The Twitter Archive

A lot of my analysis is posted on Twitter but never becomes a fully fledged blog post. This post will act as an archive of my earlier work which will hopefully be more accessible and readable than scrolling a year or more back on my Twitter profile.

This will read in chronological order from oldest to newest. I'll include a short description along with graphs here, and links to the original tweets are in the headings.

The Impact of Pitch Speed on Called Strike Rates

Fastballs which are thrown faster are less likely to get called strikes on the edge of the zone. I first noticed this as a feature of my pitch level prediction models but it shows up in the real data as well.

There could be several reasons for this, slower pitches may be easier to frame, or MLB pitchers with slow fastballs have better command and hit their spots which makes framing easier.

Pitch Tunnelling

I looked at whether there was any effect of tunnelled vs un-tunnelled pitches on my model's predictions. There were plenty of (bad) assumptions which went into this and the only conclusion that I could make was that tunnelled pitches performed better than untunnelled pitches, but that is because they had better locations anyway.

HBP+

Tim Locastro was traded to the Yankees so I created the HBP+ statistic to celebrate the occasion. This uses pitch location to estimate the expected rate of HBPs and then HBP+ measures how good a player is at getting hit more often than the model expects.

Locastro's HBP+ is over 1000!

HBPs can be split into "clear" HBPs (xHBP > 0.5) and "marginal" HBPs (xHBP < 0.5). The former shows more variation among pitchers and the latter among hitters. The aggregated results by team are shown in this thread

Effects of Losing Spin Rate on Fastball Location

In the wake of the sticky stuff ban, many pitchers lost spin on their fastballs. This not only affects pitch stuff, but can also mean that pitches sit lower in the zone making them more hittable.

Using My Pitching Predictions on Hitters

Hitter tools can be assessed by using my models. Here I look at batter whiff rates, which players have better bat-to-ball ability and which make better swing decisions.

Stuff & Command Aging Curves

How do my pitching grades change as a player ages?

Minor League Pitching Measurements

The Florida State League has Statcast in its parks which allows me to measure minor league stuff and command grades.

Machine Learning FanGraphs Articles

I trained a neural network on an archive of FanGraphs articles and found the output pretty funny, here are some highlights.

fWAR Leaders - Unknown Pleasures Style

Measuring the Difficulty of an Umpire's Game

UmpScorecards can sometimes be misleading and games with many close, high leverage calls will probably result in a low accuracy and high bias to one team.

An umpires score is more determined by the conditions of the game than their ball/strike calling quality. Although over multiple games this skill becomes apparent.

I wrote an article inspired by this tweet at Baseball Prospectus which was nominated for a SABR award.