TY - JOUR
T1 - Multiple regression with inequality constraints
T2 - Pretesting bias, hypothesis testing and efficiency
AU - Lovell, Michael C.
AU - Prescott, Edward
N1 - Funding Information:
*Michael C. Love11 is professor of economics. Wesleyan University. Edward Prescott is on leave from the Department of Economics, University of Pennsylvania. as a Brookings Economic Policy Fellow. Department of Labor. The authors are indebted to Robert. Summers, d o h W. Pratt and Arnold Zellner for helpful advice and comments on a preliminary draft. Computations were executed on the University of Pennsylvania computer. This article was supported by the Graduate School of Industrial Administration, Carnegie-Mellon University, Wrsleyan Univeraity, the Wliarton School of Finance and Comrnercc and the National Science Foundation.
PY - 1970/6
Y1 - 1970/6
N2 - This article analyzes, within the context of the standard multiple regression model, the problem of handling inequality constraints specifying the signs of certain regression coefficients. It is common econometric practice when regression coefficients are encountered with incorrect sign to delete the variables in question and reestimate the equation. This article shows that this procedure causes bias and can lead to inefficient parameter estimates. Furthermore, we show that grossly exaggerated statements concerning significance levels are likely to be made when other regression coefficients in the model are tested with the final regression obtained after deleting variables with incorrect sign.
AB - This article analyzes, within the context of the standard multiple regression model, the problem of handling inequality constraints specifying the signs of certain regression coefficients. It is common econometric practice when regression coefficients are encountered with incorrect sign to delete the variables in question and reestimate the equation. This article shows that this procedure causes bias and can lead to inefficient parameter estimates. Furthermore, we show that grossly exaggerated statements concerning significance levels are likely to be made when other regression coefficients in the model are tested with the final regression obtained after deleting variables with incorrect sign.
UR - http://www.scopus.com/inward/record.url?scp=0012667910&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0012667910&partnerID=8YFLogxK
U2 - 10.1080/01621459.1970.10481134
DO - 10.1080/01621459.1970.10481134
M3 - Article
AN - SCOPUS:0012667910
VL - 65
SP - 913
EP - 925
JO - Journal of the American Statistical Association
JF - Journal of the American Statistical Association
SN - 0162-1459
IS - 330
ER -