Trump simultaneously most and least popular he's ever been

Politics & Elections
1

As of Friday, the Rasmussen daily presidential tracking poll claims that Trump’s approval rating is over 50 percent, the highest it’s ever been.

Today, CNN’s presidential approval poll says that his approval rating is 35%, matching the lowest it’s ever been.

As far as I know, both of these organization do, in fact, do what they claim that do; namely, contact large numbers of Americans (chosen in some way intended to be random/representational), ask them if they approve of Trump, and do their best to accurately record the responses.

I’m not shocked if their numbers are somewhat different. In particular, Rasmussen seem to be tracking Trump’s approval among "likely voters’, while CNN is just contacting random Americans. But normally I’d expect the two to be somewhat in sync with each other. That is, one might be on average 5 points higher or lower than the other, but when one was going up, the other was too, and vice versa.

But at present, one is the highest it’s ever been, and one is the lowest. Which is odd.

(My general approach to presidential approval is to look at 538.com, which, not surprisingly, has Trump somewhere in the middle, although on a distinct downward slide over the last week or so, after a steady climb for the previous couple of months.)
Thoughts?

2

Using the standard “as popular as Trump” sounds like something that can never have a positive connotation. Trump can bask in his polling ups and downs and always know that he will be as popular as Trump.

3

Mmm.

Rasmussen has a significant pro-Trump adjustment in their polling. That means, for whatever reason, they tend to skew about 5% higher on the ‘Approve’ than other polls. So there’s some wiggle room there.

Contrariwise, CNN skews a bit low. 538 has them adjusting UP by 1% on ‘Approve’.

In the 538 aggregate, we see Trump now at 39.4% approve and 54.9% disapprove. He’s underwater by 15.5% after gaining some ground and being underwater by only 11.7% on February 15th. So what pushed those numbers almost 4 full points in 10 days? 10 days ago his popularly was as high as it had been since May 9 of last year.

4

The CNN poll is, in fact, the lowest it’s ever been.

The Rasmussen poll, however, is not the highest it’s ever been. According to Rasmussen Trump’s net approval is currently 13 points lower than its high mark of +14 right after Trump’s inauguration.

Both polls agree that Trump’s net favorability is currently much lower than its highest point.

ETA: Also, Rasmussen has had Trump’s approval higher than 50 many times early in his presidency so it’s not like it’s the second highest it’s ever been right now either.

5

People who consistently vote and people who consistently don’t vote generally tend to have different attitudes and priorities. Pollsters have been tripped up in the past by failing to distinguish between the two groups.

6

I made this same point about Rasmussen ten months ago in a strikingly similar thread.

Thread: Trump Approval Hits 50%

7

Hmmm,

The Parkland shooting was on the 14th. So, how he has handled that has probably not been a feather in his cap.

8

It should also be mentioned that it’s much more difficult to poll likely voters than registered voters or all adults. When polling likely voters you have to make a judgment call as to who will be voting, and there’s likely a feedback between how popular Trump is overall and the likelihood that registered voters will get out to vote for him. When polls are off it’s generally because the set of likely voters was misdetermined. Rasmussen has a poor rating, most likely because they’re not great at determining the set of likely voters.

9

Here’s another weird thing about Rasmussen. While they are pretty much the only ones to use a likely voter screen for presidential job approval, they were also pretty much the only ones doing that during the Obama era as well. The weird part is that they were not nearly as large an outlier during the Obama years.

Here’s how one can see it without putting in all that much effort…

Go to Huffpost Pollster’s Trump job approval and click on ‘[Customize this chart](http://elections.huffingtonpost.com/pollster/trump-job-approval/edit#!minpct=0&maxpct=100&mindate=2000-01-01&maxdate=2018-02-26&smoothing=moderate&showpoints=yes&showsplines=yes&hiddenpollsters=&hiddensubpops=A - D,A - R,A - i,RV - D,RV - R,RV - i,LV - D,LV - R,LV - i&partisanship=S,P,N&parties=D,R,I,N&selected=approve,disapprove&fudge=1)’.

Right now when I do that I see, Disapprove 52.5% Approve 43.0% for a net of -9.5%.

Now the reason we’re using Huffpost Pollster Custom Chart is that I can click on the Pollsters button and unselect Rasmussen. Let’s do that and see how the numbers move.

Disapprove 53.3% Approve 41.4% for a net of -11.9%!

Including Rasmussen moves the average of net approval by more than two points. Part of that is that they report their number more often than most other outfits, but there’s no denying that they are much more favorable to Trump than other pollsters.

Is this the result of the likely voter screen or are they doing things differently for Trump. Now I can’t say for sure, but since they also used a likely voter screen for Obama let’s take a look.

[Obama Approval Custom Chart](http://elections.huffingtonpost.com/pollster/obama-job-approval/edit#!minpct=0&maxpct=100&mindate=2008-01-01&maxdate=2018-02-26&smoothing=moderate&showpoints=no&showsplines=yes&hiddenpollsters=&hiddensubpops=A - D,A - R,A - i,RV - D,RV - R,RV - i,LV - D,LV - R,LV - i&partisanship=S,P,N&parties=D,R,I,N&selected=approve,disapprove&fudge=0.6)

With Rasmussen: Approve 56.1%, Disapprove 39.7%, Net +16.4%
Without Rasmussen: Approve 55.3%, Disapprove 39.7%, Net +15.6%

So Rasmussen was slightly more favorable to Obama than average at the end of his term.

This is not completely rigorous but it demonstrates the underlying point pretty well. Rasmussen was not really separate from the herd in the Obama era, and then became instantly one of the largest outliers after Trump’s inauguration. Something changed and it wasn’t the likely voter screen.

Or more specifically, it wasn’t the use of a likely voter screen that changed. I suspect the details of how they screened likely voters did, in fact, change.

10

Your links are broken – they somehow got a bunch of spaces put in.

11

They work for me, but here’s some cleaner links that will get you most of the way there.

http://elections.huffingtonpost.com/pollster/trump-job-approval

http://elections.huffingtonpost.com/pollster/obama-job-approval

When you click on ‘Customize this chart’ you’ll be able to pick and choose which pollsters are included among other things.

I’ve also noticed that near the bottom of each of these pages we see, “RSS | Spreadsheet | JSON | API Help”. The word “Spreadsheet” links to a downloadable .tsv file that contains all the polls. Once you have that you can filter however you like.

Right now I’m experimenting with Rasmussen v non-Rasmussen weekly moving average. At a glance it looks like they changed their likely voter screen at least twice since Trump took office.

Overall they’re about 6 point more Trump favorable than the average of everyone else. For the first month of the Trump presidency (~Jan 20 - Feb 20, 2017) they average +13 versus others and were never below +10.

Currently they are about +10 more favorable than all others as well, but in between now and then there were a few months where they were chilling in the +3-5 range.

12

All the links work just fine for me.

13

Consistent with the idea that Trump support is falling, the generic congressional ballot is back to a double-digit lead for Democrats.

14

Something interesting I noticed from the list of recent polls: the Q poll was the outlier at the high end of presidential approval, but the same poll was completely the opposite regarding the generic ballot.

I’m also curious why the CNN/SSRS poll doesn’t have a reliability rating, although it has a very small adjustment value (-1).

15

“I’m at the center of First and First. Jerry, I’m at the nexus of the universe.”

16

Most likely his response to Stoneman Douglas.

17

Could it be unreliable, but in an unknown direction? Like the SurveyMonkey polls, which have very low reliability, but no adjustment.

18

This is as good a place as any to put this…

Here’s something for the budding data scientist. I’m going to include a scrap of Python code to grab all the polls collected on Huffpost Pollster’s Trump Job Approval page. This requires Python 3* with the Pandas library. I’m going to put this in ‘PHP’ tags since I believe the ‘Code’ tag is still busted in the default skin. I am assuming that this is OK Huffpost since there is a direct link on the page to the .tsv I’m accessing and their ‘API help’ they say ‘We’re sharing this data under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.’

import pandas as pd

url = 'http://elections.huffingtonpost.com/pollster/' \
      'api/v2/questions/00c%20-Pres-45-Trump%20-%20' \
      'Job%20Approval%20-%20National/poll-responses-' \
      'clean.tsv'
df = pd.read_csv(filepath_or_buffer = url, 
                 sep = '	', 
                 parse_dates = ['start_date', 
                                'end_date'])

# Print first five rows
print(df.head())
# Print column names and data types
print(df.dtypes)

Note: there are some weird line breaks in the above to ensure that things are compatible with both Python’s whitespace rules and the board’s PHP tag line length rules.

Let me know if you have any questions.

ETA: and it still messed up the line breaks even after I previewed. If anyone asks I’ll post a cleaner version. The problem being that the URL is quite long. Admin please fix the code tags bug.

19

It’s not just the “likely voters versus adults” issue that leads to consistently higher Trump favorability figures at Rasmussen, as compared with other pollsters.

It’s that Rasmussen calls only landlines:

http://www.rasmussenreports.com/public_content/about_us/methodology

^ That refers to calls to landlines. They “adjust” for cell-phone users:

[ibid]

Over 50% of US households have NO land line. Majority Of U.S. Households Are Now Wireless-Only. Why Do Health Experts Track Them? : All Tech Considered : NPR

As households with landlines are overwhelmingly likely to be older and more rural than households without landlines, it’s not surprising that Rasmussen consistently over-reports favorable ratings for Trump.

20

This is not an unreasonable hypothesis, but I don’t believe it’s correct. Or at the very least it doesn’t tell the whole story.

Using the techniques I’ve described above I’ve gathered all the ‘Obama Job Approval’ polls aggregated at HuffPost for his entire eight years in office. I’ve looked at the time series of moving averages for Rasmussen polls versus non-Rasmussen polls. There’s some weird stuff in there that strongly suggest they switch up how their likely voter model works from time to time to adjust where they are relative to the field.

For example, while on any random day after Nov 2010 they had 2-6 point anti-Obama lean compared to all other pollsters, however from Sep 2013 to Mar 2015 they had a pro-Obama lean the entire time. That is to say that their two week moving average was more favorable to Obama than the two week moving average of all other pollsters every single day for 18 months. That’s not random walk noise. That’s something changing that I can’t directly observe. Furthermore, calling only landlines does not explain why this 18 month period stands out.