Picks for the 2018 Preakness Stakes or: How I Learned to Stop Worrying and Love the Predictive Model

Picks for the 2018 Preakness Stakes or: How I Learned to Stop Worrying and Love the Predictive Model

What on Earth is happening? It’s been two weeks since the Kentucky Derby where Justify emerged through record amounts of rain and slop to break the Curse of Apollo. Following behind him by 2½ lengths was Good Magic. This Saturday, the two horses are meeting again for the 2018 Preakness Stakes. Justify is the favorite followed once more by Good Magic as the second favorite valued at 1-2 and 3-1, respectively. This race is truly the tale of two horses and I, for one, am very excited. However, only Good Magic has broken the top three for the predictive model next to two longshots, on which betting is typically unwise.

Predictions for the race based on our predictive model

Two weeks ago, I made picks on the Kentucky Derby using a statistical model that predicted Good Magic to win and Justify to place. Building off of this previous model, a newer version has been created for the Preakness that now has Bravazo winning, Sporting Chance placing and Good Magic showing.

Meanwhile, Justify is nowhere to be seen. He could place lower than expected having suffered a hoof bruise during the Derby. However, Bob Baffert — Justify’s trainer who previously trained American Pharoah — claims all is well. Recent track sightings have reassured many and I might be imagining it, but I can’t stop seeing a limp. While it is rare for Kentucky Derby winners to also win the Preakness, if his Derby win is disguised for the model it does not change the predicted outcome.

This whole thing feels weird. Though, the beauty of analytics is that it removes our own personal biases and utilizes objective data from the past to make inferences for the future. Therefore, if Justify wins this race as many experts are predicting, he is a special horse.

This updated model accounts for post number, morning odds, weather, past performances, and expert picks to predict the place in which each horse will finish. The model has an overall accuracy of 94.74% with a confidence interval of 82.25% to 99.36%. While these values are relatively high, they are not 100%. This means that anything is still possible.

When modeling, it is wise to remember that no prediction is definitive. This is especially true in horse racing where your model might tell you that Mendelssohn will place 3rd and then he decides that running in the mud is no fun and comes in dead last. Oh, horses…