16
Table 2. Regressions of Expenditures on Vote Shares (LDP 2000)
Variables
Model I
Model II
Model III
Own campaign-period spending
0.08(0.02)***
--
0.06(0.02)***
Rival campaign-period spending
-0.01(0.02)
--
0.03(0.02)*
Own fund spending
--
0.00(0.00)**
0.00(0.00)**
Own party branch spending
--
0.01(0.00)***
0.01(0.00)**
Own koenkai spending
--
0.01(0.00)***
0.01(0.00)**
Rival fund spending
--
-0.03(0.01)***
-0.03(0.01)***
Rival party branch spending
--
0.00(0.00)
-0.00(0.00)
Rival koenkai spending
--
-0.01(0.01)
-0.01(0.01)
Incumbent
2.44(0.77)**
1.50(0.74)**
1.43(0.72)**
Previous terms
0.18(0.10)*
0.06(0.10)
0.07(0.10)
Number of support groups
0.03(0.02)
0.02(0.02)
0.02(0.02)
Competition
0.42(0.04)***
0.39(0.03)***
0.41(0.03)***
Number of candidates
-2.18(0.28)***
-2.03(0.27)***
-1.97(0.26)***
% Densely inhabited district (DID)
-0.10(0.01)***
-0.11(0.01)***
-0.09(0.01)***
Number of cases
271
271
271
Intercept
30.02
33.57
29.70
Adjusted R-squared
0.77
0.79
0.81
Note: Dependent variable = LDP percentage of total district vote.
* = significant at .10 level. ** = significant at .05 level. *** = significant at .001 level.
Model II excludes the campaign-period figures and is used to assess the effects of
the types of year-based expenditures on vote shares. As detailed above, own fund, party
and koenkai spending have a positive effect upon the percentage of the LDP district vote
(0%, 1%, and 1% for ¥100), whereas rival fund expenditures has a negative impact
(3%).
15
In this model, both own and rival expenditures appear to have their hypothesized
effects on the vote shares of LDP candidates.
Finally, Model III captures the combined effects of campaign-period and year-
based expenditures. Own party branch and koenkai spending leads to a small effect of 1%
and 1% of the vote share for each increase in spending of ¥100 ($.94) per elector. In
15
I checked the results for the LDP and DPJ using Cook and Weisberg’s test for heteroscedasticity. Two
of the models showed slight signs of heteroscedasticity. To correct for this, I performed regression with
robust standard errors. The coefficients and statistical significance of the corrected models differed little
in regards to the results reported above.
Convention