimates and predicted values from the regression model in Eq. (2) of Table 6. Coefficient estimates are obtained by running the model with 180 observations obtained from the 180 experiment participants. Using these coefficient estimates, the mean impact is calculated by taking the difference between the value of the regression expression when the variable of interest (Linked or Non-linked) equals one and when it equals zero, calculated at the mean values of other variables (e.g., Govindarajan and Gupta, 1985).11 For example, to calculate the estimated mean impact of linked measures in the Strategy Map Treatment, denoted Impact LSM, we estimate the regression when the Women's Store dominates on linked measures in the Strategy Map Treatment (Link-WS = 1, Treatment SM = 1) and when the Women's Store does not dominate on linked measures in the Strategy Map Treatment (Link-WS = 0, Treatment SM = 1). The difference in these estimates is the mean impact of linked measures for participants with strategy maps. The mean impact of linked and non-linked measures is similarly calculated for the Benchmark and Narrative Treatments, and mean impacts are compared across treatments. Consistent with our hypothesis, the estimated mean impact of linked measures in the Strategy Map Treatment (2.307) statistically significantly exceeds the corresponding mean impact in the Benchmark Treatment (1.489, p = 0.043, one-tailed) and the Narrative Treatment (1.484, p = 0.041, one-tailed). Also consistent with the hypothesis, the mean impact of non-linked measures is smaller in the Strategy Map Treatment (0.503) than it is in the Benchmark Treatment (1.136) and Narrative Treatment (0.670). However, only the difference between the Strategy Map and Benchmark Treatments is not statistically significant (p = 0.084, one-tailed). Together, the results indicate that participants receiving strategy maps appropriately placed significantly more weight on linked measures, but they did not place significantly less weight on non-linked measures.