.-
help for ^rglm^
.-

Generalized linear models with robust variance estimates ---------------------------------------------------------

^rglm^ [varlist] [weight] [^if^ exp] [^in^ range] [^,^ ^cl^uster^(^varname^)^ ^ms^pec ^td^ist minus(#) glm-options ]

Description ------------

^rglm^ fits generalized linear models and calculates a Huber (sandwich) estimate of the variance-covariance matrix of estimates. It can be used alone or called without arguments after a previous call to ^glm^. As with other "robust" commands, the units may be considered to fall into clusters.

Options for use with ^rglm^ --------------------------

^cluster(^varname^)^ specifies the variable which defines sampling clusters.

^mspec^ specifies that full Huber variances be used. These are robust to mis-specification of conditional means. If ^mspec^ is absent, semi-Huber variances are calculated, robust to variance mis-specification caused by overdispersion, underdispersion, heteroscedasticity and clustering, but assuming that conditional means are specified correctly by the model. (Except in the case of canonical link functions, where the semi-Huber variance is the full Huber variance. See Section 2.5 of McCullagh and Nelder (1989).)

^tdist^ specifies that P-values and confidence intervals are to be calculated assuming estimates to have a t-distribution with M-k degrees of freedom, where k is the number of model parameters, and M is the number of clusters if ^cluster^ is specified, or the number of observations if ^cluster^ is not specified.

^minus(^#^)^ specifies the ^minus^ parameter to pass to ^_robust^, to apply a finite-sample adjustment to the Huber covariance matrix. If absent (or negative), it is reset to k (the number of model parameters).

If a varlist is supplied, then all ^glm^ options are allowed. If not, then the only ^glm^ options allowed are ^level^ and ^eform^, and ^cluster^, ^mspec^, ^tdist^ and ^minus^ are ignored.

Examples ---------

. ^glm dead ln_dose, family(binomial ntest) link(logit)^ . ^rglm, ms^

. ^rglm dead ln_dose, family(binomial ntest) link(logit)^

. ^rglm dead ln_dose, family(binomial ntest) link(logit) clust(group)^

Author -------

Roger Newson, Imperial College School of Medicine, London, UK. Email: ^r.newson@@ic.ac.uk^

Acknowledgment ---------------

This program was based on a previous version called hglm, which calculated semi-Huber variances, and was kindly supplied to the author by David Clayton of MRC in Cambridge, England.

Also see ---------

Manual: ^[R] glm, [R] _robust^

References -----------

McCullagh, P. and J. A. Nelder. 1989. Generalized Linear Models. 2d ed. London: Chapman & Hall.