Candidate of the Democratic Party vs Trump: Where are we now?

 Hi,

 

I waited for things to stabilise before publishing new analyses. In this blog message, I analyze the polls conducted since June 14, two weeks before the debate between Joe Biden and Donald Trump, up until the last polls published this morning.

Here are the trends in voting intention for the candidate of the Democratic Party (Biden until July 21st and Harris afterwards) and for Donald Trump, candidate of the Republican Party. The following graph shows that,support for Donald Trump placed him in very good position before and after the debate, However since then, and particularly after Joe Biden's withdrawal and hi eventual replacement by Kamala Harris, the situation changed completely. Most polls that have been published since Harris took the lead as the possible candidate for the Democratic party favor her. According to the polls, she would be at about 52% right now. However, some polls still see Trump as leading. The next graph will show the characteristics of those polls.

 

 

The following graph presents the poll results according to mode combination. Because we need to have enough polls in each combination for our purposes, we categorized the polls in two groups, polls using web opt-in only and polls using multiple modes. There are much more polls using web opt-in conducted recently than polls using a combination of modes. It seems rather clear that, although both types of polls estimate a clear increase in support for Kamala Harris. However, the web opt-in polls, as in the 2016 and 2020 US presidential elections, estimate the support for the democratic candidate higher than the other polls. And in 2020, the "other polls" were right. Of course, meanwhile pollsters have worked on improving their methodology. Nonetheless, the difference is persistent. This difference may have several causes, not necessarily related to the quality of the polls but it is possible that the 2024 election will give us another test of mode effects.

 

In conclusion, everybody should be careful in the conclusions they draw from polls. The campaign is just starting.

 

** Methodology: 

First, I take only one poll by pollster by period. Many pollsters ask two questions, one where they present all the candidates and one where they force respondents to choose between the two main candidates. I chose to always take the poll that mentioned all the candidates because it seemed more comparable to the other polls.  

Second, the trends are calculated using local regression. I always used a parameter of .65 (the percentage of points used for adjustment) for the analysis, because it was the parameter most recommended in the literature. It gives smoother trends. However, it did not seem appropriate in a situation with much movement. I therefore changed it to 0.5 which gives trends that are more sensitive to movement.