Here's how one expert tweaked the USC/L.A. Times poll so it matches the averages

Donald Trump‘s consistent lead in the USC/L.A. Times “Daybreak” poll of the election — and the gap between the poll and all other major surveys — has drawn a lot of attention. 

Most of the comments have come from partisans, either Trump supporters praising the poll for telling them what they’d like to believe or Clinton supporters attacking it for the opposite reason.

Ernie Tedeschi, an economist and former Treasury Department official who is adept at data analysis, looked at the issue differently. He wondered what was driving the difference between the Daybreak poll and other surveys.

He found that if he took the poll’s data and weighted it differently, what had been a tie between Trump and Clinton suddenly became a Clinton lead. Indeed, his version has the poll showing her in the lead almost the entire summer and fall. 

Moreover, the reweighted version almost precisely matches the path of the Real Clear Politics average of all polls

So who is right?

As Tedeschi noted, that’s something we can’t know until the election is over. The Daybreak poll could be “picking up a signal no one else is,” he wrote.

But his work provide reassurance that although the poll differs from other surveys, its data about the trends in the election — the ups and downs in support for the two candidates — are consistent with what others have found.

That also suggests that the poll’s underlying sample — the panel of 3,200 people who are queried each week about the election — is sound.

What Tedeschi did demonstrates how much difference the fairly technical topic of weighting a poll can make.

Weighting is something that pollsters routinely do in order to make sure that their samples properly match the U.S. population, as we explained in our Frequently Asked Questions about the poll.

There’s no single, universally agreed upon way to weight a sample. The process involves numerous decisions, all of which pose tradeoffs: Weight too little, and the poll can fail to represent the diversity of the U.S. population. Weight too much, and the poll can be distorted.

The details on how the poll is weighted are available from USC. Tedeschi also described how he did his reweighting of the data.

The main difference between the original weighing and what Tedeschi did is that he dropped some weights the USC pollsters use, most notably their decision to weight the sample to match people’s reports of whether they voted in 2012 and, if so, for whom.

It’s also worth noting that Tedeschi’s work underscores the importance of USC’s decision to make all the data from the poll available to the public — something that, unfortunately, not all pollsters do.

Source link