Haynes, Stephen E., 1945-2005-09-022005-09-022005-01-15https://hdl.handle.net/1794/131511 p.This note explores the insidious empirical trap posed by two common equality restrictions in regression analysis. The trap is that restricted coefficients can lie outside the interval of unrestricted coefficients and even reverse sign when negatively correlated regressors are added to one another or when positively correlated regressors are subtracted from one another.170659 bytesapplication/pdfen-USEquality restrictionsSign reversalsInvalid restrictionsThe Empirical Trap of Sign Reversals with Equality RestrictionsWorking Paper