Statistical Analysis of the Impact of Voter ID in November 2012 Election

This page includes material supplemental to the article titled:

Pennsylvania AFL-CIO Analysis Indicates That Over 35,000 Voters Were Disenfranchised In 2012 General Election As A Result Of Voter ID

Definitions of data and terms associated with voters:

  • 2012 Vote Propensity: This score is a proprietary calculation of relative voting likelihood, and is based on a variety of publicly available data.  It was calculated prior to the 2012 general election, and is used in this analysis to ensure that we compared treatment and control voters who had relatively similar voting tendencies.  There are some voters who, either because of a lack of voting history or other factors, did not have a 2012 vote propensity score calculated for them.  For the integrity of our analysis we excluded any such voters.
  • Treatment Voters: These are voters who were initially identified by the Pennsylvania Department of State in the spring of 2012 as potentially lacking a valid State-issued ID.  This list of approximately 758,000 voters included the ones targeted by the State for additional informational mailings about the Voter ID law.  Not all of these voters lacked ID, but they did receive letters and postcards from the State indicating that there was concern over their ID status and that they may not be able to vote as a result.  This list was obtained from the Pennsylvania Department of State in July of 2012.
  • Control Voters: These are the remaining registered voters in Pennsylvania, who were matched to State-issued IDs in the PennDOT database, and were not targeted by the State for additional mailings about their ID status.  For the purpose of this analysis we used a copy of the voter file obtained from the Pennsylvania Department of State in the Spring of 2012, near the time that the Department of State conducted their match with PennDOT records.  Because of this, we are not including in our analysis voters who registered after this time.
  • 2012 Vote History: The data regarding which voters cast ballots in the November 2012 election was obtained from the Pennsylvania Department of State in July of 2013.  Because of the need to match voters from our database with their vote history in the 2012 election, any registered voters from our earlier database who were removed from the voter rolls prior to July of 2013 were not included in our analysis.
  • Note About Sample Size: Because our analysis only included voters who were present in both the 2012 and 2013 voter file exports, and because we included only voters who had a previously calculated 2012 vote propensity score, our full voter universe contained 7,379,811 registered voters, rather than the full list of voters in Pennsylvania.  We did not attempt to project our results from this sample to the full population of registered voters, meaning that we certainly underestimate the number of disenfranchised voters in this analysis.

Data Table

Class Control Voters Treatment Voters Control Voters Predicted Turnout Treatment Voters
2012 Vote
Registered Votes Cast Registered Votes Cast Turnout
99% CI* Turnout
Vote Shortfall
(99% CI*)
0 – 9.9  151,353 15,588 42,013 2,300 10.3% 10.1% to 10.5% 5.5% 1943 to 2111
10 – 19.9 242,908 41,325 31,008 3,136 17.0% 16.8% to 17.2% 10.1% 2079 to 2200
20 – 29.9 255,003 56,769 33,836 3,577 22.3% 22.1% to 22.5% 10.6% 3884 to 4027
30 – 39.9 280,002 74,279 23,742 3,831 26.5% 26.3% to 26.7% 16.1% 2416 to 2518
40 – 49.9  348,556 125,579 31,638 5,949 36.0% 35.8% to 36.2% 18.8% 5384 to 5516
50 – 59.9  353,551  167,426  28,956  9,908  47.4%  47.1% to 47.6%  34.2%  3742 to 3867
60 – 69.9  442,966  256,506  34,531  16,326  57.9%  57.7% to 58.1%  47.3%  3604 to 3736
70 – 79.9  852,367  594,950  60,463  37,147  69.8%  69.7% to 69.9%  61.4%  4979 to 5133
80 – 89.9  1,267,740  1,086,697  86,191  69,948  85.7%  85.6% to 85.8%  81.2%  3865 to 4003
90 – 99.9  2,566,292  2,459,302  246,695  232,987  95.8%  95.8% to 95.9%  94.4%  3344 to 3502
Totals  6,760,738  4,878,421  619,073  385,109  72.2%  n/a  62.2%  35,239 to 36,613

*99% CI: Confidence Interval – this measure indicates that with a 99% degree of statistical certainty, the true value of the indicated measure would fall within the range calculated, and only a 1% chance that the true value would be higher or lower than this range.

The 99% Confidence interval was calculated using a standard statistical test shown below.  Voters from each 2012 vote propensity category were analyzed independently so that we were only comparing control and treatment groups who had similar vote propensities.  After the predicted turnout was calculated for each tier, we calculated the shortfall between the anticipated vote and the actual vote among treatment voters.  Those shortfalls ranges for each tier were totaled to arrive at the total number of disenfranchised voters.

p = the observed proportion

n = the sample size
z = 2.57 (standard for calculating a 99% CI)

Graph 1: Distribution of voters in the control and treatment groups by their vote propensity scores.  Propensity scores were rounded to the nearest whole number, and the graph indicates the percent of voters from each group with that score.  While our analysis standardized for vote propensity when comparing turnout between the control and treatment groups, it is nevertheless interesting to observe how similar the vote propensity distributions are for the two subsets of voters.

Graph 2: Observed voter turnout, by vote propensity category, for control and treatment groups.  The much more pronounced impact among what would be considered ‘sporadic’ voters, and the relatively low impact among high propensity voters is consistent with what would be expected if the results were due to confusion about the status of the Voter ID law, where higher propensity voters may be more likely to seek out additional information and find out that the law had be blocked.