New lows, but are they sticky?

Today produced two new lows for Trump approval, one in the normally Trump-friendly Rasmussen poll (above) and one in the normally less friendly Quinnipiac poll (below.)

I wrote on Twitter today about the Rasmussen drop before seeing the Quinnipiac result, but the central point is the same for both. “New lows” are often not sticky. A poll may hit the new low only to rebound a bit from that. If polls are subject to random noise (they are, due to random sampling and other reasons) then a “new low” may well just be a unusual bit of noise that will go away next time.  For this reason, before getting too excited (or upset) about “new lows” we should be skeptical until confirming evidence arrives, in the form of repeated lows by the same poll and by multiple polls showing the same trend. (This applies equally well when we see “new highs”.)

Here are my (lightly edited) comments from Twitter early this afternoon:

I wrote yesterday that polls often bounce back up a little after hitting “new lows”. Monday Rasmussen hit 39% Trump approval for 1st time. Then on Tuesday Rasmussen was unchanged at 39% Trump approval, not bouncing back up.

Now Wednesday, Rasmussen is at 38% Trump appoval, a second new low in 3 days, and the opposite of a bounce back up.

Meanwhile Gallup has read 37, 37 & today 36. Gallup has been in the 36-40 range since early June so these numbers are not “new” for them.

But Rasmussen has until now been consistently among the highest polls for Trump approval. (See first chart at the top of this post.) The decline is therefore striking.

I still expect Rasmussen to hit a low and bounce back up, but if Gallup should also break through 36 to 35 (registered once on 3/28) or below, we would have evidence that there is a genuine new low, not just a short-term bounce down.

Now that we have the new Quinnipiac poll in hand there is another piece of evidence. Quinnipiac’s new approval reading is at 33%, a new low for their poll.  But is this indicative of a drop in approval? It is certainly down from Quinnipiac’s previous poll which was at 40, but that one was actually a little high. Q-poll had previously been in the mid-30s. So the current 33, is a new low, but not one far from most of their previous polls.

As you can see in the chart below (repeated from above) the Quinnipiac polls averages 5.1 percentage points below my trend estimate as of the days the Q-polls have been completed. With a current trend estimate of 37.4, we’d expect the Q-poll to be at about 32.3, quite close to the 33 of the actual poll.

In contrast, Rasmussen averages +4.8 points ABOVE the trend. So their current 38% is very low for them. Based on past deviations from trend, we’d expect the Rasmussen poll to be closer to 42.2 than to 38. For this reason, I think the new low for Rasmussen is unlikely to persist. Why they have fallen so much more than would be expected given the overall trend is a bit of a mystery. It could be substantive. Rasmussen samples likely voters while most other pollsters are using adult voter or registered voter samples right now. Likely voters are more attentive to politics, so it is possible that greater attention has brought some decline among the Rasmussen respondents. But if so, why this week rather than last week, or the week before? The more circumspect evaluation is that for some reason the samples of the last three days have had an unusually strong negative view of Trump, and that as Rasmussen’s sample moves to new respondents, the overall approval will likely return to the usual +4.8 over trend, or thereabouts.

All that said, the approval trend accounting for all polls has declined a bit in recent days, to the current 37.4 that I calculate.  Real Clear Politics puts it at 38.4 approve and 56.9 disapprove while HuffPost/Pollster has it 37.5-58.4 and 538 is 37.0-57.5. There is a lot of agreement across our different methodologies, and all are showing a slight decline in the trend estimate in recent days.

It is always good to compare trends across pollsters, rather than only looking at one poll or only looking at the overall trend estimate. Here are Trump approval & disapproval trend comparisons across all pollsters with 2 or more polls since January.