Archive for October, 2018

Why I Don’t Trust Pre-Election Voter Surveys

October 29, 2018

Models and history no longer work, any more than weather forecasts after a huge volcanic eruption, because the assumptions under which they work no longer apply. (Yes, Trump IS the volcanic eruption.)

Surveys no longer can be relied on because the samples they pull are no longer representative.

Random telephone book samples no longer work, because too many potential respondents no longer have landline phones, and those that do have caller ID and don’t answer calls from unknown numbers.

Cellular phones are not reliably listed anywhere in numbers sufficient to produce answers that are statistically valid, and users also tend not to answer unknown numbers, further reducing the respondent pool.

The geographic confines of a cellular number sample are no longer valid, because cellular users can retain their phone numbers across area codes and can’t be used to reflect opinions in a given Congressional district, county or even state.

Until representative samples can be found that are large enough to be statistically valid (1,000, at a bare minimum, for a geographic area or demographic group), voter preference surveys cannot be trusted. As of now, the only way to get that kind of sample is to send teams out into the field, knocking on doors. That process is so time-consuming and expensive that a sample may have changed based on events or publicity.

I would rather bet on my instincts than rely on the results of any pre-election survey.

Oh, and I did take a Stats class in grad school. Passed it. Not by a lot, but I passed it.

Advertisements