Hi,
The question is whether the Democratic convention modified the trends in voting intention that had appeared since Joe Biden withdrew from the race. I present an analysis of the polls conducted since June 14 in order to be able to understand the impact of different events, that is, the Biden-Trump debate on June 28, Biden's withdrawal on July 21 and the Democratic convention that ended on August 22, where Harris was officially nominated as the Democratic candidate.
There are 173 polls conducted since June 14. A word about our data source. We have relied on Wikipedia from the beginning and, in 2020, there were no major difference between Wikipedia's report and fivethirtyeight.com. However, this time, for the same period, we found 35 polls that are present in fivethirtyeight.com but not in Wikipedia. Most of these polls are public, but not publicized, that is they are conducted by political groups or individual pollsters who are not sponsored by any organisation or media. I checked that the results did not differ when we use only the polls referenced in Wikipedia or all the polls (see methodology at the end of this post).
The first figure shows the trends in support for the Democratic candidate -- Biden before July 21 and Harris afterward -- and for Trump. It is computed on the "two-party share", that is on the sum of the support for both candidates. In addition, when pollsters use two questions, one that asks for vote intention for all the candidates -- including Kennedy, Stein, etc. -- and one that asks only about support for Trump or the Democratic candidate, we took only the former question in order to make sure not to give more weight to some polls. So, every poll appears only once.
Figure 1 shows that almost all the polls favoured Donald Trump before and it was even more evident after the debate. The situation changed completely after Biden's withdrawal from the race. Since the beginning of August, almost all the polls favour Kamala Harris over Donald Trump. Voting intention for Harris would therefore be around 52.5% now, according to the polls. There is no clear sign yet that there was an increase in support for Harris following the convention but there is no sign of decrease either.
What about modes?
The second question is whether there is a difference according to mode of administration and source of data. With the information provided, I categorized the polls in three groups: a) Polls that use only web opt-in panels -- around 64% of the polls --, b) polls that use multiple modes (25%) and c) polls that use a random or quasi-random single mode (telephone polls conducted by interviewers, or web polls using a probabilistic frames, for example). Since there are not enough single mode quasi random polls, I grouped them with polls that use multiple modes since these polls normally use at least one random or quasi-random source.
Figure 2 shows the trends in estimates of support for the democratic candidate according to the combination of mode of administration and data source. Last week, there was a difference between web opt-in polls and polls using multiple modes but there were not that many multiple mode polls carried in the last days before my blog post. With a few new polls entered, we see that all types of polls show the same portrait of the current support for Harris, that is around 52.5%. In a way, this is good news for citizens, ... and for the Democrats. When different methods show the same estimates, we become more confident in the results. We still have to stay modest because there are not that many polls that are using multiple modes and random data sources.
Conclusion
For now, we can be quite confident that support for Kamala Harris is higher -- 4-5 points -- than support for Donald Trump. It is even more evident considering that we do not see any poll, whatever the method used that has Trump ahead of Harris for the last 10 days. How does this translate in Electoral votes and in support in specific states, this is not my competence. The real campaing is just starting. I will follow the polls from now on with blog posts planned about every two weeks unless there is something important going on.
*** Methodological note: Robustness check: If I use only the polls that are listed in Wikipedia, the trends are a bit different, with Harris about half a point lower but the same other assessments remain, that is, there is no poll showing Trump ahead of Harris in the data.