Three weeks before E Day: Deep fog?

 Hi,

 Polls seemed to be somewhat contradictory lately so I waited to be able to include the last week-end polls in order to see if things get clearer. I present the analyses using local regression for the polls conducted since September 1st, the date usually considered the real start of the campaign. It also allows me to compare with previous campaigns where I always started my analyses with the polls conducted after September 1st. 

Notice also that the polls are positioned in the graphs in the middle of their fieldwork. This is important because some polls are published "the day after" but others are published up to six days after the fieldwork ends. It is somewhat mindboggling. Why? One thing is sure, graphs should use the fieldwork date.

The first graph shows the trends in voting intentions for Trump and Harris, over the two-party share. The two vertical lines are for the presidential and vice-presidential debates. The graph shows that there was likely an increase in Harris's share of voting intentions between the two debates with some polls estimating her share at 54%. However, the graph also shows a substantial variation in polling. On September 22-23 for example, one poll shows Trump slightly over 46% of the two-party share and another poll at close to 52%. 

Since the vice-presidential debate, most of the polls show Harris ahead of Trump. The graphs show that the trend is now flat, with Harris slightly under 52% of the two-party share (and therefore Trump at slightly over 48%).

 
 

If I compare with the 2020 campaign, this campaign appears quite different.  In 2020, the peak in support for Biden was reached around October 10 -- at 55% of the two-party share -- according to the polls that used multiple modes or quasi random single mode methodologies.  Web opt-in polls showed an almost flat trend at 55%. Biden won 52.2% of the two-party share in the end, with only the multiple mode polls giving an accurate estimate, on average.

We cannot be sure however that the two situations atre comparable since Harris became candidate late in the "general campaign" and she was the current vice-president.

What about modes now?

As I noticed before, it is difficult to figure out differences between modes. It is even more difficult because research has shown that the main difference is not necessarily in estimates at a given point or during a given period, but differences in the trends portrayed by each mode (Durand & Johnson, 2021). And to analyse trends, it is preferable to have as many polls as possible. With a reduced number of polls that are not conducted using Web Opt-in (72% of the polls in September), the trends are sometimes "fuzzy". Since there are much more web opt-in polls than polls using another mode, the trend that they portray is smoother. 

The following graph shows support for Harris (on the two-party share) increasing between the two debates and stabilizing since the vice-presidential debate. This is on average because since the vice-presidential debate, for example, three web opt-in polls show Harris at par or one point below Trump. The others estimate her share from 51% to 53%. The only poll estimate from a poll that do not use web opt-in methodology estimates Harris's share at 51.5. 

One thing is sure, there is much variability in poll estimates, whatever the methodology used.

Another way of looking at modes and estimates is the following. The table presents the results of all the polls that started in September.

It shows that 72% of the 75 polls that started in September used a web opt-in methodology. Four poll estimates out of five (81%) favored Harris. The others (14 polls) favored Trump or put the two candidates at par and nine of these polls used either multiple modes or a quasi random single-mode methodology (like live telephone or web probabilistic). It is possible that the "bump" that we see between the two debates is due a compositional effect because of a higher proportion of polls conducted with multiple modes or quasi-random single mode methodologies.

 

 

Conclusion 

For now, we have polls for five complete weeks of the main electoral campaign. While it shows that a large majority of the polls estimate that Harris support is larger than Trump's, the gap between the two is not that substantial. Besides, there is a discrepancy in estimates according to the methodology used. There are still three weeks to go. The only thing to do is monitor the situation 😏. It is certainly impossible to say right now that one or the other candidate will win. And we need to keep in mind that the main unknown is how the estimates at the national level translate into votes in the electoral college.