A few weeks ago I had written about the VA Governor’s race before going on vacation - in that time it seems as though Terry McAuliffe’s campaign had lost a lot of steam and Youngkin made up a lot of ground in the final weeks of the campaign. At the time of this writing, it seems overwhelmingly likely that Glenn Youngkin will become the next governor of Virginia. To avoid some of the galaxy-brain takes that will inevitabely wind up twitter, I thought I’d distract myself by following up on my previous post.
Firstly, I should share the updated polling average. A few weeks ago, McAullife appeared to have a sizeable lead in the polls:
FiveThirtyEight's poll tracker for the VA governor race (https://t.co/HKVw7RcsJN) shows the moving averages - as a weekend coding project, I revamped with a weighted avg., 95% CI, and win probability based on an election day distribution📈 pic.twitter.com/hM6vzLUZnM— Mark Rieke (@markjrieke) October 3, 2021
However, as of election day, the race had significantly tightened to effectively a coin-toss:
Final update - p_win ~ 55% McAuliffe; again, the win probability is just the portion of the polling avg. distribution above 50% for each candidate. For *actual* models, go check out the work by @lxeagle17/@Thorongil16 or @jhkersting, among others pic.twitter.com/W9bDf1OumZ— Mark Rieke (@markjrieke) November 1, 2021
As I mention in the above tweet, the win probability isn’t a true forecast, just the portion of each candidate’s election day distribution above 50%. That being said, actual forecasts similarly had the race down to a near 50-50 split as of this morning:
🚨FINAL #VAGov FORECAST🚨— Lakshya Jain (@lxeagle17) November 2, 2021
At the end of a long campaign, here's where the model made by me and @Thorongil16 stands. The race is rated as Lean Democratic, with a forecasted margin of D+3.6 and a win probability of 67%. An interactive map is over at https://t.co/Kmvj6sRElC. pic.twitter.com/kuaJc82HML
While I definitely plan on utilizing a more scientific poll-weighting methodology in the future, I do find it interesting that even a simple averaging method can produce relatively accurate results in line with the majority of other forecasters.
Regarding post-hoc analysis of why McAuliffe lost, I won’t dredge up any of my own (partially because it’d be irresponsible & pundit-y to do so without referencing any data and partially because it’s getting late & I’m a bit tired), but I’ll point out a few tweets from Nate Cohn that show that the results appear to show a near uniform shift across precincts and different voting groups. This would suggest that McAuliffe’s loss is tied more closely to the national environment, rather than shifts amongst specific groups/counties.
Update at nearly 1500 precincts: Youngkin is still on track for victory, running a ahead of what he needs across basically every dimension worth looking at.— Nate Cohn (@Nate_Cohn) November 3, 2021
But it's worth noting just how close things are to 'expectations,' which is simply shifting the 2020 result to the right pic.twitter.com/Mbe3pg7NFB
And btw, if you're doing pundit things I'd look at that red column. McAuliffe fell short of expectations--a more-or-less uniform shift--by basically the same amount, just about everywhere— Nate Cohn (@Nate_Cohn) November 3, 2021
This won’t stop the networks from ascribing the win/loss to very specific campaign issues (I’ve already seen quite a few folks ascribe Youngkin’s win to education, race, suburban-reversion, etc., without any evidence to back up such claims). Until there are deep dives into data regarding the election, I’d treat any comments rom pundits with a hefty grain of salt.
Some closing thoughts
That’s all for me today! I’ll be back in a few weeks with some non-political content, looking at a machine learning model predicting the price of a diamond in the diamonds dataset.