Geographic concentration is central to a possible weakness in Michigan’s Senate polling: an issue can sharply affect one community while barely moving otherwise similar voters elsewhere. Statewide demographic balance may miss that local response.
Michigan’s Democratic primary provides a warning. Call the Map reports that late polls suggested a double-digit El-Sayed victory; his actual margin over Haley Stevens was approximately one point. Stevens led by 34.7 points in West Bloomfield, while El-Sayed led by 22.3 points in Royal Oak—a 57-point difference.
The entire 57-point gap cannot be attributed to the difference in the size of the Jewish population. The population of Royal Oak is younger than West Bloomfield and that also increased El-Sayed’s share of the vote.
These communities have broadly comparable populations and nearly identical per capita incomes, but West Bloomfield has a much larger Jewish community. My hypothesis is that geographic concentration amplified the electoral response to the Temple Israel attack and El-Sayed’s reaction through shared institutions and community networks. The contrast warrants investigation, although age, ideology, and Stevens’s local connections offer alternative explanations.
In “When Should You Adjust Standard Errors for Clustering?”, published in the Quarterly Journal of Economics, Abadie and colleagues explain why clustering must be evaluated through sampling and research design.
The concern tonight is whether polls adequately capture this geographically concentrated electorate’s preferences and turnout.
I am considering a journal article on this topic but wanted to share this preliminary concern on election night.
Author’s note: I occasionally consult on economic issues and for political campaigns.