Ten days after the debate

 Hi,

In this post, I present an analysis of the polls of the US 2024 presidential election, first starting June 14 and second, after Harris became the likely candidate, globally and by mode combination. 

How had support for the two main candidates changed?

The debate between Kamala Harris and Donald Trump took place on September10.   Most media have concluded that Harris won the debate and opinion polls show that around two-thirds of the respondents thought the same. However, did it change voting intentions? The short answer seems to be that voting intentions did not move substantially. However, the trend up for Harris is still showing.

The first graph shows voting intentions for the Democratic candidates -- Biden before July 21st, Harris afterward -- starting mid-June. The four vertical lines show the Biden-Trump debate, Biden's withdrawal from the race, the end of the Democratic convention and the Harris-Trump debate. It is obvious that the replacement of Biden by Harris changed voting intentions substantially as I showed in my previous post. Before the replacement Trump was ahead of Biden in most polls. Harris rapidly appeared to be preferred to Trump is most surveys since then.

The graph shows a very small trend in increased support for Harris since around the convention. There is no plateau in support however.

 

If we use only the data gathered after Biden's withdrawal, the small trend up for Harris appears more evident. The second graph allows for a zooming on the post-debate polls. On the 20 polls conducted since the debate -- starting on September 11 or after -- three showed Trump ahead of Harris and one showed both candidates at par. All the other polls have Harris ahead. The graph shows an increase in voting intention for Harris that brings her still around 52.5% however, five points ahead of Trump in the two-party share, identical to what I had in my preceding post.

 

 

Mode effects?

As I have shown for the US 2020 election, mode effects may differ during a campaign. While in 2020, I analysed the data only from September 1st, this time, I gathered all the polls since January first. I was not sure to find the same patterns. Some pollsters have changed mode and/or sampling source since 2020. However, in too many cases, it is difficult to figure out exactly the methodology used by the pollster. I categorized the polls in three combinations of mode and sampling source. a) Web Opt-in only b) single mode probabilistic or quasi-random source -- web probabilistic or live phone, for example -- and c) multiple modes -- various combinations of IVR, Web opt-in, Web probabilistic, Live phone, sms-to-web, etc. 

Since June 14, 22 different pollsters conducted 137 web opt-in only polls, 10 conducted 22 single mode quasi-random polls  and 16 used multiple mode polls (53 polls). One pollster - Activote -- is not categorized because its methodology is "unorthodox" by all criteria. Some "pollsters" are in fact organizations that sub-contract polls but they do not provide sufficient information to know who conduct the polls. Nineteen pollsters conducted only one or two polls since June 14. Many of them are in the single mode quasi random category.

We need to have enough different pollsters in each combination to be able to attribute mode differences to mode-source combinations. If it were not the case, the differences could be attributed to other methodological pratices of each pollster, like specific weigthing procedures, likely voter models, etc. At least for now, there are not enough pollsters and polls using a quasi-random single mode to keep them apart. The trends for these polls would be meaningless. Therefore, I grouped them with the multiple mode category.

This first graph shows the trends by mode combination since June 14. It shows that, as in most of the 2020 campaign, Web Opt-in polls systematically estimate support for the Democratic candidate higher than polls that use multiple modes or a quasi random single mode.

If we zoom on the period since Biden withdraw and all the polls measure support for Harris vs Trump, there are 82 Web Opt-in polls, 29 polls using multiple modes and 14 using quasi-random single mode. Web Opt-in polls are the only ones showing a trend up for Harris. However, since the debate, only three polls on 20 polls conducted show Harris trailing. One is a web-only polls, one is a quasi-random single mode poll and one a multiple mode poll. One quasi-random single mode poll shows the two candidates neck-to-neck. The 16 other polls show Harris ahead of Trump by 2.04 to 6.25 points.

Conclusion

We have to be very careful since, although I use rather robust analytical procedures, changing the period of analysis may have an impact on the trends and the forecast. Each poll is one point in the analysis which means that, for example, the conclusion on the after debate period is based on only 20 data points. A few more data points could change the conclusion somewhat. However, if we look at the whole period, the main conclusions of the analyses remain. Support for Harris is higher than support for Trump. There is a small trend toward an increase in support for Harris, at least in the Web Opt-in polls. There is much variability in the polls, between and within mode combinations. We should not look at polls as it they are all equal. We cannot conclude which ones are "good" and "bad" polls. We will only be able to conclude whether some mode combinations tend to lead to better estimates, and this only after the election is over.