A month before E day

 Hi,

It has been 25 days since the Harris-Trump debate. There was a Walz-Vance debate last Tuesday but, according to observers, it will not likely move voting intentions. Therefore, it is a good time to assess the current situation.

We know that Biden's replacement by Kamala Harris drastically changed the race. Donald Trump was leading before the replacement. Tamala Harris was rapidly shown in the lead by most surveys afterwards. But what happened since then?

The analysis uses 156 polls conducted since July 21 by 47 pollsters and organizations. Since the Trump-Harris, debate, 43 polls were conducted by 27 different pollsters or organizations. A comment here that U.S. seems to be in a unique situation among electoral democracies. Usually, there are not more than a few pollsters, most of them specialized in electoral polls, who conduct and publish polls in any given electoral campaign. The large number of pollsters in the US may be partly due however to the presence of academic pollsters, a situation rarely if ever seen elsewhere.

Voting intention, since Biden's withdrawal and since the debate

The first figure shows that the trends in voting intention do not appear to have changed much since my last post two weeks ago. The trends are flat, that is, there is no trend in one direction or another since the end of the Democratic party's Convention on August 22. However, contrary to what we have seen just after the debate, there are polls that estimate that Trump is ahead of Harris or right on the 50% line (see the red dots on the graph). There are seven of these polls on 43 polls. We now have Harris forecasted -- if the election was held tomorrow -- at 52%, four points ahead of Trump in the two-party share.

 

 

If we use only the polls conducted since the debate, there appears to be a rapprochement between the two candidates. We have to be very careful because the analyses are based on relatively few polls. The figure now shows Harris forecasted at 51.5%. Only the next polls will allow to validate this trend, or not. Right now, the few last points appear influential in the analysis.

 

 

What about modes?

I analyze the differences between modes using all the polls conducted after July 21. I decided not to perform the analysis on the polls conducted after the debate only because there are really not enough polls to perform a reliable analysis by mode. Since July 21, there are 108 polls (69% ) that use Web Opt-in only -- or at least this is what we conclude from the information provided. There are 32 polls (20%) using multiple modes and 16 (10%) using a quasi random single mode methodology. Given these data, I grouped together the polls using multiple modes and those using a single mode with a quasi-random sampling source in order to have enough data points to perform a reliable analysis.

Following a discussion with the director of Activote, Victor Allis, I decided to include Activote in the Web opt-in category, though it does not use, as such, that type of methodology. I checked that, whether I include or exclude Activote, the trends are the same.

The figure shows that the web opt-in only polls tend to estimate support for Harris higher than the other polls. However, the trends portrayed by the two mode categories are the same, i.e., they are flat. The forecast from the web opt-in polls gives about 52% to Harris while the forecast from the other types of polls is very slightly over 50%. Since the debate, only nine web opt-in polls (on 108) estimated Harris's share lower or equal to Trump's share. On the opposite, five of the 16 single mode quasi random polls and ten of the 32 multiple mode polls do the same. 

If we focus on the 43 polls conducted since the debate, there are two web opt-in polls on 31, two multiple mode polls on seven and three single-mode quasi-random polls on five that put Trump ahead or at par with Harris. Therefore, we need to conclude that there is a systematic difference between modes. This difference can hardly be explained by other factors since there are multiple pollsters, each conducting rarely more than one poll during the period. Sixteen pollsters using the Web Opt-in only methodology conducted at least one poll since the Harris-Trump debate, five pollsters using single-mode quasi-random polls and six conducting multiple mode polls. It is highly unlikely that, besides using the same type of mode combination, they all use another similar methodological feature. This being said, it does not mean that, within each category, all pollsters are the same. I am interested in the impact of modes of administration and sampling source, not in individual pollsters and I do not participate in the "Who was best" contest.

 

 

Conclusion: A swallow does not make a summer...

I use to change this expression to "A poll does not make a summer". Funnily, in French the same expression is "A swallow does not make Spring" and now, perhaps we should change it again to "A poll does not make an Autumn"... The point is that, although in the US presidential elections, we have much more polls that we usually have in any other electoral democracy, there is uncertainty left, because of the different estimates that are proposed according to modes of administration and sampling sources AND because the two preceding US presidential elections have shown that some methodologies -- namely those that use multiple modes or single mode with quasi random samples -- seem to have lead to a better forecast of the results. This would mean that Harris and Trump are most likely at par right now. Of course, this does not give the portrait of what is happening in the swing states or of what will happen in the next weeks.