Are Polls Accurate Predictors of Election Outcomes?

The NY Times reports polling results in the battleground states as “extremely close,” Nate Cohen, the Times polling guru is writing about polling in his column, The Tilt, almost every day.  Of course Nate was “wrong” in 2016 and 2020 and to his credit takes a mea culpa, he tries to explain why the polling universe, were so wrong, he meanders through a long list of possible reasons, actually a tortuous journey. If the polls were wrong in 2016 and 2020 do we have any confidence they’ll be better in 2024?

What is a poll?

Polls predict outcomes at a moment in time, with a statistical margin of error. The pollsters create a microcosm of a pool of “likely voters” and the pool mirrors characteristics of the population, referred to as a “stratified random sample.” The greater the number of characteristics the more accurate the poll: perhaps age, race, gender, level of education, employment status, income, etc.

The pollster randomly selects “likely” voters from the sample and asks a series of questions, The questions can be which candidate do you support and other questions, perhaps most important policies choices, depth of support for candidate, etc., every poll has a margin of error expressed as + and – a number, plus or minus 4% is an 8% swing.

Unfortunately the NY Times fails to report the term “statistical tie,” a one percent difference is a statistical tie. if a margin of error is greater than 1%.

Pollsters use similar methodologies and in 2016 a week before the election Cohen predicted a Hillary win.

In our era of cell phones and caller ID who is more likely to answer a call? a senior, a woman? a new voter?  Pollsters “weight” samples hoping to increase accuracy.

Statisticians are leery of polling as a science, and ask,

To what extent are pollsters using algorithms to select a polling population? And if so, does it increase accuracy?

If you’re a visual learner the Khan Academy posts on U-Tube an explanation of “random sampling” and other elements of polling.

 The Pew Trust takes a deep dive into the science, or not so scientific world of polling here and is highly critical and lists many flaws. Worth a read!

The number of polling data providers has increased sharply, polling data is sold and polling data gets high clicks on the web.

The impact of polling data itself may influence election outcomes.

How many voters use reporting on polling to drive their candidate selection?

To what extent do candidates use reported polling data to drive policy decisions?

Of course it’s only a matter of time before the microchip in our earlobe will enable us to vote on every important issue, then again that assumes elections will continue