Previous: Format:

Up: DISTS

Next: Graphics:

Purpose:

DISTS gives detailed information about your model's district-level results. It enables you to draw conclusions about the partisan makeup of each individual district and the probability that each district will elect a Democrat or Republican. (For example, you might wish to evaluate the partisan makup of districts that have incumbents, or of ``minority districts'' drawn to adhere to the Voting Rights Act.) You also can evaluate the fit of your model to the data, or its success in predicting future election results, by comparing lists or plots of observed and expected (or predicted) district-level results.

Our model predicts district-level election results, relative to a specified level of the average district vote, which may be input directly by supplying VBAR, or indirectly through DELTA, the average statewide partisan swing to the Democrats. If neither VBAR nor DELTA is specified, JudgeIt assumes DELTA=0 (i.e. no statewide swing outside of the systematic changes expected because of values indicated by the explanatory variables in XNEW).

There are three general situations in which you might want to use a DISTS statement:

1. Evaluation--to view district-level results from a model applied to data from a past election. To do this, specify YVOTE with a variable name, and XVARS (with no XNEW statement) before DISTS. If you choose DISTS LIST, the following information will be printed: district number, observed Democratic proportion of the two-party vote, expected vote proportion, standard error of the expected vote proportion, and the probability that this district will elect a Democrat in hypothetical repetitions of the same elections. If you choose DISTS PLOT, a graph will be printed with expected vote printed horizontally and observed vote vertically; districts are represented by little circles. You can use this plot to assess the fit of your model to the data. See Section 9, Subsection 9.2 for a command file that implements this procedure.

2. Prediction--to predict the likely district-level results for a future election possibly under a new redistricting plan. To do this, first estimate coefficients based on data from elections that have already been held. Thus, specify YVOTE (with a variable) and XVARS. If you wish to see the results of this preparatory analysis, call some analysis procedure (such as DISTS, SUM, SVCURVE, or SEATS), or issue a REG; command.If you only wish to see the results of this preliminary regression, specify REG ON; before calling REG; Then, for the prediction you wish to see, specify YVOTE PREDICT to indicate that this is a prediction problem, and use XNEW to identify the explanatory variables you are using for prediction, and finally the DISTS command. For a command file that illustrates a predictive use of the DISTS command see Section 9.4 in Section 9.

   If you are using DISTS to evaluate your predictions, and therefore have the data for the election year you are predicting, specify a YVOTE2 variable with this information. If you choose DISTS LIST, the following information will be printed: district number, observed Democratic proportion of the two-party vote for the election you are predicting (if you specify YVOTE2), expected vote proportion, standard error of the expected vote proportion, and the probability that this district will elect a Democrat in hypothetical repetitions of the same elections.

   For prediction problems, DISTS PLOT is only valid if you do specify YVOTE2. In this case, a graph will be printed with predicted vote horizontally and observed vote vertically; districts are represented by little circles. You can use this plot to assess the adequacy of your model's predictions.

3. Counterfactual evaluation--to evaluate district-level results under a specified counterfactual situation (i.e., a situation that could have occurred but did not), in an election that already has taken place. To do this, specify YVOTE with a variable, XVARS, and XNEW. XNEW specifies the counterfactual you wish to evaluate. For example, you could see whether the Democratic candidate would have been likely to win district 43 had the Republican incumbent in that district decided not to run for reelection. To do this, use an XNEW statement that is the same as the XVARS statement, except that you would change the incumbency code in only district 43 from Republican incumbent to open seat.

By default, all standard errors are based on the total variation, which is appropriate for evaluating model predictions (since if your model is reasonable, you should expect to get the predicted or expected vote proportion correct to within plus or minus two standard errors for about 95% of the districts). If you wish your standard errors to be based only on the variability in your estimation of the expected (or predicted) vote proportion, add the EXPECTED option to the DISTS statement. See the discussion under SVCURVE for the distinction between the two kinds of standard errors.