Most studies of organizational performance define performance as a dependent variable and seek to identify variables that produce variations in performance. Researchers who study organizational performance in this way typically devote little attention to the complications of using such a formulation to characterize the causal structure of performance phenomena. These complications include the ways in which performance advantage is competitively unstable, the causal complexity surrounding performance, and the limitations of using data based on retrospective recall of informants. Since these complications are well-known and routinely taught, a pattern of acknowledging the difficulties but continuing the practice cannot be attributed exclusively to poor training, lack of intelligence, or low standards. Most researchers understand the difficulties of inferring causal order from the correlations generated by organizational histories, particularly when those correlations may be implicit in the measurement procedures used. We suggest that the persistence of this pattern is due, in part, to the context of organizational research. Organizational researchers live in two worlds. The first demands and rewards speculations about how to improve performance. The second demands and rewards adherence to rigorous standards of scholarship. In its efforts to satisfy these often conflicting demands, the organizational research community sometimes responds by saying that inferences about the causes of performance cannot be made from the data available, and simultaneously goes ahead to make such inferences. We conclude by considering a few virtues and hazards of such a solution to dilemmas involving compelling contradictory imperatives and the generality of the issues involved.