This study investigates coefficient bias and heteroscedasticity resulting from scale differences in accounting levels-based research designs analytically and using simulations based on accounting data. Findings indicate that including a scale proxy as an independent variable is more effective than deflation at mitigating coefficient bias, even if the proxy is 95 percent correlated with the true scale factor. In fact, deflation can worsen coefficient bias. Also, deflation often does not noticeably reduce heteroscedasticity and can decrease estimation efficiency. White (1980) standard errors are close to the true ones in regressions using undeflated variables. Replications of specifications in three recent accounting studies confirm the simulation findings. The findings suggest that when scale differences are of concern, accounting researchers should include a scale proxy as an independent variable and report inferences based on White standard errors.