Improving the EAC's surveys

Yesterday Thad and I participated in some interesting discussions and conversations with a number of folks regarding how to possibly improve on the survey research efforts of the Election Assistance Commission (EAC). In informal remarks to the assembled individuals (including some election administrators, users of the EAC survey data, and representatives from the EAC), I addressed four questions.

  1. What are the challenges in collecting election data? Here I said that there were five basic challenges that we have encountered in recent years trying to collect election data:

    1. Data availability — all too often, important data is not available, even simple statistics like the number of ballots cast at the county level in some states.
    2. Inconsistent reporting — as we are learning, many states have different definitions and conceptions for various procedures, for example, what it means to be a registered voter and how they treat different types of registration status when they report registration data.
    3. Errors in data — sometimes data is reported incorrectly.
    4. Aggregation — in most cases, important precinct-level data is not available.
    5. Election administration — we have very little reliable data about basic administrative issues, like governance and cost.
  2. What are the key data elements for social science research? Here I discussed three key data elements for current research:

    1. Election outcome statistics: registration, participation, votes on all candidates and measures.
    2. Data by voting mode (precinct voting, early voting, absentee voting, UOCAVA voting).
    3. Administrative data (costs, personnel, training, voting equipment, for example).
  3. How are election survey used and abused? Here I primarily concentrated on some problems we have seen in the use of elections data:

    1. Aggregation and ecological inference — the use of aggregate data to make inferences about individual-level behavior.
    2. Statistical abuses — model overfitting and the ignorance of historical data.
    3. Incorrect use of demographic statistics — in most cases, demographic statistics are reported for populations, not for the registered or voting electorate (for example, we rarely know the numbers of white or nonwhite registered voters in a jurisdiction, unless we try to estimate those quantitites).
    4. A lack of experimental data.
  4. What possibilities exist for collaboration on survey data collection? Here I talked about a number of ideas. One idea is to simply note that election officials can and should reach out to the research community for help; as I noted, we often are quite interested in collecting data, and we sometimes have the financial support and personnel to support data collection. Another idea was to exploit possible collaborations between social science infrastructure for data collection, for example, the American National Election Studies or the General Social Studies. A third idea was to think outside the box about how we can use new technologies to collect data (there will be more discussion of this in our report).

The good news from these conversations yesterday is that the EAC is committed to continuing it’s efforts to collect and distribute high-quality election and election administration data, in 2006 and 2008.

Thad and I also presented a draft report, “Improving the Election Day Survey”. We will finalize and release the final report in the next few days, and will of course write more about our conclusions regarding the EAC Election Day Survey at that time.