{smcl}
{* 19nov2002}{...}
{hline}
help for {hi:leastlikely}{right:(SJ2-4: st0022_1)}
{hline}

{title:List least likely observations}

{p 8 13}{cmd:leastlikely} [{it:varlist}] [{cmd:if} {it:exp}] 
[{cmd:in} {it:range}] 
[{cmd:,} {cmd:n(}{it:#}{cmd:)} {cmdab:g:enerate(}{it:varname}{cmd:)}
[{cmdab:no:}]{cmdab:d:isplay} {cmdab:nol:abel} {cmdab:noo:bs}
{cmdab:do:ublespace}]


{title:Description}

{p 4 4 2}For regression models for categorical dependent variables, 
{cmd:leastlikely} lists the in-sample observations with the lowest 
predicted probabilities of observing the outcome value that was 
actually observed.  For example, in a model with a binary dependent 
variable, {cmd:leastlikely} lists the observations that have the lowest 
predicted probability of {it:depvar=0} among those cases for which 
{it:depvar=0}, and it lists the observations that have the lowest predicted 
probability of {it: depvar=1} among those cases for which {it:depvar=1}.  The 
least likely values represent relatively deviant cases that may warrant 
closer inspection.

{p 4 4 2}{cmd:leastlikely} works with estimation commands for models of binary 
outcomes in which option {cmd:p} after {cmd:predict} provides the predicted 
probability  of a positive outcome (e.g., {cmd:logit}, {cmd:probit}), but the 
dependent  variable must be coded as 0 and 1.  Likewise, {cmd:leastlikely} 
works with estimation commands for models of ordinal or nominal outcomes 
in which option {cmd:outcome(}{it:#}{cmd:)} after {cmd: predict} provides the
predicted probability of outcome {it:#}.  Exceptions are commands in which the
predicted probabilities are probabilities within groups or panels or for
"blocked" data; {cmd:leastlikely} will produce an error message if executed
after {cmd:blogit}, {cmd:bprobit}, {cmd:clogit}, {cmd:glogit}, 
{cmd:gprobit}, {cmd:nlogit}, or {cmd:xtlogit}.  

{p 4 4 2}{cmd:leastlikely} lists the observation number and the
predicted probability (as Prob or as the variable name specified by the
{cmd:generate} option).  Values of variables in {it:varlist} will also be 
specified.


{title:Options}

{p 4 8 2}{cmd:n(}{it:#}{cmd:)} specifies the number of observations to be listed
for each outcome.  The default is 5.  If multiple observations have the same
probabilities, more than the specified number will be listed to include all
of them.

{p 4 8 2}{cmd:generate(}{it:varname}{cmd:)} specifies that the probabilities
of observing the outcome value that was observed should be stored in
{it:varname}.  If not specified, the variable name {cmd:Prob} will be created
but dropped after the output is produced.

{p 4 8 2}[{cmd:no}]{cmd:display} forces the format into {cmd:display} or 
tabular ({cmd:nodisplay}) format.  If you do not specify one of these two 
options, then Stata chooses one based on its judgment of which would be most 
readable.

{p 4 8 2}{cmd:nolabel} causes the numeric codes rather than the label values
to be displayed.

{p 4 8 2}{cmd:noobs} suppresses printing of the observation numbers.

{p 4 8 2}{cmd:doublespace} produces a blank line between each observation in
the listing when in {cmd:nodisplay} mode; it has no effect in {cmd:display}
mode.


{title:Examples}

{p 4 8 2}{cmd:. logit low age lwt race2 race3 smoke ptl ht ui}{p_end}
{p 4 8 2}{cmd:. leastlikely}{p_end}
{p 4 8 2}{cmd:. leastlikely age lwt, n(10) g(prob)}{p_end}


{title:Author}

      {browse "mailto:jfreese@ssc.wisc.edu":Jeremy Freese}
      Department of Sociology,
      University of Wisconsin-Madison