The 0ptimus Election Forecasting Simulator is intended as a fun tool to play around with spending decisions by outside partisan groups in the 2018 midterm elections. You can choose to re-allocate GOP money, Democratic money, or choose "Both Parties" mode and re-allocate both to better simulate the real-time chess match that goes on during election season.
The math behind it is all based off of our 2018 House and Senate models- which performed very well and had about 97% accuracy in predicting individual seats. In changing the independent expenditure numbers, especially by unrealistic amounts, you’ll sometimes get strange results- but for the most part, this tool will reflect realistic probability changes that organizations like the NRCC and DCCC tried to optimize across the country- and now you can try to beat them at their own game.
Our House and Senate models are ensembles of several models we run (for more info on this, check out our model methodology). For this exercise, we use our logistic regression model with 30 variables selected using Select K Best feature selection, and the additional two variables that do the work- Democratic IEs and Republican IEs, which consist of the total spending by outside groups for the general election on each side. In a logistic regression model, a coefficient associated with a variable quantifies the change in the log odds produced for a unit change in that variable while all other variables are held constant. The old probabilities are obtained from our ensemble model. Using these coefficients we calculate the change in log odds produced by the IE changes that the user introduces and then, using this change and the old probability, we estimate the new probability of victory.
Favored Seats - A race is said to be favorable towards the GOP if its GOP win probability is greater than 0.5; otherwise, the Democratic candidate is the favorite.
Expected Seats - Based on the probability of each seat, we calculate the expected number of aggregate seats for each party. For example, if 5 races each have an 80% chance of Democratic victory, we would expect 4 Democratic seats and 1 Republican seat from that group.
The solutions we provide are meant as possible ways of maximizing a certain outcome. Some of these are just one of several potential ways to solve the problem- there are many different ways to “save California,” for instance. These are the specific ways we went about maximizing favored seats and expected seats:
Optimized solutions for expected seats - We first remove all money spent from all races, and then we add money in chunks of $10,000 to a race whose GOP/Dem win probability increases the most by adding that amount. We do this until we exhaust all the available money. In this way, our algorithm always spends money on a race where it would increase expected seats the most.
Optimized solutions for favored seats - First, we remove all money spent from all races, and calculate the probability of GOP/Dem victory. We do not re-add money into any races that are already above 50% probability of victory at this point. In this way, we do not waste resources on races which do not require any external support to already be favored. In the second step, we allocate money in chunks to races which our ensemble model with actual IEs had above a 50% chance of victory, until they are once again above 50%. In the third step, we take the leftover money from the second step and allocate it to races based on their win probability as predicted by the ensemble model (races that were close to 50% already receive money first). Some races never cross 50% even if we pour a millions of dollars into them. After identifying such black-hole like races, we reallocate wasted money from such races to the other races below it. Finally, with any leftover money that would no longer flip any races at all, we add money in chunks of $10,000 to the remaining races below 50% whose win probability increases the most by adding that amount. We do this until we exhaust all the money. In this way, our algorithm always redistributes independent expenditures to get the maximum number of favorable races.
Optimized to Lean R/D - Similar to our solutions for favored seats, these scenarios use the same methodology but instead maximize the number of seats where the party has a 60% chance of winning or more- surpassing the toss-up category and reaching Lean R/D.
Likely R/D Reallocation - What if we were to re-allocate money spent in races that were likely lost and re-allocate it to toss-ups instead? This scenario does that for each party; races where the party had less than a 25% chance of winning had their independent expenditures removed and instead added to races near the middle, where IEs make a bigger impact.
VA10 Reallocation - This scenario envisions where the GOP, Democrats, or both could have re-allocated money spent in VA-10 more efficiently across the country, to maximize the seats where they would be favored.