Here is a graph showing polling for Bernie Sanders and Joe Biden. See below for some important details.
The numbers used for this graph come from 38 national polls asking for voter preference about a varying number of candidates. There is a large variation across the polls in how many answered something other than a particular candidate (like “none”). These two factors cause useless and distracting variation in the actual percentage value given to a candidate for a given poll. You can imagine that if a certain candidate gets 23% of the “votes” in a given poll, that number could change a lot if non-answers were excluded, or the total number of candidates was different. An imperfect but still improved way to calculate the percent value for a given candidate is, then, to only look at a subset of the candidates across all the polls, and recalculate the percentage of polling for each candidate using only those numbers. That is what this graph shows, for these candidates only:
Why that particular list? Well, I noticed that if you look across all the polls, one minor candidate (minor in terms of percent in the collection of polls) seemed to vary from the middle of the middle tier to the bottom of the middle tier, but was never in the lowest lowest tier, and also, was polled from early on: Klobuchar. So, I took the RCP average at about the time of the debates, and applied the Klobuchar Factor. If you were below Klobuchar, you were out of consideration. Since then, the candidates have moved around a bit, and a present day Klobuchar Factor would produce a different list. But I don’t really care, because I just needed to have a cutoff somewhere.
The regression analysis suggests that about 56% of the variance seen in each canidates’ polls is explained by time (i.e., there is a pretty robust trend where time matters). I’ve extended the regression line out 20 days into the future, which would be the end of July.
So, getting back to the story of these two candidates. I want to consider each candidate separately. The reason they are both in the same graph, and blog post, is because they are the two candidates with the highest number across the entire data set, so the graph makes sense for their scale, and the process is cleaner of we separate out candidates by scale.
The story of Joe Biden is this: He started off high, around 50%, and ended up much weaker, closer to 30% with some of the most recent polls showing 25%. He halved, almost. Or at least, looking at the extended projection, he is in the process of measuring out his polling half-life, as it were. He was probably artificially high partly due to name recognition, and lost ground as other candidates gained. He also started out in a different sort of artificial high, as a well known and widely loved guy where policy had not been vetted, and has lost among Democrats in that way as well. But this is Biden, and this is how he has performed in his earlier presidential campaigns. Biden watchers are not surprised. Biden watchers will not be surprised if he isn’t really a factor in this campaign by the end of the year.
The story of Bernie Sanders is interesting. His numbers show the second lowest amount of variance, scaled by magnitude, of all the candidates. He started of around 20%. He is still around 20%. Bernie is not moving up, Bernie is not moving down. Well, maybe a tiny bit down. What he seems to be doing, really, is slowing down just a bit as Elizabeth Warren is passing him, much like a car going 45mph slows down a bit when a faster car is passing them on the highway. Though that is of course a bad analogy because the intentionality of events is very different.
In short, Biden is gliding to a campaign ending landing, while Sanders is flat-lining. The latter observation is, I think, the most significant. It tells us something, maybe, about Sanders campaign. His base is unmoving. This is expected, I think. I just hope that should Sanders not get the nomination nod, that base sees fit to support the nominee in 2020, all of them, different than what happened in 2016.