15
available. The only cell with greater than 10% cross-over is the percent of respondents in 1996
who were closer to Clinton, ideologically, but who campaigned for the Republican party. And,
moving three columns to the right, this figure falls from 10.62% to 8.77% when one restricts
activities to candidate specific campaigns. In general, Republicans seem to cross-over a bit less
than Democrats, but the differences are not substantial. Also, it is apparent that restricting
analyses to respondents who report voting for one of the major parties would eliminate some
cross-over. The amount of cross-over in the “vote” cells is somewhat lower than that of the
“proximity” cells. However, the gain in eliminating cross-over participation must be balanced
against the restriction of subjects to major party voters. While some cross-over is apparent in
Table 1, the rate of cross-over is similar across parties and generally low. Ideally, I would have
more detailed participation information, but the danger of assuming respondents participate for
the most proximate candidate appears to be quite low.
6
***INSERT TABLE 1 ABOUT HERE***
John Zaller has generously made his data on the ideological locations of presidential
candidates available on-line.
7
Zaller (n.d., 17-26) uses Rosenstone’s (1983) survey of experts to
locate presidential candidates in ideological space, from 1948-1980 and he uses the most
informed 15% of NES respondents’ candidate location responses to locate candidates from 1972-
2000. Zaller then combines the two series, using the three overlapping elections to calibrate each
series into the same zero-centered, seven-point scale, metric (see the Appendix for candidate
locations). These candidate locations are in the same seven-point metric as the respondent self-
6
As additional robustness checks, I estimated all models restricting the sample to major party voters. The results
were nearly identical. I also estimated the 1996-2000 model using signed participation (since it was available in the
Rapoport and Stone panel) and signed proximity scores. The results were an even more robust confirmation of
proximity motivated participation. This last robustness check suggests that the non-partisan model is the more
difficult hypothesis test. The cross-over errors appear to bias the models against finding robust evidence of
proximity motivated participation.
7
Downloaded from
http://www.sscnet.ucla.edu/polisci/faculty/zaller/Floating%20Voters,%201948-2000/
on May 9,
2003.
Convention