1. To investigate our predictions, we replicate and extend the task developed by Luft and Shields (2001; hereafter L&S)13 for three reasons. First, the L&S task includes complex components but is also intuitive. In particular, the L&S task is based on a real-life profit-forecasting task that avoids the simplistic concerns outlined by Bonner and Walker (1994) because (1) the L&S task does not contain perfectly predictable outcomes, (2) the L&S task contains four decision cues across 20 observations, and (3) the decision cues are not universally diagnostic.14 While prior feedback studies include some of these complex components, we note that the L&S task is relatively unique in that there is a lagged relationship between the decision cues and profit outcomes, adding an extra level of difficulty to the task investigated in this study. A second benefit of investigating the L&S forecasting task is a clear opportunity for improvement (as implied by the experimental results of L&S). Finally, we are able to extend the original L&S design by one period in order to include
repetition-based OFB.15
2. In the L&S setting, research participants were informed that a multi-plant manufacturing company had recently implemented a quality improvement program. The company wanted to determine what effect, if any, discretionary quality improvement spending would have on the gross profits of each plant. Each plant produced the same product using the same technology and participants were told that quality improvement effects should be comparable across plants.
3. Participants were provided with learning data that included historic quality improvement spending (four consecutive quarters) and actual profit from the most recently completed quarter. Each participant viewed results from 20 company-owned manufacturing plants.16 After studying the learning data, participants received quality improvement spending data for 20 additional (but very similar) plants within the same company. Participants were asked to predict gross profit for each plant (as shown in Appendix A). All of the experimental materials used in the learning phase and in the first task iteration (comprising the replication) are identical to the materials used by L&S.17
4. We add a second task iteration in order to investigate the effects of OFB and incentives. As described below, we use a between-participants design to manipulate financial incentives and the presence of OFB. In order to illustrate the similarity of data provided to participants, Appendix B shows the correlation matrices for each of the three data sets utilized in this study. As in the L&S data sets, actual gross profit (i.e., the realization of participants’ predictions) is most closely correlated with the quality improvement spending in the earliest quarter presented (three quarters prior to the most recently completed quarter). The three-quarter lagged quality improvement spending and current gross profit have a correlation of 0.90 in the learning data, 0.95 in the first task iteration, and 0.91 in the second task iteration (see Appendix B). In all three data sets, the three quarter lagged quality improvement spending is significantly associated with realized profit (p ,0.05), but correlations between profit and quality improvement spending in the other quarters (as well as the correlations among the separate quality improvement spending amounts) are not statistically different from zero. Therefore, participants who learn the lagged relationship between quality improvement expenditures and current period profits should outperform participants who do not learn the lagged relationship.