------------------------------------------------------------------------------- help forcomplogit(Glenn Hoetker) -------------------------------------------------------------------------------

Compare coefficients across logit equations

complogitvarlist,group(varname)[numerator(varname)denominator(varname)specific(varname)]

Description

complogitallows for the accurate comparison of coefficients across groups in a logit equation by automating the calculations discussed in Hoetker > (2004; 2007). This help file will make little sense without reference to those articles.

In its basic application, it performs the requested logit equation on all observations, on cases where the

groupvariable=0 and on cases where it equals 1. It then performs a logit for all observations, allowing residual variance to differ across groups. It then tests whether residual variation differs across groups using a likelihood ratio and Wald chi-square test. Lastly, it uses a likelihood ratio test to test the null hypothesis that the true coefficients are the same across groups versus the alternative thatat least onecoefficient differs.

Optionally, the user may request a test of test of the null hypothesis that the > true coefficients are the same across groups versus the alternative that

a specificcoefficient differs. To do so, specify the desired coefficient in thespecificoption. Note that this test assumes that all other coefficients are equal across groups.

Optionally, the user may request a test that the ratio of two coeffiients is the same across groups. To do so, specify the desired variables in the

numeratoranddenominatoroptions.

OptionsSee above

Remarks

Version 8 of Stata is required. There is minimal error checking.

complogitml.adocontains the maximum likelihood routine and is of little direct use. It is based on code contained in Allison (1999). Users may also wish to s > ee theoglmcommand of Richard Williams and his associated papers.

ReferencesAllison, P.D. 1999. Comparing logit and probit coefficients across groups.

SMR/Sociological Methods & Research. 28(2): 186-208.Hoetker, G. 2004. Confounded coefficients: Accurately comparing logit and probit coefficients across groups. Working paper. http://www.business.uiuc.edu/ghoetker/documents/Hoetker_comp_logit.pdf

Hoetker, G. 2007. The use of logit and probit models in strategic management re > search: Critical issues.

Strategic Management Journal. 28(4): 331-343.

Examples. complogit y x1 x2, group(type)

. complogit y x1 x2, group(type) num(x1) den(x2)

. complogit y x1 x2, group(type) sp(x1)

AuthorGlenn Hoetker, College of Business, University of Illinois at Urbana-Champaign Email: ghoetker@uiuc.edu