Very interesting. Does each data point represent the top 25% fastest pitches grouped by pitch count? And do these pitchers represent the ones who gained/lost the most steam as pitch count rises?

Justin verlander is a beast!

Justin verlander is a beast!

You could say that again, Jeff.

I had a feeling Jonathan Sanchez would be on this list. Yikes, he really turns into a different pitcher by the end of the game.

By the way, I'm glad to see a post about this, because I've been thinking about "gaining and losing" velocity within starts for a while now. Do you know if there would be some accurate way - in terms of a metric - to determine which pitchers gain or lose the most? I was thinking about running some sort of simple linear regression between pitch count and velocity, and then taking the slope of that to determine a general trend. I would love to hear your thoughts.

Albert, I took all pitches 2007-2010, partitioned them by pitcher and pitch count, and then broke them into 4 percentiles. I used only the top 25%.

As for how I chose the pitchers, I liked the how the graphs looked. The players were all interesting and near the very top/very bottom of my list. I don't know the best way to quantify such a measure, but I ran a correlation between pitch count and velocity for pitches 1-100 for everyone. Sanchez was -.97.

This is very interesting work, Jeremy. A few questions:

1) Were the top 25% of pitches still classified as the same pitch type from the beginning to end of the game?

2) Would this explain some pitchers having consistent FIP/xFIP vs. ERA splits? For instance, The top 3 players tend to have FIP/xFIP above their ERA, whereas the bottom 3 players tend to have xFIP and FIP below their ERA. Could this because the players whose velocity declines get most of their strikeouts early and then give up most of their runs late, thus leading to greater ERA's than would be expected from their strikeout and walk totals?

DSMok1, I didn't look at pitch types at all.

4 of the top/bottom 5 follow your suggested ERA-xFIP split (Shields, Duke, Sanchez, Beckett do, Trevor Cahill does not. Jackson, Lilly, CC, uroda do, Livan Hernandez does not.), but overall, there's zero correlation. The original reason I looked at these charts was to see whether size/build related to loss of velo, but I didn't find anything by looking at that, either. Except that I could make nice charts.