Misclassification Bias in Strategy Research: Analysis and Application

Many issues in strategy and technology management are categorical. Errors in classification of the dependent variable cause systematic bias in regression estimates. In the context of the linear probability model, we characterize the misclassification bias as being proportional to the difference between two measures. One is the “bias” in the explanatory variable, conditional on false negative, multiplied by the probability of false negative. The other is the “bias” in the explanatory variable, conditional on false positive, multiplied by the probability of false positive. Further, the misclassification bias decreases with the variance of the explanatory variable. We apply the theoretical proposition to an analysis of the effect of human capital on the mobility of scientists and engineers. We show that LinkedIn profiles provide more accurate employment histories than patent records. The difference in the estimated coefficients between regressions using employment histories from LinkedIn and patent records exactly equals the bias predicted by our proposition. ∗Corresponding author: Ivan Png, NUS Business School, National University of Singapore, [email protected].