As much as I feel anxious about the election, I'm excited for the polling drama this time around. Feels like the whole industry is on the verge of meltdown: you've got the influx of low-quality partisan polls; you've got aggregators who are accusing the polling industry of hedging (even Silver got on this earlier this week after defending it); you've got the suspicious relationship between polls and betting markets (again, Silver at the center of it); response rates have dropped from 80% in the 90s to 10% a decade ago to apparently 1% for some pollsters now. Pollsters have been trying to keep up by increasingly focusing on weighting poll results based on anticipated turnout of different demographics, but in an extremely entrenched election where it's all about turnout, how do you even anticipate turnout when the poll response rate is so low?
If I could just totally rebuilt the industry, I'd entirely separate out the process of conducting surveys and the process of building polling models. Have survey data go into an industry-wide collective aggregate that all pollsters can then select based on their preferred methodologies, and build their own turnout models. This creates a layer of transparency and reproduceability of results. Stop talking about margin of error, which is an antiquated idea that creates a false sense of security.
|