Previous: Purpose:

Up: PRELIM

Next: PRELIM AUTOMATE

Options:

As we describe above, LAMBDA and SIGMA should be averaged over several elections. Thus, in the initial run, you can simply use PRELIM; after each definition of YVOTE and the rest of the data for each election you plan to analyze. The results will be printed each time on the screen and your output file. NOTE: if val1 is less than zero or greater than one, it should be truncated to within this interval. But also check to see if something is wrong with your data or model. One common problem which leads to this situation is when you omit an incumbency status variable from the XVARS command and the electoral system you are evaluating has a strong incumbency advantage; another problem is omitting a partisan control variable (such as that created by the PTY! operator). If val2 is less than zero or greater than one, then you almost certainly have some problem with your model or data. Check to make sure that the YVOTE variable is a proportion (ranging from zero to one) and not a percentage (ranging from zero to 100). You can check this by running the STATS command to get a summary of the data after only specifying YVOTE. In practice, we find that average values of LAMBDA have a fairly large range (0.2 to 0.98) but SIGMA is almost always between 0.02 and 0.12. Note also that a very large variation in LAMBDA and SIGMA for different elections you are analyzing may indicate a problem. One such problem is a large variation in uncontested elections. In this case, make sure you recode these (using the CHG! operator or IMPUTE statement) and/or include an explanatory variable for uncontestedness status (using the UNC! operator) in XVARS and also XVARS2.

To simplify the comparing and averaging of your estimates of SIGMA and LAMBDA, you can use PRELIM Pnum; after you define the data. This will print the output, just as with the PRELIM; version of the command, but it will also save it. For each election analysis, you can issue another PRELIM Pnum command, and these will all be saved. At the end of the run, you should then issue a PRELIM WRITE; command; this takes all the previous PRELIM Pnum; results, sorts them by num (and, within values of num, the year specified on the YVOTE command), and prints them all out in a convenient table. For example, you may have one set of elections immediately following redistricting periods for which you have only a few explanatory variables, and many other elections for which you have much better explanatory variables. Call these Type 1 and Type 2 elections, respectively. In this situation, you would probably do PRELIM P1; for Type 1 and PRELIM P2; for Type 2 elections. Then when you do a PRELIM WRITE; command, the results will be printed in chronological order for Type 1 and then for Type 2 elections. This will make it easier to examine the data for patterns before you decide whether to average LAMBDA and SIGMA once for Type 1 elections and once for Type 2 elections.

After you have averaged your estimates of LAMBDA and SIGMA, and chosen a value of each for each analysis you plan to perform, you should change the PRELIM command already in your JudgeIt command file to the final format: PRELIM LAMBDA num1 SIGMA num2;. With this form of the command, PRELIM prints out nothing, and only sets the values of LAMBDA and SIGMA for subsequent analyses. (See the discussion of the ALIAS command, which will enable you to change the alias at the top of the file instead of each individual request, and otherwise make the commands simpler.)