Here is a guest post from a friend of PunditTracker.
The American Political Science Association (APSA) has issued their latest quarterly publication on the front end of Election 2012. Within PS: Political Science and Politics, a collection of political scientists shared their predictive modeling for the current presidential campaign. And if it lacks the drama of Drudge’s flashing red light, it nonetheless is an intriguing read.
All thirteen of the featured academicians are predicting a close race. Five scholars are calling an Obama victory in the popular vote. Five forecasters are modeling a win for Romney. Three are predicting a toss up.
|1. Abramowitz (Emory)||Toss Up|
|2. Campbell (U of Buffalo)||Obama|
|3. Cuzan (West Florida)||Romney|
|4. Erikson (Columbia) and Wlezien (Temple)||Obama|
|5. Hibbs (retired)||Romney|
|6. Holbrook (Wisconsin)||Romney|
|7. Lewis-Beck (Iowa) and Tien (Hunter)||Romney|
|8. Lockerbie (East Carolina)||Obama|
|9. Norpoth (Stony Brook) and Bednarczuk (Wisconsin)||Obama|
|10. Montgomery (Wash U), Hollenbach (Duke), Ward (Duke)||Toss Up|
|11. Berry (U of Colorado) and Bickers (U of Colorado)||Romney|
|12. Jerome (Paris 2) and Jerome-Speziari (Paris 2)||Obama|
|13. Klarner (Indiana State)||Toss Up|
The forecast range is from 53.8% for Obama to 53.1% for Romney. The predictions were finalized anywhere from 57 days (both Campbell/Montgomery and Hollenbach/Ward) to 299 days (Norpoth/Bednarczuk) prior to the 6 November election date. The certainty levels range from 10% (Hibbs) to 88% (Norpoth/Bednarczuk).
Predictive modeling within the political realm is enjoying a multi-decade ascent. The bull argument, as framed by University of Buffalo’s James Campbell, is threefold. One, there are “general influences” that consistently impact electoral behavior. Second, they are both knowable and readily measured. Third, this data has a historical basis allowing for comparability over time.
There are of course limitations to what we can reasonably predict about the political future. Three key challenges, amongst many complexities, include:
- 1) Data Selection: Even if you believe there is a meaningful relationship between the economy and elections, a diverse array of modeling choices still remain. Should we use a single indicator or multiple indicators? Which indicator or set of indicators from U-3 unemployment to gross disposable income to GDP should be leveraged? The choices are manifold and, as we see in above forecast ranges, have a statistically significant predictive impact.
- 2) Data Accuracy: Both preference polling and even seemingly hard data like economic indicators also have clear sampling gaps. On the former, we see dynamics like the so-called House Effect where certain polling teams have a bias (from question framing to turnout assumptions). On the latter, even with hard data sets like Gross Domestic Product (GDP), we minimally see advanced, preliminary and final reports each quarter. And of course, the truth is the data is serially revised for many years.
- 3) Campaign Impact: Finally, campaigns matter. A confluence of variables can complicate even the best forecasting model, including candidate performance over the lifecycle of the race, fund-raising prowess and growing sophistication around microtargeting and mobilization.
We have a tendency to overstate our ability to predict, particularly in building models that allow for uncertainties. Notwithstanding that, the work of these academicians to bring some of the precision of the hard sciences to the political realm is a laudable one. And without Drudge’s red ink, bolded type or flashing lights, we will have a chance to reflect on their level of predictive success in just 21 days.
Donald Trigg is a “friend of PunditTracker.” Prior to his time in the private sector, Trigg spent a decade in the public policy space including his work on the 2000 Bush for President campaign in Austin, Texas.