Last updated August 20, 2017 07:50:21 PM

What does data tell us when we continuously average President Trump’s approval rating polls, “correcting” for house bias and putting more stock in the most recent polls? Here’s what we know.

Wondering how Trump’s numbers compare to those of past presidents? Our historical approval ratings tracker lives here.

How Many Americans (Dis)Approve of President Trump?

Approve: 37.4%

Disapprove: 58%

Net Approve: -20.6%

Long Term Trump Approval

The above numbers come from a “correcting” of polls for their varying biases, which we presume come from the different firms that conduct polls with differing methods. I then average all of those polls over the past two weeks, giving more weight to more recent polls released.

Trump Approval by Pollster

*Note: the polls. shown below are a selection of the highest-quality and most frequently fielded. If you want the full picture, check the table below!

All Polls of Trump Job Approval

Methodology (Short Version)

(Long version here)

The idea that polling firms have differences unique to themselves is nothing new. Political scientists since the 1980’s have been documenting these effects. These so-called “house effects” have come to the foreground of political statistics in recent years, and as such are the major focus on our Trump Tracker.

Yet, the house effects aren’t the only component here. Determining Trump’s popularity is a three stage process, namely:
  1. Collect polls of Trump’s job approval
  2. Adjust polls based on their house effects
  3. Average the polls, with those recent polls counting more than others

The first step is easy, as is the third (we use a very simple exponential weighting formula similar to this) — which leaves the second for evaluation here. Below I briefly discuss how we determine house effects for this tracker, plus a few other discussions of systematic variance in Trump approval polls.

We compute the house adjustments used above by looking at the differences among polling firms with comparable survey methods who release polls at the same point in time. For example, Rasmussen and Public Policy Polling are some of the only pollsters who consistently use IVR/Online samples and apply a likely voter model to their top line. Rasmussen’s house effect is the difference in their polls from the average IVR/Online + likely voter poll.

The graph below plots the survey house effect for any pollster who has conducted two or more polls of Trump’s job approval.

Another way of looking at this is a simple box plot of Trump approval polls over the past two weeks. Here, the differences across survey houses are plainly illustrated.

As I mention in the the original analysis by Alexander Agadjanian and myself, there are also systematic differences among polls taken online, by automatic phone dialing, and by live interview over the phone. Those differences are listed here (note: these differences aren’t accounted for in the adjusted Trump approval rating above — that’s only house effects).

Over the next 4 years, more than 2,000 polls of President Donald Trump’s job approval will be released. Check back here each day for automatic(!) updates to this interactive. How popular will Trump be in 2019? We’ll be the first to know.

The code for the analysis behind this interactive can be found here. Polling data from