Online Appendix for The Letter After Your Name : Party Labels on Virginia Ballots published in State Politics and Policy Quarterly Alex Garlick Ph.D. Candidate, University of Pennsylvania www.alexgarlick.com List of Figures 7 Charlottesville, Virginia Sample Ballot 1989.................. 8 8 Charlottesville, Virginia Sample Ballot 2013.................. 9 9 Virginia Newspaper Attention to Campaigns: 1997-2001........... 10 10 Change in Precinct Registration Counts.................... 10 11 Loudon County, Virginia Sample Ballot 1997.................. 11 12 Fairfax County, Virginia Sample Ballot 2013.................. 12 1
Appendix Examples of Virginia Ballots Figure 7 is a photocopy of a 1989 sample ballot. 1 Candidates for state-wide office were only listed by name. Figure 8 is a 2013 sample ballot downloaded from the Charlottesville Board of Elections website. It demonstrates that after the 2000 ballot reform, candidates were now listed by name, with a letter representing their party affiliation: R indicates Republican candidates, D indicates Democratic candidates. Figure 1 featured excerpts of Figure 7 and Figure 8 prepared by the author. Figure 11 is a 1997 sample ballot without party labels in Loudon County. 2 Figure 12 is an electronic ballot sample downloaded from the Fairfax County website from 2013 showing the policy being applied on electronic voting machines. 2000 Presidential Election The empirical model employed in this paper uses George W. Bush s vote share in the 2000 General Election to proxy for the national partisanship of voters in Virginia precincts. As Levendusky, Pope, and Jackman (2008) note, using the average of a few presidential elections is similar to Converse s measure of the national vote. I re-ran the analysis with 2000 and 2004 average GOP Presidential vote share and there is no significant difference from the results reported. The question then becomes, was the 2000 Presidential General Election representative of the national partisanship of Virginia voters? On a national scale, this election was unusual, featuring the infamous butterfly ballot in Florida (Niemi and Herrnson, 2003). However, it was representative for Virginia. Table 7 shows that it was consistent with the proximate general elections in this era. 1 I thank Anne Benham at the University of Virginia Library for forwarding this copy of a ballot maintained by the registrar at the Charlottesville Board of Elections. 2 I thank Judith Brown, the Central Registrar for Loudon County for sending me this photocopy. 2
Table 7: Two Party Vote Shares in Virginia General Elections Year Republican Votes Democrat Votes GOP Vote Share 1996 1,138,350 2,092,060 51.06% 2000 1,437,490 1,217,290 54.15% 2004 1,716,959 1,454,742 54.13% Media Coverage of Virginia State Politics 1997-2001 A content analysis of the coverage of Virginia campaigns in the three years of my sample demonstrates the lack of attention paid to state-level races, like Lt. Governor or the House of Delegates, compared to national politics by newspapers. Figure 9 shows the number of stories in the regional Virginia papers in the LexisNexis database on the campaigns for President, Governor, Lt. Governor and the House of Delegates. A one percent random sample of the state-level stories shows that 90 percent of the articles in the sample directly address the campaigns in question. These article counts are from the Virginia News Sources in LexisNexis for the six months preceding Election Day. The count is of articles containing the word: Campaign, name of the office: President, Governor, Lieutenant Governor, or House of Delegates, and the last name of the Republican and Democratic candidates (for executive offices). The primary newspapers in the LexisNexis database for this region are: Richmond Times-Dispatch, The Washington Post, & Washington Times. The dates of the content analysis are: May 3, 1997 to November 3, 1997; May 6, 2000 to November 6, 2000; May 5, 2001 to November 5, 2001. Change in Registration Counts Figure 10 demonstrates the change in voter registrations in Virginia precincts during the years in my sample. Changes could have resulted from a round of redistricting that went into effect before the 2001 election. Also, there is population growth and voter registration 3
efforts that can drive up registration counts. A value of 1.0 in the figure indicates there was no change in registrations from 1999 to 2001 or 2001 to 2003. The data are censored for precincts that increased or decreased by 50 percent as it indicates that a substantially different set of voters were in that precinct across elections. Results are not substantively different including these outlier districts in all models. Clustered Standard Errors In the results section, findings were presented with robust standard errors meant to correct for heteroskedasticity in the data. Another concern is that the residuals are correlated with another. Specifically, the model could be vulnerable to spatial correlation in the GOP presidential vote share variable. Table 8 reports the intraclass correlations (Angrist and Pischke, 2008, ch. 8) for the four main models in the results section, clustered by locality. While the intraclass correlations reported in Table 8 are not extreme, a method to account for this type of correlation is to cluster standard errors. However, there are two concerns using that method with these data. First, Angrist and Pischke (2008) warn against clustering with too few units as it can bias the results. This is a particular concern for the state legislative races, as Table 8 shows that precincts would be grouped into as few as 40 localities. A second concern is that the size of the logical cluster units (locality) are not uniform. Virginia s localities, which are counties and larger cities, have as few as 4 precincts (in Buena Vista County) and as many as 272 precincts (in Fairfax County). In Tables 9 & 10 the topline results are replicated alongside models with clustered standard errors. In some cases, clustering increases the standard errors substantially. In the case of the House of Delegates, the coefficient on the key interaction term is not robust to this method of calculating standard errors. It should be noted that the result for Lt. Governor remains significant when the standard error is being clustered by a sufficiently high 133 units. Due to the concerns relayed above, the clustered standard errors are not employed in the 4
main results of the paper and only presented here to relay a conservative estimate of the upper bound for standard errors in the partisan voting models. Table 8: Intraclass Correlations for Clustered Models in Tables 9 & 10 Office Gov. Lt. Gov VA Senate VA House Obs. (Precincts) 3860 3860 774 908 Clusters (Localities) 133 133 40 47 ICC in %Bush 2000 0.437 0.437 0.544 0.507 ICC in Residual 0.266 0.350 0.663 0.540 ICC calculated with moulton package in STATA. 5
Table 9: Association with National Partisanship: Executive Offices Years 1997-2001 Office Gov. Lt.Gov (1) (2) (3) (4) Robust Standard Errors Yes Yes Yes Yes Clustered Yes Yes Post2000-0.088-0.088-0.094-0.094 (0.004) (0.010) (0.006) (0.010) %Bush2000xPost2000-0.028-0.028 0.105 0.105 (0.007) (0.017) (0.011) (0.025) Constant 0.559 0.559 0.510 0.510 (0.001) (0.002) (0.001) (0.004) Observations 3860 3860 3860 3860 Clusters 133 133 Cluster Units Locality Locality Standard errors in parentheses. All models have precinct fixed effects. p < 0.05, p < 0.01 6
Table 10: Association with National Partisanship: State Legislative Races Years 1999-2003 1999-2001 Office VA Senate House of Delegates (1) (2) (3) (4) Robust Standard Errors Yes Yes Yes Yes Clustered Yes Yes Post2000-0.073-0.073-0.085-0.085 (0.022) (0.040) (0.029) (0.063) %Bush2000 x Post2000 0.188 0.188 0.192 0.192 (0.041) (0.085) (0.056) (0.123) Constant 0.469 0.469 0.494 0.494 (0.002) (0.007) (0.003) (0.007) Observations 774 774 908 908 Clusters 40 47 Cluster Unit Locality Locality Standard errors in parentheses. All models have precinct fixed effects. p < 0.05, p < 0.01 7
Figure 7: Charlottesville, Virginia Sample Ballot 1989 8
Figure 8: Charlottesville, Virginia Sample Ballot 2013 9
Figure 9: Virginia Newspaper Attention to Campaigns: 1997-2001 3389 Stories 0 500 1,000 3,300 452 907 263 134 120 1997 2000 2001 173 Governor Lt. Governor House of Delegates President Figure 10: Change in Precinct Registration Counts Density 0 2 4 6 Censored Censored < 1/1.5 > 1.5/1 0.5 1 1.5 2 2001 2003 10
Figure 11: Loudon County, Virginia Sample Ballot 1997 11
Figure 12: Fairfax County, Virginia Sample Ballot 2013 12
References Angrist, Joshua D, and Jorn-Steffen Pischke. 2008. Mostly harmless econometrics: An empiricist s companion. Princeton university press. Levendusky, Matthew S., Jeremy C. Pope, and Simon D. Jackman. 2008. Measuring District-Level Partisanship with Implications for the Analysis of U.S. Elections. The Journal of Politics 70 (7): 736 753. Niemi, Richard G, and Paul S Herrnson. 2003. Beyond the butterfly: The complexity of US ballots. Perspective on Politics 1 (02): 317 326. 13