Finding significant gaps in univariate distributions            [STB-21: sed8]
----------------------------------------------------

        ^wgap^ varlist [^if^ exp] [^in^ range] ^, w^gap^(^prefix^) z(^prefix^)^

measures the significance of gaps or separations in univariate distributions.
The difference between adjacent observations is weighted to highlight gaps in
the middle of the distribution.  An approximate z-score also can be calculated
for each weighted gap.  These z-scores can be used during exploratory data
analysis to identify unusual or suspicious separations in the data. Typically,
z-scores greater than 2.25 are considered candidates for further study, 
although this cutoff can be adjusted higher if you are looking for multiple 
gaps.  ^wgap^ is somewhat robust to heavy-tailed distributions, but it can be
confused by very heavy-tails (for example, uniform distributions) and by
numerous ties.







Options
-------
At least one of the two options, ^wgap()^ or ^z()^, must be specified.  Each
of these options specifies a prefix to use for storing results.  For example,

        ^wgap price, wgap(pgap)^

will store the weighted gaps in the new variable ^pgap^.  Typically, the
z-scores are of more interest than the gaps.

If the varlist contains multiple variables, a number is added to the prefix.
For example,

        ^wgap a b c, wgap(gap) z(z)^

will store the weighted gaps and z-scores for variable ^a^ in ^gap1^ and ^z1^,
respectively, the gaps and z-scores for variable ^b^ in ^gap2^ and ^z2^, and
the gaps and z-scores for variable ^c^ in ^gap3^ and ^z3^.


References
----------

Wainer, H. and S. Schacht. 1978. Gapping. ^Psychometrika^ 43: 203-212.




Author
------

	Richard Goldstein
        Qualitas, Inc.
        richgold@@netcom.com


Also see
--------

    STB:  sed8 (STB-21)