Thread by @chrishanretty

A number of people (n > 2) have now asked me about the effect of rain on turnout, given that it is forecast to be wet on Thursday (1/n).

This is not a stupid question. There is academic research on the precipitation/turnout link, either for its own sake or as an instrumental variable (2/n).

The most cited paper on this topic finds that *an inch* of rain over the course of the day reduces turnout by almost one percent: myweb.fsu.edu/bgomez/GomezHa… (3/n)

But an inch of rain is an awful lot. At the time of #EURef , a day which featured some localised storms, the *wettest* local authority received less than an inch of rain (23mm) (4/n)

The *average* local authority that day received around 3 mm of rain. So if we make the "average local authority" bone-dry, then turnout in that area might have increased by (3mm/1 inch) * 1 = 0.12% (5/n)

When I investigated the link between rainfall and turnout in the referendum, I found no statistically significant link: medium.com/@chrishanretty… (6/n)

Pat Leslie and Baris Ari disagree, and (after including a lot more covariates) find that an extra inch of rain would have decreased turnout by 2.45 percentage points doi.org/10.1016/j.polg… (7/n)

(This would mean that our average-to-bone-dry local authority would have seen turnout fall by 0.3 percentage points). (8/n)

Turnout effects of less than a percentage point are also unlikely to lead to very different vote shares. (9/n)

Suppose 66% of the population always votes, 33% never votes, and 1% votes only if it's clear. (10/n)

Suppose the always votes population votes

42 Con/33 Lab/25 Other

and the sometimes-votes population votes
33 Con/42 Lab/25 Other

which a big difference (11/n)

If it rains, the vote is

Con 42/Lab 33/Oth 25.

If it's dry, the vote is
Con 41.86 [(66*42 + 1*33)/67]
Lab 33.13 [(66*33 + 1/42)/67]

12/n)

So: the effects of weather on turnout are small, if they exist, and the effects on vote share are tiny (less than one fifth of one percent in an extreme case). (13/n)

If you want to fuss about *something*, fuss about whether Wednesday's news coverage will portray the result as a foregone conclusion. Expected closeness is well known to promote turnout (14/14)

p.s. If anyone can hit me up w/ a detailed raster map of election day rainfall, that would be great...

You just *knew* it was going to happen: I wrote on the link between constituency level rainfall and turnout in this election medium.com/@chrishanretty…