Data Gathering

This pilot study aims to capture data in 30 precincts in three counties: Alameda, Napa, and San Mateo (each representing a different technology).  Our sample size is small (n≈923), but should yield statistically important results.  In the three counties we have chosen, the average number of voters per precinct is 577 (min = 503, max = 680).  From a practical standpoint, every 15 second difference in mean ballot completion time translates into an extra hour of waiting at the end of the day.  Therefore, we conducted a power analysis based on the expected number of voters (1,325,667 voters)¹ with a mean of 3.75 minutes² and a standard deviation of 15 seconds.  By observing 10 precincts in each county, we will be able to detect small to medium effects of machine type on wait times when compared between counties (r≈.20).³

                                

Our data gatherers will record 1) voter arrival rate, 2) the exact time every 5th voter checks in, enters and leaves the voting booth, and 3) the frequency and point of origin of lines.  See Figure 1.  We will use a regression model to examine relationships between characteristics of interest and wait times.  We will also update elections queuing models with more reliable baseline data to predict wait times under simulated parameters.

                                      

Finally, we will augment our study findings with cost estimates of various components of the voting process, with special attention paid to voting technology, as it tends to be the significant driver of costs.  We conducted a cursory cost comparison of a DRE touch-screen system and a paper ballot, optically scanned at the precinct-level and found that the cost-per-vote using a DRE was 2.4 times larger than for the paper ballot with optical scan.  See Figure 2.

---

¹   Number of registered voters in the four counties as of February 10, 2007.  See http://sos.ca.gov/elections/ror/ror_oddyr/county.pdf.
²   If voters were equally distributed and arrived at a constant rate, each would have to cast his ballot in 3.75 minutes to avoid any lines at the end of the day.
³   Based on Cohen’s power table for one-way ANOVA testing with four distinct groups.