Of course, managers try to improve on random choices and make the best decision contingent on the constraints and opportunities facing the firm, and their efforts tend to trip up the researcher who would "second-guess" the policy choice. Policy A might be the best for a firm stuck with an inferior set of opportunities; policy B may be a viable but not ideal choice for a firm blessed with good opportunities. The investigator who does not control for the firms' different opportunities concludes that policy B beats policy A, whereas the shortchanged firms are those that chose B. The more canny researcher tries to duck the problem by employing controls for differences in the decision maker's opportunity by means of a set multivariate model of the determinants of performance. That move will often not suffice, however, because the policy choice is dependent on, and hence correlated with, the indicators of the business's opportunities, making it unlikely that the researcher can estimate their separate influences precisely. In statistical estimation multicollinearity inflates standard errors and, in practice, makes the results frustratingly sensitive to small changes in the measurement of the variables and/or specifications of the model. Independent of this problem, any random disturbance or uncontrolled structural condition that affects both choice of policy and performance will bias the estimated effect of the policy on performance. Solutions to these problems do exist, for example, a two- pass procedure in which the researcher first relates the sampled businesses' policy choices to their opportunity sets. That yields a prediction of how apt each choice was, given the decision maker's opportunities and the average choice pattern of all decision makers in the sample. The second step is then to test whether policies that appear well matched to the firm's opportunities outperform those that are mismatched. This design unfortunately is hard to implement in research on international business, where firms' opportunity sets are hard to define and characterize accurately and likely to be highly heterogeneous. The design also depends, of course, on firms not doing too good a job in making the observed choices. If no mismatches occur in the observed sample, the researcher hits a blank wall, because the differential effect of superior over inferior policy choices by identically situated firms cannot be observed. That is, all business decisions are correct, but the researcher has no way to test and confirm this.