Saturday, August 5, 2017

Trump and the polls

Many stories have been written this year saying Trump voters are still happy with Trump. Almost none have been written about Clinton voters still pissed that she won by nearly 3,000,000 votes and over 2% of the popular votes and Trump was still installed as president by the seriously anti-democratic Electoral College system.

In contrast to these anecdotes, polls  make the attempt to gauge the general public opinion using something approximating scientific methods. The big failure of poll-based prediction in 2016 makes me less confident in these numbers, but statistical methods never promise certainty. That said, the polling numbers for Trump's popularity after six months in office show a public growing quickly disenchanted.

I follow 22 different polling companies, getting their results from the Pollster page funded by The Huffington Post, but I have more confidence in looking at six companies that poll every week or even more often. The two tracking polls that update almost every day are Gallup and Rasmussen. The four polls that give weekly numbers are Politico, SurveyMonkey, YouGov and Ipsos/Reuters. I never consider any one polling company to be the most reliable, but I would rank these six at least as reliable as the companies that poll only two or three times a month or even less and much more reliable than the very sporadic pollsters.

The graph speaks for itself. While there are ups and downs in the average written in blue and the median written in red, the general trend is downhill. Since the middle of July, the numbers have taken a steep fall. On July 11, Trump's net popularity averaged -11 percentage points and the median was -13.5 points. As August began, those numbers sunk to -19.7 on average and a median of -21.5 percentage points.

On the left is slope graph for the six companies, showing their net numbers on January 31 and July 31. Two points are difficult to read due to exact overlap. On the far right, both YouGov and Ipsos/Reuters had Trump at -1 point net in January, while Rasmussen and Gallup now concur that Trump is at -22 percentage points when the unfavorable number is subtracted from favorable.

The first and most obvious point is that everything is downhill. Politico, represented by the light blue line at the top, has been consistently the kindest to Trump, but currently even they have his net favorable numbers at -10 percentage points, worse than even Gallup had at the end of January. The steepest fall is the yellow line, representing Rasmussen, a poll well known throughout this century as being very kind to conservatives. In January, only Politico and Rasmussen gave Trump a net favorable score. Now, Rasmussen is tied with Gallup giving Trump a -22 point rating, only surpassed in the negative direction by Ipsos/Reuters at -24.

Let me repeat that no poll is perfect and even a collection of polls won't always give us an accurate read. For example, in last year's polls of Pennsylvania, not even one company gave Trump the lead, which made his win there all the more shocking. But having written that, I present this data as an antidote to anecdotes. For all the reporters who can find Trump voters still happy with their choice, the polling companies can find large masses of voters who realize they made a horrible mistake in November.

No comments: