Addressing Selection Bias Attributable to Dropped Observations in Event Studies and Missing Data in Studies Using large Commercial Databases

We know smaller firms are underrepresented in many databases. This study shows that when the impact of the variable of interest (VOI) differs across firm size, the use of size control variable does not address selection bias issues associated with disproportionate representation of different sized firms in the sample. Similarly, when researchers delete contaminated observations in event studies, big firms become underrepresented, thereby resulting in similar selection bias issues. We recommend routine use of interaction term between VOI and firm size. This is an easily implementable, widely understood, simple and practical solution that allows for statistically correct and reliable inferences.