Information Technology and Firm Profitability: Mechanisms and Empirical Evidence

Table A2 shows the results of the Nijman-Verbeek test for sample selection in panel models. This test examines the possibility of selection bias in unbalanced panel data. We construct two indicator variables. First, the lagged selection indicator “inlastyr” indicates that a firm present in year t of the sample period is also present in year t – 1. Second, the forward selection indicator “innextyr” indicates that a firm present in year t of the sample period is also present in year t + 1. If profitability is biased by the number of times a firm is present in the sample, these selection indicators will be significant when included in the main panel regression models. The forward selection indicator is particularly useful to test the existence of bias due to attrition, while the lagged selection indicator is useful to test the existence of bias due to the possibility of systematic differences among firms that appear in the data set for the first time. The results of the Nijman-Verbeek test show no evidence for selection bias due to the structure of the unbalanced panel, as is evident in the nonsignificant coefficient estimates of the lagged and forward selection indicators. Columns 1 and 3 show fixed and random estimations, respectively, when we include the lagged selection indicator “inlastyr” in the model. This suggests that inclusion of the firm in the previous period has no significant effect on profitability, suggesting that selection bias in the unbalanced panel is not a problem. Columns 2 and 4 use a lead of the selection indicator “innextyr.” Nonsignificance of coefficient estimates of the lead select indicator suggests that attrition is not a source of bias in the estimates.