Appendix X

Supplemental Report on the Racial Impact of the Rejection of Ballots Cast in Florida’s 2000 Presidential Election and in Response to the Statement of the Dissenting Commissioners and Report by Dr. John Lott Submitted to the United States Senate Committee on Rules in July 2001

Dr. Allan J. Lichtman, Professor
Department of History
American University
Washington, DC 20016

August 2001

INTRODUCTION

This supplemental report provides additional evidence confirming the finding in my first report of wide disparities between ballot rejection rates for blacks and non-blacks in the presidential election of 2000 in Florida. It also examines issues raised in the statement of dissenting commissioners and the accompanying statistical report by Dr. John Lott submitted to the Senate Committee on Rules in late July of 2001.[1] In particular, this report comprehensively examines the question of whether other factors such as poverty, income, education, literacy, and first-time voting account for racial disparities in ballot rejection. The supplemental report demonstrates the following:

1. The finding in my initial report of major racial disparities in ballot rejection rates in Florida’s 2000 presidential election is confirmed by additional evidence of what actually happened in voter precincts in three additional counties.

2. The dissenters’ statistical consultant admitted before the Senate Committee on Rules that “a greater percentage of black and Hispanic people are turned away than, or don’t get to vote, than white people.” The dissenters concede that African Americans in Florida had their ballots rejected at a rate at least triple that of non-African Americans.  

3. Racial disparities in ballot rejection rates cannot be explained by differences between blacks and non-blacks in education, income, or any other factor pointed to by the dissenters.

4. The relationship between race and ballot rejection remains substantial and statistically significant even within comprehensive models with much greater explanatory power than any of the models presented by dissenters.

5. Although the dissenters offer education and literacy as explanations for ballot rejection, the statistical models developed by their consultant do not show the importance of these variables. These models also exclude other key variables, include redundant variables, explain relatively little of the variation in ballot rejection among counties, and are contradicted by precinct-level results.

6. The dissenting opinion, which relies heavily on Dr. Lott’s improperly designed and conducted statistical report, provides no credible discussion of the issues posed by the study of ballot rejection in Florida’s presidential election. 

1. THE FINDING OF MAJOR RACIAL DISPARITIES IN FLORIDA’S 2000 PRESIDENTIAL ELECTION IS CONFIRMED BY ADDITIONAL EVIDENCE AND IS NOT CONTRADICTED BY ANY ALTERNATIVE FINDINGS

In my initial report I wrote, “It should be stressed that the purpose of this study was to determine whether there existed in the Florida 2000 presidential election disparities between the ballot rejection rates of blacks and non-blacks. The purpose was not to establish the causes of any such disparities.” The results of my analyses, both of county-level data and of precinct-level data within several counties, demonstrated that there were major differences in the rate of ballot rejection for blacks and non-blacks in Florida’s 2000 presidential election. There is not a single alternative finding in the dissenting opinion or statistical report that even purports to show the lack of such race-linked disparities in ballot rejection. Indeed, the dissenters’ statistical consultant, John R. Lott, Jr., admitted the existence of such disparities in his testimony before the Senate Committee on Rules. The only numerical findings on the ballot rejection rates for African Americans and non-African Americans in either the dissenting opinion or the accompanying statistical report are numbers copied from my initial report. Thus the conclusions of my initial report stand without contradiction by any alternative results.

Since completing the original report I have gathered additional precinct-level data for Broward, Escambia, and Gadsden Counties.[2] These are important additions: Broward is the second most populous county in Florida. Escambia is a county with optical scanning technology recorded by precinct. Gadsden is the only majority-black county in Florida and had the highest ballot rejection rate in the state. In Broward County 14 percent of registered voters are black and 2.5 percent of ballots were rejected, slightly below the average for counties using punch card technology. In Escambia County 16 percent of registered voters are black and 3.6 percent of ballots were rejected, tied for second place among counties with optical scanning technology recorded by precinct. Escambia County had the largest number of rejected ballots (4372) for such counties, accounting for about a quarter of all rejected ballots for counties with optical scanning technology recorded by precinct. The analysis of Escambia County offers the first detailed glimpse using precinct-level data of the relationship between race and ballot rejection for counties with optical scanning technology recorded by precinct. It confirms the finding suggested in my first report of racial disparities in ballot rejection rates even among counties with the best available technology. In Gadsden County 54 percent of registered voters are black and 12.4 percent of ballots were rejected. The precinct-level data (for Broward, Duval, Escambia, Gadsden, Miami-Dade and Palm Beach Counties) now includes 59 percent of the statewide total of rejected ballots and 54 percent of the black registered voters in the state.

Precinct-by-precinct rejection rates and black voter percentages for Broward, Escambia, and Gadsden Counties are reported in Graphs 1S, 2S, and 3S (attached at the end of this report). These graphs also include the linear regression line to depict the relationship between race and ballot rejection. As indicated by the results of ecological regression and extreme case analysis reported in Table 1S and Charts 1S through 6S, the estimated rejection rates derived from precinct-level data in these three counties confirm the findings of the first report of major racial disparities in ballot rejection rates in Florida’s 2000 presidential election.

For Broward County, as demonstrated in Table 1S and Chart 1S, the rate of rejection for ballots cast by blacks was an estimated 6.2 percent, compared to an estimated rate of 1.8 percent for votes cast by non-blacks. As demonstrated by Table 1S and Chart 2S, results of extreme case analysis for 90%+ black and non-black precincts confirm the findings of ecological regression analysis for Broward County. In precincts that were 90 percent or more black the overall rate of rejection was 6.5 percent, compared to a rate of 2.0 percent for precincts that were 90 percent or more non-black.

For Escambia County, as demonstrated in Table 1S and Chart 3S, the overall rate of rejection for votes cast by blacks was an estimated 16.8 percent, compared to a rate of 1.7 percent for votes cast by non-blacks. As demonstrated by Table 1S and Chart 4S, results of extreme case analysis for 90%+ black and non-black precincts confirm the findings of ecological regression analysis for Escambia County. In precincts that were 90 percent or more black the overall rate of rejection was 13.7 percent, compared to a rate of 2.2 percent for precincts that were 90 percent or more non-black.

The Escambia County results powerfully confirm race-related discrepancies in ballot rejection even among counties with the best available technology in Florida’s 2000 presidential election. These results also sustain the conclusion that improved technology is not the complete answer to reducing rates of ballot rejection and diminishing race- related discrepancies. Press reports indicate that Escambia County might have turned off its precinct check on overvoting for reasons of economy, confirming the importance of resources and training. The findings for Escambia County reveal large, statistically significant differences in the rates of rejection for blacks and whites. The difference of 15 percentage points in estimated rates of ballot rejection for black and white voters exceeds the discrepancy of about 13 percent estimated from county-level data for all Florida counties, and is far greater than the discrepancy of about 4.5 percent estimated from county-level data for counties with optical scanning precinct-recorded technology.

For Gadsden County, as demonstrated in Table 1S and Chart 5S, the overall rate of rejection for votes cast by blacks was an estimated 21.6 percent, compared to a rate of 4.4 percent for votes cast by non-blacks. The difference of 17 percentage points in estimated rates of ballot rejection between black and white voters is second only to Duval County among the six counties examined. As demonstrated by Table 1S and Chart 6S, results of extreme case analysis for 90%+ black precincts confirm the findings of ecological regression analysis for Gadsden County. In precincts that were 90 percent or more black the overall rate of rejection was 22.8 percent. There were no 90%+ non-black precincts in Gadsden County, although the two counties that were 88% non-black had rejection rates of 5.5 percent.

For all six counties examined with precinct-level results, the rate of rejected ballots by African Americans ranged from about 6.5 percent to about 24 percent. For all six counties, the unweighted mean rejection rate for ballots cast by blacks was 16.9 percent. In contrast, the ballot rejection rate for non-African Americans ranged from about 2 percent to 6 percent. The unweighted mean rejection rate for ballots cast by non-blacks was 3.7 percent, for a gap of 13.2 percentage points. For all six counties, the weighted mean rejection rate for ballots cast by blacks was 14.0 percent, meaning that nearly one of seven African Americans who entered the polling booth in these counties had their ballots rejected as invalid. The weighted mean rejection rate for non-blacks was 3.5 percent, for a gap of 10.5 percent. These results are comparable to the difference in ballot rejection rates of 12.8 percent for blacks and non-blacks statewide derived from the county-level ecological regression analysis of my first report. The racial divide of 10.5 percent net means that as compared to non-blacks 30,000 additional African Americans had their ballots rejected in these six counties alone. Statewide, a racial gap of 10.5 percent would mean that as compared to non-blacks, more than 60,000 additional African American voters had their ballots rejected in Florida’s 2000 presidential election.

In response to the county-level and precinct-level findings in my initial report of major racial disparities in ballot rejection rates, the dissenters and their statistical expert present not a single alternative numerical estimate of the ballot rejection rate for African Americans in Florida’s 2000 presidential election. The dissenters and their statistical consultants inexplicably fail to examine any of the precinct-level data that is available in Florida, even though such data provides crucial tests of county-level models and direct information on actual ballot rejection rates for overwhelmingly black and non-black precincts included within counties where purportedly the same voting technology was used in each precinct.

2. RACIAL DISPARITIES IN BALLOT REJECTION RATES CANNOT BE EXPLAINED BY DIFFERENCES BETWEEN BLACKS AND NON-BLACKS IN EDUCATION, LITERACY, INCOME, POVERTY OR ANY OTHER FACTOR CITED BY DISSENTERS

The only statistical analysis presented by dissenters is in their accompanying statistical report prepared by John R. Lott. However, Lott’s report ignores the central issues of racial disparities in ballot rejection for Florida’s 2000 presidential election. Instead it addresses the separate issue of whether such disparities can be attributed to other factors. The analyses presented below will demonstrate that differences in literacy, education, income, or poverty do not account for the major differences in ballot rejection rates for African Americans and non-African Americans in the presidential election of 2000. The relationship between race and ballot rejection remains substantial and statistically significant even after controlling for such variables as well as for many other factors, including measurements of first-time voting.

Table 2S below reports a county-level multiple regression analysis that, unlike the statistical report presented by dissenters, controls directly for literacy, education, poverty, and income for all counties. In addition, these equations also control for differences in technology: whether a county uses punch cards, optical scanning centrally recorded, or paper ballots or machines. The influence of these technological variables is measured against the remaining system used in Florida: optical scanning by precinct. An additional analysis, presented in Table 3S, examines the issue of the influence of education and first-time voting on ballot rejection rates for precincts within Miami-Dade County.[3]

The county-level findings reported in Table 2S show that racial differences in ballot rejection rates in the 2000 presidential election are not reducible to differences between blacks and non-blacks in income, poverty, education, or literacy. To the contrary, as indicated by the preliminary analysis presented in my initial report, controlling for socio-economic factors fails “to diminish the relationship between race and ballot rejection or to reduce the statistical significance of the relationship.” As demonstrated in Table 2S, even after controlling for a wider array of socio-economic variables than any of the dissenters’ models, the relationship between race and ballot rejection remains substantial and statistically significant at levels beyond the stringent .01 standard used in social science.[4] The regression coefficient for the percentage of black voters, controlling for all variables in Table 2S, is .140, which corresponds to a difference in ballot rejection of 14.0 percentage points between blacks and non-blacks, holding constant the variables included in the equation. This means that independent of income, poverty rates, education, literacy, and the technology for voting, there is a double-digit difference in ballot rejection rates between African Americans and non-African Americans. This result is comparable to the difference in ballot rejection rates of 12.8 percent for blacks and non-blacks derived from the ecological regression analysis of my first report.[5]

The findings of an analysis of precinct-level data from Miami-Dade County, reported in Table 3S, confirms that the relationship between race and ballot rejection is independent of educational levels. In Miami-Dade County, which includes numerous black and Hispanic voters, a greater percentage of Hispanics than African Americans had less than a ninth grade education—the only socio-economic variable that was statistically significant in the analysis reported in Table 2S above. The percentages are 27 percent for Hispanics and 18 percent for African Americans. Moreover, Hispanics face additional language barriers to voting. If education were responsible for differences in ballot rejection, the Hispanic rate of ballot rejection in Miami-Dade County should be comparable to or even greater than the black rate. As demonstrated in my initial report, these expectations are not confirmed. The heavily African American precincts have a much higher rejection rate than the heavily non-African American precincts. However, the heavily Hispanic precincts have a lower rejection rate than the heavily non-Hispanic precincts (many of which are African American).

Table 3S provides a more refined analysis of the influence of race and ethnicity on ballot rejection in Miami-Dade County. It reports the results of a multiple regression analysis, based on precinct-level data, which estimates the percentages of rejected ballots of African American, Hispanic, and white voters in each precinct. The results of analysis shows that the black rejection rate is substantially higher not only than the white rate but also than the Hispanic rate as well. The rejection rate for whites is only 1.7 percent. The Hispanic rate is higher at 4.2 percent, whereas the African American rate is 10.0 percent, almost 6 percentage points higher than the Hispanic rate and more than 8 percentage points higher than the white rate.

The data available in Miami-Dade County also provides an opportunity to extend the regression model to include a measure that in part captures the phenomenon of first-time voting: the percentage of registrants aged 17 to 20. Virtually all of those voting from this group will be first-time voters in 2000, although, of course, first-time voters could also belong to other age groups. The results of analysis, reported in Table 4S, show that when controlling for race, the coefficient for the percentage of 17- to 20-year-old registrants is negative. In contrast, the coefficient for black voters is positive, substantial, and statistically significant, indicating a 9.3 percentage point gap between black and white rejection rates, controlling for first-time registrants. The coefficient for Hispanics reveals a smaller, but still statistically significant gap of 2.7 percentage points.

In sum, the results of analyses at both the county-level and precinct-level decisively reject the proposition—asserted but never fully tested by the dissenters—that differences between African American and non-African American rates of ballot rejection are a function of socio-economic factors. To the contrary, racial differentials in ballot rejection rates are virtually unaffected by controls for literacy, education, income, and poverty. Moreover, at least a partial control for first-time voting, using precinct-level data from Dade County shows no influence on the relationship between race and ballot rejection.

3. THE RELATIONSHIP BETWEEN RACE AND BALLOT REJECTION REMAINS SUBSTANTIAL AND STATISTICALLY SIGNIFICANT EVEN WITHIN COMPREHENSIVE MODELS WITH FAR GREATER EXPLANATORY POWER THAN ANY MODELS PRESENTED BY THE DISSENTERS

The results of estimating a more comprehensive model of ballot rejection in Florida counties than that presented in Table 2S is reported in Table 5S. This model surpasses the effort to control for socio-economic factors by also including the increase in the vote cast between the elections of 1996 and 2000, the turnout of registered voters in 2000, the percentage of the presidential vote received by the Democratic candidate, the ratio of voters to precincts in each county, and whether the election supervisor is Republican or Democratic.

This model does a far better job than any of the models in the dissenters’ statistical supplement in accounting for changes from county to county in the percentage of rejected ballots in Florida’s 2000 presidential election. With an R² value of .866, this model accounts for 86.6 percent of the variation from county to county in ballot rejection rates for the 2000 presidential election. The models in the dissenters’ statistical report (Table 2) explain only from 73.4 percent to 79.5 percent of the variation from county to county in ballot rejection rates for the 2000 presidential election.

Despite the stringent controls included in this model, the relationship between race and ballot rejection is substantial and statistically significant at levels beyond the stringent .01 standard used in social science. The regression coefficient for the percentage of black voters, reported in Table 5S, is .143, which corresponds to a difference in ballot rejection of 14.3 percentage points between blacks and non-blacks, controlling for the variables in the equation. This coefficient value is almost identical to the coefficient reported above without the additional variables.[6] None of the socio-economic variables in this comprehensive model, however, have a statistically significant influence on ballot rejection rates. Neither does the variable measuring changes in voter turnout between 1996 and 2000, a variable that would partly capture the phenomenon of first-time voting.

Similarly, the political identity of election supervisors has no discernible influence on ballot rejection rates in the comprehensive model. Variables measuring whether the supervisor is Republican or Democrat both have negative signs (relative to non-partisan supervisors), and fail to approach conventional levels of statistical significance. In contrast, controlling for the factors included in the equations of Table 5S, the relationship between ballot rejection rates and the Democratic vote in the 2000 presidential election is negative and statistically significant at .05. This indicates that as Gore strength declines, ballot rejection rates are higher than would be expected based on the other variables in the model.

The findings of this study that the relationship between race and ballot rejection remains substantial and statistically significant even under stringent controls is confirmed by other independent analyses, including one performed by Philip A. Klinkner, Associate Professor of Political Science of Hamilton College, and submitted to the United States Senate Committee on Rules.[7] All his models explain far more variation than any of the models in the statistical report presented by the dissenters. Professor Klinkner found that for every model studied, the relationship between the percentage of black registered voters and the percentage of rejected ballots remained substantial and statistically significant. Professor Klinkner concludes, “While my data and findings were arrived at independently, these findings are essentially the same as those of the U.S. Commission on Civil Rights (USCCR). Thus, my data and findings contradict the accusations that the USCCR conducted a biased survey with inaccessible data.”[8]

Based on an analysis of Florida’s individual voter files for 2000, Professor Klinkner has also provided county-by-county percentages of first-time voters.[9] Results reported in Table 6S demonstrate that substituting this direct measure of first-time voting into the model for change in voting between 1996 and 2000 produces virtually no change in the relationship between race and ballot rejection.[10] According to Table 6S, the coefficient for the percentage of black voters is statistically significant beyond conventional levels and has a value of .137, corresponding to a difference in ballot rejection of 13.7 percentage points between blacks and non-blacks, controlling for the variables in the equation. The coefficient measuring the relationship between first-time voting and ballot rejection is negative and falls far short of statistical significance. None of the other variables from Table 5S show any substantial change in Table 6S. The model in Table 6S also explains 87.4 percent of the variance in rejection rates, far greater than any of the models presented by Dr. Lott. Thus the hypotheses presented by the dissenters regarding the alleged effects of income, poverty, education, literacy, or first-time voting do not withstand scrutiny. The inclusion of these variables in an analysis estimating ballot rejection rates does not diminish the relationship between race and ballot rejection in Florida counties for the 2000 presidential election.

4. THE STATISTICAL REPORT PRESENTED BY DISSENTERS PROVIDES NO CREDIBLE MODELS OF BALLOT REJECTION IN FLORIDA’S 2000 PRESIDENTIAL ELECTION 

The statistical report commissioned by dissenters includes only one table that provides the results of estimating models of ballot rejection in the presidential election of 2000. These estimates, which are for county-level data, are in Table 3 of the dissenters’statistical report, which includes 16 models. The results of Models 1 through 8, which include the literacy variable, are fully reported in Table 3. The results of Models 9 through 16, which do not include literacy, are only partially reported in Table 3. These models are nearly, but not quite identical, to the 8 models presented in Dr. Lott’s original report. Dr. Lott’s models, at a minimum, suffer from the following flaws:

1. The dissenters’ models lack conceptual foundation. Among other problems they omit key variables that are essential to hypotheses advanced in the dissenters’ written opinion, including measures of education and first-time voting.

2. The dissenters’ models include duplicative measures of the racial composition of county, destroying the integrity of the effort to gauge the independent influence of race on ballot rejection rates.

3. The dissenters’ models explain far less of the variation in ballot rejection than the models developed by Professor Klinkner and myself.

4. The dissenters’ models produce results that are internally contradictory and conflict with what we actually know happened at the precinct level in Florida’s 2000 presidential election.

The dissenting opinion—as opposed to the statistical report—in both its initial and revised form cited education, literacy, and first-time voting as the key explanatory factors accounting for the relationship between race and ballot rejection in Florida’s 2000 presidential election (see pp. 21–24). Despite this emphasis on education, literacy, and first-time voting, of the 8 statistical models of ballot rejection in the 2000 presidential election in Dr. Lott’s initial report, not a single model included variables measuring education, literacy, or first-time voting. He does not explain his exclusion of education and first-time voting, but does attempt to justify his omission of literacy by claiming that my initial report “does not reference data on literacy rates.” (Lott report, p. 4). Yet in my initial report I fully defined my literacy variable, precisely reported its source, and provided printouts of the data for each of the 67 Florida counties. I provided the same information for my use of educational data. Indeed, the dissenters’ own report, as distinct from Dr. Lott’s statistical supplement, includes an extensive discussion of my use of literacy data. In the current revision of his report, Dr. Lott somehow discovers the literacy data and includes it in an additional eight models that he presents in his revised report.