{smcl}
{* 27jul2004}{...}
{hline}
help for {hi:clogit_score}    
{hline}

{title:Obtaining likelihood scores after {cmd:clogit}}

{p 8 22 2}{cmd:clogit_score} {{it:newvarlist} | {it:stub}{cmd:*}} [{cmd:if}
{it:exp}] [{cmd:in} {it:range}] [{cmd:,} {cmdab:now:eight} 
{cmdab:nooff:set}]

{title:Description}

{p 4 4 2}
{cmd:clogit_score} generates new variables containing likelihood scores 
after estimation using {cmd:clogit}.  You must either specify {it:k} new
variables, where {it:k} is the number of independent variables in your model,
or specify {it:stub}{cmd:*} and {cmd:clogit_score} will
automatically generate the required number of variables, and name them
{it:stub}{cmd:1}, {it:stub}{cmd:2}, etc.

{p 4 4 2}
In the case of {cmd:clogit}, likelihood calculations take place at the group
level, where the grouping variable is specified to {cmd:clogit} via options
{cmd:group()} or {cmd:strata()}.  As a result, the likelihood scores are also
calculated at the group level, and placed into the first observation of each
group.

{p 4 4 2}
By default, {cmd:clogit_score} will calculate scores for the estimation 
sample.  You can override this default (in the rare case that you would 
want to predict scores out of sample) by using {cmd:if} and {cmd:in}.

{title:Options}

{p 4 8 2}{cmd:noweight} specifies that if weights were used in fitting 
the model, they be ignored when calculating scores.  You should rarely have
to specify this option since the resulting scores wouldn't have any useful
interpretation.

{p 4 8 2}{cmd:nooffset} specifies that if a linear offset was used in fitting
the model, it be ignored when calculating scores.  You should rarely have to 
specify this option.

{title:Examples}

{p 4 8 2}{cmd:. clogit low lwt nonwhite smoke ptd, group(pairid)}{p_end}
{p 4 8 2}{cmd:. clogit_score sc*}

{p 4 8 2}{cmd:. clogit low lwt nonwhite, group(groupid) or}{p_end}
{p 4 8 2}{cmd:. clogit_score s_lwt s_nonwhite}

{title:One note on speed}

{p 4 4 2}{cmd:clogit_score} works by looping over the groups in the data
and obtaining scores as the likelihood gradient vector treating each group
as the entire dataset.  As a result, for large datasets {cmd:clogit_score}
is what you would call a "get up and go get a cup of coffee" command.

{title:Also see}

{p 4 13 2}
{hi:[R] clogit}