1/26/2025

A few days ago, I published the first version of the 2025 Virginia Gubernatorial Election Poll Tracker. If you've checked the site within the past 24 hours, you would've noticed that it wasn't being updated and had a maintenance message. Today, I've finished the second version of the tracker and heavily reworked its methodology. The new model also no longer tracks the difference in two-party vote share, but instead each candidate's current projected percentage. Spanberger is ahead by 4.8%. This is a half-point increase from the previous model. More importantly, this model would've projected the winners of the 2017 and 2021 gubernatorial elections.

Methodology

Polls are weighted based on their median surveying date, sample size, and 538 pollster ratings of the firm that published the poll. All 3 of these metrics are adjusted to be on a scale of 0-1 and have specific parameters for how each value is weighted.

The first metric is the Date Influence Score (DIS), which is assigned based on a cubic equation that correlates how far back a poll's median surveying date was, relative to the most recently published poll. If a survey has a Date Influence Score ≤0, the poll automatically loses its influence over the tracker. The DIS is the heaviest factor in weighting each poll, as the model is a daily tracker (essentially a nowcast if it weren't for the fundamentals aspect of the tracker). Out of the 3 poll weighting metrics, the Date Influence Score is generally weighted 3:1:2 (with 3 being the Sample Score and 4 being the 538 Pollster Rating).

Poll Date	Days prior to most recent poll	Date Influence Score =1-[0.001*(x)^3]
1/10/2025	0	1-[0.001*(1)^3] = 1
1/5/2025	5	1-[0.001*(5)^3] = 0.875
12/17/2024	24	1-[0.001*(24)^3] ≤0 Poll is nullified

After a Date Influence Score is assigned, the pollster's value is weighted according to 538's Pollster Ratings. 538 uses a 0-3 scale in their ratings, which results in the model dividing the score by 3 in order to adjust the metric to be on a 0-1 scale. Currently, the last poll weighting factor in this model is the sample size, which should ideally be 1,067 for a poll with a 3% MOE and 95% confidence interval of any state's population.

Poll Date	Date Influence Score	538 Pollster Score	Sample Size	Sample Score
1/10/2025	1.00	0.87	806	0.76

After calculating its 538 Score, Sample Score, and DIS, the three biggest pieces of the poll's weight have been decided. The next step is to calculate the poll's overall score, which is done by using the weighted average described earlier (3:1:2).

Date Influence Score	538 Pollster Score	Sample Score	Data Times Poll Score
1	0.87	0.76	0.89

The maximum possible individual poll score is 1, which will make the poll listed above hold significant weight in the daily average until more polls are released. Once the poll score is added to the data table, it becomes part of the overall sum of all of the scores for each poll used in the model. This sum is used as a divisor for all polls to calculate their final "Influence" score that weights them in the adjusted polling average.

Polling Firm	Poll Score	Influence
Mason-Dixon	0 (DIS is ≤0)	0%
Emerson College	0.961	36.39%
CNU	0.900	33.84%
VCU	0.804	29.77%

The adjusted polling average is exactly 85% of the puzzle. In this initial version of the poll tracker, fundamentals make up the remaining 15% of the final average. These fundamentals consist of currently only two factors, though this will likely be expanded upon in the future.

Cook PVI	2016-2017 Shift
D+ 3.0	D+ 3.6

Although the Cook PVI includes the current governorship and state delegation, it doesn't account for Virginia's trend away from Donald Trump in 2017. It's currently unclear how significant this 3.6-point shift will be in the 2025 election, but it's currently being weighted as half of the overall fundamental forecast (7.5% of the total average). The data table below represents the current snapshot of the figures used to make the full poll tracker as of 1/26/2025. Although the tracker will be updated daily, the example tables will remain the same unless the model is further altered.

Weighted Poll Average	Fundamentals Only	Full Tracker
D+ 4.9	D+ 3.3	D+ 4.7