19
LDP, JSP, JCP and OTHER camps was competing.
12
Although our substantive interest is
mostly on the early period during which the LDP reduced the number of candidates, we have
decided to analyze the entire period of SNTV elections (except for the 1993 election), partly
again for the reason of consistency and partly because we believe that the mechanism with
which the number of LDP candidates was determined under SNTV must have been the same
even beyond the initial period of LDP’s power consolidation. As we discuss below, however,
we will later add a set of dummy variables that measure year-specific fixed effects, in order
both to test the robustness of our results and preempt some criticisms of our model
specification.
We pursue our analysis in this section for three-member, four-member, and
five-member districts separately, and exclude those exceptional one, two and six-seat
districts. As discussed in the previous section, a varying district magnitude is expected to
generate different electoral dynamics and a different set of parameters. Note that during the
period of investigation, some districts experienced an increase/decrease in the district
magnitude due to reapportionment of seats. We drop those districts whose district magnitude
in the current election is not the same as that in the previous election. In order to deal properly
with the time-series, cross-sectional structure of our data, we use a linear regression with a
robust variance estimate that adjusts for within-district correlation (Rogers 1993).
13
12
In fact, we have replicated the analysis including partially-contested districts, but the results did not
differ substantially and they are thus not reported.
13
We have tried various other specifications, including a model by feasible generalized least squares
(FGLS), but the results did not change substantially.
Convention