Rainfall and the 2019 election
tl:dr no substantively significant association, but there may be a weak negative link
Yesterday’s general election was the first December general election since 1923. Although there was no significant snowfall, the weather was pretty dreich across much of the country. This wet weather led to concerns about turnout being lower than it might otherwise have been, had the election been held in the spring.
Turnout in the election was down, despite changes to the electoral register which ought in theory to have boosted turnout by removing duplicate entries and deceased voters. Although I cannot assess whether turnout would have been greater had the election been held in another month, I can assess whether the weather was a factor, by comparing areas which received higher or lower rainfall and checking whether turnout in those areas was higher or lower.
Weatherquest, a spin-off of my former employer, generously provided me with a map of rainfall over the period 7am to 10pm yesterday. Using this, I was able to extract the average levels of rainfall per constituency.
The average level of rainfall was just over 8 millimetres; the standard deviation, 3.6 mm. Looking across the different regions, rainfall was highest in the South West (avg 10.5mm) and Wales (avg 9.8 mm), and lowest in Scotland (3.2mm) and the North East (6.9mm).
As you can see from the graph, there is no substantively important
relationship between rainfall and turnout (left panel), or between
rainfall and the change in turnout (right panel) — that is, it fails the intra-ocular impact test.
If we want to go beyond scatterplots, we can estimate a regression model of turnout including rainfall as a predictor. Our conclusions depend on the details of the regression model (see table below).
If we include just past turnout, then there is a (statistically) significant negative effect of rainfall. An extra millimetre means a fall in turnout of 0.05 percentage points. Moving from no rain to average rain (8 mm) would be associated with a fall of 0.4 percentage points.
If, alternately, we include 2017 majority and region effects, we find, if anything, a small positive effect. An extra millimetre of rain is associated with 0.03 percentage points in turnout. However, the effect is not statistically
significantly different from zero, and we cannot rule out a negative
effect as “large” as -0.018 points per millimetre, or half a point
down per inch of rainfall.
Thus, if you are a rain booster, you may want to say that turnout was 0.4 percentage points lower than it could have been, had everything been bone dry. You'd have to say why you believe a parsimonious model which ignores differences between regions and closeness of contest (a known predictor of turnout).
If, conversely, you are skeptical about rain effeccs, you will want to focus on how including regional effects turns a statistically significant negative association into a positive one much closer to zero. When I previously looked at this in the context of the 2016 referendum, the conclusions didn't change when I included controls for region and expected closeness: this time they do.
This is obviously a very minor point in the context of a very significant election. There are far broader questions about the composition of the Conservative and Labour electorate which will take weeks and months to unravel.