What Can Be Done to Fix the Sox?

Like most players Holt is a streak hitter, but he tends to have those streaks when he doesn't play regularly. He is what he is - a really good utility player who should be used that way, not as a regular. At least not with the Sox.

Holt at 3B, De Aza in LF ... what a relief.
The goal is average, but mediocrity will be tolerated.

2 weeks. That's the length he needs to go to swing a hot stick and have everybody love him.

I think Sandoval will be better than Crawford. Maybe SF will take him back. I think Panda will be about what he was on paper. He is not worth his contract.

Hanley will be fine once he gets more healthy.

Have better luck on balls in play. They're at least 3 wins short already, and that's ignoring Fenway Park.

I've got a nice regression model to predict team BABIP, with r = .62. Yes, the Sox have baseball's worst Soft%, but Soft% is meaningless, because there's no difference in terms of contributing to team BABIP between it and Medium%.

The two things that really drive team BABIP are Hard% and Popup% (FB% * IFFB%). Together, they explain 57% of the explainable part of team BABIP. And, yes, even though the Sox are only 22nd in Hard%, they're next to worst in Popup% and next to worst in the interaction between it and Hard% (it's Hard% * (Hard% - k * Popup%)).

But, wait, there's more.

The next most important factor (13%) is going to the opposite field, and the Sox are 14th in MLB at that.

Then comes IFH% (10%). The Sox are second. They have 21 more infield hits than average, which has added 14 points to their BABIP. Without them, they'd have a .256 BABIP. Playing in Fenway Park.

Low GB% contributes 9% in this model. Remember that infield hits have already been accounted for, so it's teams that have a high IFH% relative to GB% that really get a BABIP boost, which makes some sense. The Sox are 22nd in lowest GB%, bad (for this metric) but not terrible.

Low K% contributes 7%. The Sox are second best in MLB (who knew?).

High BB% contributes 4%. The Sox rank 8th. I may check later for an interaction between K% and BB%, which would make some sense, and if there is one, we probably look even better.

The model predicts a .295 BABIP in a neutral park, which is a point below this year's average and 2 points below the 2002-14 average. In other words, the relatively low amount of hard contact and very high popup rate should have been nearly offset by all the infield hits, with a little bit of boost from terrific plate discipline.

A .295 BABIP versus the actual .270 is 37 more hits than we've had, and each hit is about .8 runs. I think that when you factor in Fenway, it's probably a 4 win deficit.

Oh, and the model says Mookie is 10 hits unlucky, which would bring his BA up to .289. He's basically hit his Steamer projection (.286 / .350 / .428) on the nose, he's just done it right at fielders.

Thanks for the analysis Eric, very interesting. 
However, I do have a couple questions. 
How did you determine that Hard% and IFFB% are the driving factors of team BABIP?
I remember an article on Fangraphs about trying to make sense of the Hard%, and in the comments I came across this graph:

cdn.fangraphs.com/blogs/wp-content/uploads/2015/06/Non_Normalized_BABIP_HH-.png

It seems here that there's no correlation between Hard% and BABIP, so does the IFFB% really make that big of a difference?
Sorry if I'm missing something here, I don't know that much about statistics but I am curious.

Eric can answer for himself, but even in that graph there's a decent positive slope. R^2 doesn't really tell you anything about the strength of the relationship, only about how important that factor is, relative to other factors (and random noise). It's kind of a signal-to-noise ratio. Also, once you control for other factors, it's very possible that the effect of Hard% could go up (let's say teams that have a high Hard% also had a high Popup%, when you control for that the effect of Hard% on BABIP would go up).

Quick question for Eric - is your regression predicting current period BABIP or next period BABIP?

That graph actually shows a very statistically significant correlation, but one that only explains a tiny part of BABIP. Too much noise, which is to say, individual player luck, but thousands of data points.

Using team data is just a clever trick. All the players on a team in a year are sharing the same park factor, so it's legit to lump them together, and when you do that, you remove a lot of the noise. The luck of a group of players will tend to even out to the point where you can measure it. Each sample in my analysis averages 6216 PA, or about 10 years of data for an average player, only it's composed of many players with different styles. It's highly unlikely that such a group if players has a real factor (other than their park, which adds noise) that the regression can't find. And a bunch of other things that I looked at it were not statistically significant: LD%, Soft%, wGDP, Pull%, Med%, Cent%, OF-FB%, BUH%, wSB, UBR, HR/FB. All of these are already accounted for by the other variables.

I used multiple linear regression to do this. I may try ANOVA later, which will get me park factors, and then redo the regression. This was a quick study, just a couple of hours of work.

They need to draft better and get a new GM. Ben has a bad trade record. I am happy the Sox did not give the Phillies what they wanted to get Hamels.

And they just wasted 3 years on WMB.  What should they have done?  I would have preferred Headley, but that didn't look like it was an option.

Headley didn't appear interested in leaving NY.  That's why he wasn't an option.  The cute nickname crap is a bunch of garbage.  Are you channeling godot?