##  "Last Updated: Thursday, 24 Mar, 2022 10:57"
Probability of Winning Election
This is the probability of each party winning the election, given their current polls and what we’ve seen historically. A PL win is around 90% likely. This is roughly the probability Biden had of winning the 2020 US Presidential Election on it’s eve. The probability of a PN win is around 8%, or slightly more than the chance one picks up an ace from a full deck of cards at random.
To arrive at these probabilities, public polls were collected from MaltaToday, Esprimi/Lobeslab, Sagalytics and MISCO. A smoothing line based on a GAM spline was fit.
Historical absolute error (the difference from the polls to the actual election results in 2017 and 2013) was calculated.
The election was then simulated 40,000 times, with each iteration having a random injection of both poll absolute error and the variability of the polls (measured through the spline’s error). For most cases, it was what was around what was historically expected, but in some cases, larger errors, such as one would see with polling misses were also simulated.
The vertical lines are each respective party’s polling average on the latest day.
Polls Through Time
These are how the polls have evolved over time, with each point being a specific poll, and the line being the fitted spline. This is the primary input to the model.
Estimated Election Day Result
And here is how the vote share on election day is expected to pan out. The bar value is the median vote observed in the 40,000 observations, and the error bars to each side are the 80% credible intervals: 80% of all simulations run fell within these lines.
How has this probability evolved over time?
We can also run the model back historically to see if it changed significantly. The truth is a rather uncompetitive race.
A full methodology writeup is available here.
Needless to say, I borrow pretty heavily from The Economist’s recent forecasts, with chunks of this copy pasted from the French Election’s GitHub repo here.