Estimation results in linear regression models are sometimes in contrast with what was expected on the basis of a certain set of hypotheses or theory, in the sense that one or more parameters have the "wrong sign". One could be inclined to think that this is due to collinearity across explanatory variables, suggesting one should leave out one or more of the collinear variables. In this note we show that this is not a valid approach. Additionally, we show that "wrong signs" can occur because of correlations between included and omitted variables, so that "wrong signs" may occur if the model is not correctly specified. That is, if we find 'wrong signs" we should start questioning our model choice, not the data.

collinearity, misspecification, parameter estimation
Statistical Decision Theory; Operations Research (jel C44), Business Administration and Business Economics; Marketing; Accounting (jel M), Marketing (jel M31), Compensation and Compensation Methods and Their Effects (stock options, fringe benefits, incentives, family support programs, seniority issues) (jel M52)
Erasmus Research Institute of Management
ERIM Report Series Research in Management
Copyright 2002, P.H.B.F. franses, C. Hey, This report in the ERIM Report Series Research in Management is intended as a means to communicate the results of recent research to academic colleagues and other interested parties. All reports are considered as preliminary and subject to possibly major revisions. This applies equally to opinions expressed, theories developed, and data used. Therefore, comments and suggestions are welcome and should be directed to the authors.
Erasmus Research Institute of Management

Franses, Ph.H.B.F, & Heij, C. (2002). Estimated Parameters Do Not Get the "Wrong Sign" Due To Collinearity Across Included Variables (No. ERS-2002-31-MKT). ERIM Report Series Research in Management. Erasmus Research Institute of Management. Retrieved from