Thread by @DrEricDing

FOLKS—Polling sucks. Amateur-“pro”-political polls as we know it—is done. Polls had said Biden led FL w/ 70% chance of winning—it became a wide Trump win.

➡️Most pollsters have no expertise in adv epidemiology & biostatistics. I would change it.🧵 #polls thehill.com/homenews/media…

2) Many epidemiologists have long worried political polling is overconfident, uncertainties underestimated, & state/local factors poorly adjusted. Now we’re seeing it play out—again.

I know someone who paid $25k for what ended up as crap n=400 poll that’s worthless.

3) Many sources of unaccounted for uncertainties in polls:

📌SAMPLING (landline vs cell vs other?)

📌RESPONSE BIAS (shy respondent)

📌TURNOUT MODEL ASSUMPTIONS (who shows up)

📌STATISTICAL WEIGHTING adds extra variance (error)

📌UNDERSAMPLING minorities for reliable weights

4) another problem, as my better half @andreafeigl1 points out, is the classic “interviewer effect” (also called interviewer variance or interviewer error)—distortion of response to a personal or telephone interview...

5) ...”which results from differential reactions to the social style and personality of interviewers or to their presentation of particular questions.”

Basically data gathered from interviewing people is affected by the behavior (real or perceived) of the interviewer / method.

6) Maybe it’s a Trump era phenomenon too with shy Trump voters. But maybe not shyness per se, but a different accent of the pollster interviewer making the rural respondent uncomfortable (think Yankee speaking accent to a Southern listener) about disclosing true views on things.