Abstract
A reexamination of the data revealed that the data set included multi-member districts in the 2002 and 2004 cycles for Vermont and West Virginia. Incumbents in multi-member districts tend to receive lower vote shares than incumbents in single-member districts. Thus the inclusion of incumbents in multi-member districts resulted in a lower average incumbent vote share in those two states in 2002 and 2004, and a higher number of candidates running for election than in other years. At the same time, 2002 and 2004 were years when Vermont had low contribution limits (while there was no change in contribution limits for West Virginia). As a consequence of the error associated with the Vermont data, the author's analysis overstates the importance of low contribution limits for the competitiveness of election.
Table I shows the correct results. For comparison purposes, Table II, a summary of the tables from the original article, also appears below. Each of these tables has rows labeled Table 2, Table 3, etc. that refer to the tables in the original article.
Notes: Standard errors in parentheses and are clustered by state. *significant at 10 percent; **significant at 5 percent; ***significant at 1 percent. Control variables such as election cycle and state indicators, as well as state-level demographic, income, and partisanship controls are included in the regressions but not reported. Number of candidates is the log number of candidates. Columns one, four, and five report OLS estimates. Columns two and three report marginal effects from logit estimation.
Notes: Standard errors in parentheses and are clustered by state. *significant at 10 percent; **significant at 5 percent; ***significant at 1 percent. Control variables such as election cycle and state indicators, as well as state-level demographic, income, and partisanship controls are included in the regressions but not reported. Number of candidates is the log number of candidates. Columns one, four, and five report OLS estimates. Columns two and three report marginal effects from logit estimation.
After correcting for the data error, the estimation results continue to show that having contribution limits increases the competitiveness of election (Table I, row Table 2). Further, the results for both the log contribution limit amount and for whether there is a limit in place have effects that are similar to those reported in the original article (Table I, row Table 3). However, the findings for low contribution limits show a different picture. For states with contribution limits (Table I, row Table 4, column 1) the new results show that a limit of $500 or less reduces an incumbent's margin of victory by 8.2 percentage points (instead of 14.5 percentage points as originally reported), a limit between $501 and $1,000 reduces that margin by 7.7 percentage points (instead of 9.5 percentage points), and a limit between $1,000 and $2,000 reduces the incumbent's margin by 4 percentage points (instead of 5 percentage points). Thus the most significant change per the data correction is reflected in the results for low limits. This is also the case when examining the other measures of competitiveness. Table I, row Table 4, column 2 shows that limits of $500 or below increase the likelihood of a narrowly fought election by 7 percent instead of 15.8 percent as originally reported, while column 3 shows that these limits decrease the likelihood of a lopsided race by 8.8 percent instead of 15 percent. Column 4 of this same table shows that low limits result in a four percent, instead of a ten percent, chance of incumbent defeat, and column 5 shows that low limits have no effect on the number of candidates, rather than a statistically weak effect as originally reported. Also for these competitiveness measures, the largest correction has to be made for low limits, not the other limit categories. In Table I, row Table 5 the pattern of coefficients is similar to that in row Table 4. In the new specifications, some of the point estimates change in statistical significance.
The author regrets the error.