There has been a lot of work behind the scenes to our daily projection methodology since my last update in May. The impact has been very positive – e.g., June represented our most accurate month ever for projecting DFS hitter points.
If you haven’t signed up already, I implurge (implore + urge) you to sign up for a month of DFSBot as for $9.99, you will get our DFS projections for DraftKings (and three others providers), the Streamonator + HitterTron for dominating Roto AND free access to our partner LineupLab’s MLB lineup optimizer leveraging our projections.
Here is a summary of the recent projections methodology improvements:
- Hitter Plate Appearances – Plate Appearances impact every hitting counting stat so it is a critical projection to get right. I have made several enhancements in May/June culminating in the latest/greatest methodology that released today. The improvements have worked to date (the correlation for PA vs actual has increased every month this year) but I think this latest version – which incorporates recent lineups, platoon patterns, opposing pitcher, and opposing bullpen – is the best yet.
- Hitter Slugging/Total Bases – I overhauled the underlying projections to estimate singles, doubles, and triples (instead of just hits, HRs, and SLG). This should lead to an improved projection as it: 1) Allows use of 1B/2B/3B park adjustment factors vs a blended ‘Hit’ park factor. The impact of park factors are smaller for singles than extra-base hits so using a blended ‘Hit’ factor leads to some inaccuracy for hitters more prone to singles or XBHs than average and 2) Creates a slightly better adjustment vs pitcher as certain pitchers are more prone to certain types of hits (e.g., ground ball pitcher gives up more singles, less XBHs).
- Hitter Runs / RBIs – I made an enhancement in late May that seems to have worked (best monthly Run and RBI results going back to May 2014) but I took it up another level by leveraging recent lineup data and the 1B/2B/3B/HR events to create projected Runs/RBIs per batting event (which includes BB, HBP, and ‘Outs’ as well) based on projected lineup spots (e.g., a player like Billy Hamilton would receive a blend of leadoff and 9th place run/rbi factors). If July isn’t our best month ever for R/RBI projections, I will be disappointed.
- Pitcher Stats – I have made improvements across the board including Innings Pitched estimates and opponent Hit and BB/HBP factors. I am optimistic these improvement will lead to some tangible accuracy gains.
There were three areas that I researched/tested and chose not to incorporate. I could just as easily add them and make our projections sound more awesome but, as you all know, I prefer to call it as a I see it (i’m looking at you hitter streakiness):
- Pitcher Wins/Losses – Earlier this year, we started projecting bullpen performance which enabled Team ERAs (subscribers can see them in Teamonator) and projected winning percentages per team (using Bill James’ Pythagorean Run Formula). I tested this data as a supplement and a replacement to our current Win/Loss projection formula for Starting Pitchers – surprisingly, it did not provide a lift.
- Weather – Thanks to LineupLab, I now have access to daily weather data. While Park Factors account for regional advantages (e.g., Texas heat, Colorado altitude), it would seem like certain gameday weather conditions would be more conducive to hitting than not. After analyzing a month’s worth of data, I think that a weather-based adjustment could eventually lead to slightly better projections. But it is a tricky one that will require a season-long analysis vs a single month (not surprising given the weather differences per month). I will be reviewing in the off-season.
- Home Plate Umpire – I stumbled upon some data on the interweb showing some significant differences in K’s, BB’s, and even runs based on home plate umpire. I tested these home plate umpire factors against our starting pitcher K/BB projections across a number of weeks of June results to see if home plate umpire could help explain some of the differences. The results were very underwhelming – the home plate umpire factors added nothing. I may revisit in the offseason but I have little faith that this is a meaningful variable.