(Page 9 of 16) - Term Limits and Partisan Swings authored by Clucas, Richard.

important variable in explaining the Democratic Party’s success in gaining legislative seats was
how it did in the presidential election. A one-percent increase in the Democratic Presidential
Vote was associated with a .956 increase in the percentage of seats held by Democrats. This was
not only large, but statistically significant at .01. The percentage of seats held by the Democrats
prior to the election and the surge in voter turnout are also important in explaining the party’s
gains and losses, both substantively and statistically. For every one percent of seats held by
Democrats prior to the election, the party experiences a decline of .332 percent. With an average
increase in voter turnout from 2002 to 2004 of 744,046 voters and an average decline in the
democratic presidential vote of 8 percent, the coefficient for Turnout Surge means that the
changes in turnout were responsible for almost an 11 average decline in the percentage of seats
held by Democratic Party. Changes in PCI did not affect the percentage of seats held by the
Democrats.

The results for the alternative models I ran were almost identical to table 2, even though

they used alternative interactive variables. Looking at the results for 2000 shown in table 3, the
growth in per capita income in states with term limits provides the largest variance in the
dependent variable. Given the 2.7 increase in real PCI in 2000, the coefficient indicates that the
growth in the economy was associated with an average percentage gain of 8.5 Democratic seats
in these states. This coefficient is also statistically significant. In table 4, the Democratic Vote for
president in term limited states explains the most variance, followed by the surge in turnout in
these states. The size and direction of the other coefficients in both of these models closely
parallel with those in table 2. After the interactive variables, the coefficient for the Democratic
Presidential vote explains the most variance. The other coefficients have minimal impact at best
on the results.

While the interactive variables were consistently the most important in explaining

Democratic seat gains in the 2000 election, they were not relevant in explaining legislative seat
changes in 2004. By far, the most important factors in explaining seats changes in 2004 were the
presidential vote, the number of seats held prior to the election, and the surge in turnout. The size
and significance of the coefficients in table 3 and 4 are almost identical to those in table 2,
though I have changed the interactive variables in each equation.

In addition to these models, I also examined two other models. The first one included the

four base variables along with the interactive variable for Turnout Surge. The second one
included the four base variables along with two interactive variables, one for the Democratic
Vote in states with term limits and the other for PCI Growth in states with term limits. In the first
model, I found that the interactive variable for Turnout Surge had no effect on the dependent
variable in 2000, unlike the other two interactive variables described above. The model including
that variable also had a very poor fit (Adj. R

= -.018). The use of this interactive variable did not

change the results for the 2004 elections. The model using both of the two interaction terms
suffered severe multicollinearity in both years. I also ran a regression combining the results from
both elections. However, there was a strong correlation between PCI growth and the dummy
variable I included in the equation to control for differences between the two years. As a result, I
have just reported the results for the analyses of the separate years.

The Effect of the Term Limit Cycle

Finally, the last question I examined was whether the changes in the percentage of seats