0-Truncated Negative Binomial Regression
Command line:
^trnbin0^ depvar [varlist] [weight] [^if^ exp] [^in^ range] [, ^lt(^#^) ir^r ^off(^var^) nol^og^ rob^ust ^score^(var)^ cl^uster(var)]
^lt(#)^ = tolerance for convergence ^irr^ = parameter estimates displayed in exponential form; standard errors and CIs adjusted ^off(str)^ = offset variable - log raw var prior to using as offset ^nolog^ = no display of iteration log ^robust^ = displays robust variance estimator; auto w cluster ^cluster(str)^ = identifies variable used for clustering ^score(str)^ = creates a variable with score value of the primary LL function
Results may be redisplayed by entering the command. Exponential form and C.I. level are allowed as options without reestimation.
Use: A 0-truncated negative binomial model is appropriate when modeling count data which have no possibility of having 0 values. This is to be distinguished from data sets without 0 values, but which may have 0's. For example, length of hospital stay data (LOS) begins with patients havin > g a LOS of 1. 0 values are not allowed. This is a situation where a truncated model may be appropriate. Examples - --------
. ^trnbin0 studytim died drug2 drug3, irr nolog^ . ^trnbin0 infected cases cd4 cd8, nolog irr off(cases) cluster(prov)^
. ^trnbin0, level(90)^ [redisplay]
Also see - --------
Manual: [5s] poisson, glm, cnreg On-line: help for @poisson@, @glm@, @cnreg@, @trpois0@
.- Joseph Hilbe Arizona State University hilbe@@asu.edu