HIST graphs a ``kernel density estimate'' of your
district-level observed YVOTE variable or your district-level predictions. A
kernel density estimate is a smooth version of a histogram.See Gary King
(``Constituency Service and Incumbency Advantage," British Journal of
Political Science, 21, 1 (January, 1991): 119-128) for an example from the
political science literature. A kernel density estimate puts a distribution
around each district and averages all the distributions to get the plotted
graph. The larger the variance of these individual distributions, the smoother
the whole graph will be. If the variances are zero (which is not allowed in
HIST), then each district will be represented by a spike at its vote
proportion. For the individual distributions, HIST uses a normal kernel
truncated from below at zero and from above at one. The key point to remember
in interpreting the graph is that the total area under the curve is equal to
1.0 (i.e. all the districts). Thus, the proportion of districts more than 70
percent Democratic can be seen on the graph by looking at the proportion of the
area under the curve to the right of 0.7. Also printed on the graph are
``whiskers,'' which are short lines at the bottom indicating the vote
proportion for each individual district.If you cannot see the whiskers for
a run, specify the MAX option. The graph drawn by HIST shows the overall
distribution of partisanship among districts in a state. For example, if few
districts are near the 0.5 point, then the state has relatively few competitive
districts. If most of the districts clump over on the right side of the graph,
then the state is predominately Democratic. Another pattern, which might be
produced by a Democratic gerrymander, would have a few Republicans winning by
very large margens and many Democrats, each winning with just a little over
of the vote.
Since HIST works with observed and predicted results, you can compare results
from an actual election with that under proposed redistricting plans. For
actual election results, you only need to define YVOTE. For predicted results,
first set up a prediction: YVOTE year var 
dataset;
XVARS vars dataset;, with XVARS defined for the most
recent election for which you have data. You must also issue a REG;
command or run any of the analysis procedures. You then can make a prediction
by defining YVOTE with PREDICT, followed by XNEW.
Gary King
2006-01-07
