Reverse causality: change and dynamic modelsFirst, Liker et al. (1985)and Finkel (1995)suggest that change-on-change models have substantial advantages over cross-sectional data in improving causal inferences. As such, I examine whether changes in facilitation payments (∆Facilitation Payment) are associated with changes in discretionary accruals (∆|DA|). Column (1) in Panel A of Table 5pertains to change-on-change models. The results report a positive and significant relationship between changes in facilitation payments and changes in discretionary accruals, confirming the causal impacts of facilitation payments on earnings management levels.8Second, McKinnish (2000), Hahn et al. (2007), Jayaraman and Milbourn (2012), and Burke and Emerick (2016)imply that long-difference change estimators are superior to both the panel and dynamic models in dealing with reverse causality. Consistent with McKinnish (2000)and Jayaraman and Milbourn (2012), I present change-on-change models over a 3-year interval in column (2) of Panel A, where changes variables are calculated between year tand t-3. The results remain robust to the long-differencing estimation.