For some time I’ve been wondering if the right wing propaganda machine could game the polling averages at sites like Real Clear Politics (RCP) and 538. We all know that polls like Rasmussen are deliberately slanted toward Republicans. Others apparently have had the same concern.
Not all polling averages are created equal. RCP, from what I understand, pretty much takes a straight average over several days. As of the 25th, they had Biden with a lead of 6.6%. FiveThirtyEight has a secret sauce formula that adds a weight to each poll based on the quality of the poll and the age of the poll. They have Biden’s lead at 7.1%
We can have an endless argument about whether this weight or that weight is the right weight, or how far back in time one should go, or an endless other number of factors that effect the poll. You simply take your best shot at what are the best criteria and you see how it works out.
Well, having the goal of trying to remove deliberately slanted polls is not easy. You simply can’t say this poll or that poll is slanted and I’m not going to use it. So what to do? I decided that if you just tried to remove outliers, you should be able to catch both the polls that are trying to game the averages, and true outliers that are also influencing the averages.
So here’s the gory details of the approach, which you can skip this paragraph if you don’t want your eyes to glaze over. I used a 20 day moving average of the polls listed at the 538 website. Pollsters that reported more than once in that period were averaged together. Polls were weighted according to their age in a simple linear fashion, a poll 20 days ago got 1/20th the weight of a poll reporting today. First I average the polls together, than I removed the 5 polls that were furthest from that average and averaged again with the remainders.
So a bit of a drumroll please. So what did I find with this approach?
First, let’s take a look at the averages of Biden’s lead starting on June 20.
The blue line is the average before the 5 outliers are removed, and the orange line is the average after the outliers are removed. As of the 25th, Biden had a 7.14% lead with outliers, and a 7.75% lead without outliers.
A word about the outliers. If the outliers were truly simple statistical outliers, I would think there would be a similar number of high outliers as low outliers over time. So who were the outliers?
Outliers
| Pollster |
Biden Lead |
| Emerson College |
3 |
| Climate Nexus |
11 |
| Targoz Market Research |
3 |
| Rasmussen Reports/Pulse Opinion Research |
1 |
| EKOS Research Associates |
1 |
Before the last month or so, the number of high outliers and low outliers were either 3-2 or 2-3 pretty regularly. But over about the last month we’re seeing a lot more low outliers than high outliers.
Just out of curiosity, I wondered what would happen if you used the median (the value where half the values are higher and half are lower) rather than average both before and after removing outliers.
Again, blue is before and orange is after the outliers are removed. There really isn’t much difference between the two, which makes sense since most of the polls would be clustered near the median. Using the median, we see that Biden has a 7.75% lead before removal of the outliers and an 8% lead after.
So what can we conclude from this exercise?
1) The RCP, 538, and this average, all show Biden falling slowly throughout the summer. But not by much and it seems to have leveled out or even reversed in the last week or so. After the unheard of unifying of the Dems after the primary, and the great Dem convention, most people expected that those lofty levels couldn’t be maintained.
2) This particular technique shows there are more outliers on the low side than the high side and it’s dragging down the average by a little less than 0.5%. I’m going to put my tin foil hat on and say I’m a bit suspicious of some of those low end polls.
3) This technique, along with the RPC and 538 averages are all within about 1% which I think is pretty significant for using some form of averaging.
4) Most importantly, even though this technique shows Biden about 1% higher than RCP and 0.5% above 538, it really doesn’t matter if Biden is up 8 or 20, we have to work like he’s down 5. EVERYTHING is on the line in this election and this is the election where I don’t care who I piss off or which friends I lose, if I have a chance of convincing someone to vote Biden and Dems, I’m going to bug the hell out of them.
Vote early in-person if you’re in WI., MI.
