Working paper
On the Use of Absolute Residuals as Dependent Variables in Accounting Research
SSRN
10/10/2023
DOI: 10.2139/ssrn.4572590
Abstract
We analyze and discuss the regression procedure using the absolute values of residuals generated from a first-step OLS regression as the dependent variable in a second-step regression. This procedure is used in numerous accounting settings, particularly in studies of discretionary accruals and investment (in)efficiency. Due to the absence of an accepted alternative estimation procedure, the use of absolute residuals as dependent variables remains prevalent in the accounting literature. We first show that the use of absolute residuals can produce both inflated and attenuated coefficient estimates and can impair inferences and we demonstrate the potential magnitude of the bias in two accounting settings. Next, we provide two solutions to avoid the bias associated with using absolute residuals as dependent variables. Finally, we discuss the problems with a new regression procedure that uses the absolute value of the first-step dependent variable (e.g., the absolute value of accruals) as the dependent variable in a single-step regression.
Details
- Title: Subtitle
- On the Use of Absolute Residuals as Dependent Variables in Accounting Research
- Creators
- Wei Chen - University of ConnecticutPaul Hribar - University of IowaSam Melessa - University of Iowa
- Resource Type
- Working paper
- DOI
- 10.2139/ssrn.4572590
- Publisher
- SSRN
- Number of pages
- 58 pages
- Language
- English
- Date posted
- 10/10/2023
- Academic Unit
- Accounting
- Record Identifier
- 9984476570902771
Metrics
114 Record Views