Please explain analytics to me

[TJseven7]

Sooo... this discussion has not gone well up to this point. I'll leave it that at that.

 

I realize math is concrete. I believe statistics often times are interpretive. The further you venture down the road of variables the more convoluted results become. I don't argue math... I do argue the merit of used and discarded variables.

 

Yet, I value analytics. I think there will come a day that they rule league wide. I also believe they aren't iron clad enough in their current state. I try to be an educated fan and I'm well aware I do not know baseball. I've put my time in other sports. Some of you, especially the analytics crowd have put your time in baseball. You know more here. I'll admit it. That doesn't stop me from seeing issues with a way of thought... or stop me from questioning why.

 

The only part of analytics I want to discuss is run generation. I belong to neither school of thought.

 

I'm under the impression that the numbers used by analytics folks do not take into account the situation as it applies to the current game. IE: low scoring/high scoring tied, down 1 or up 10... all occurances lumped together. Statistics in a vacuum. Season(s) seen as 1 event where numbers hold true.

 

1...Is that true? If so, why do analytics people assume there are no quantifiable variations in regard to game situation?

 

I've heard often that the bunt success rate is terrible. Heard many conflicting numbers. I hear constantly... well hes a bad bunter so bunting him is bad.

 

2... Are there any stats (preferably without pitchers included) that show bunt success rate diminishing over the years? We have to assume bad and good bunters (position players) were used to bunt in old school baseball. Is there any proof stating bunters are league wide getting worse?

 

I also hear that the pitcher was nasty so getting a bunt down would be hard to do. While I understand and acknowledge that I wonder why no one brings up the fact that the nasty pitcher also makes it harder to hit.

 

3... Is there any proof that a power arm or nasty pitcher makes it harder to bunt beyond the point of equal correlation to the degree said pitcher makes it harder to hit?

 

I'm asking you to make sense of why these perceived holes in logic aren't a reality.

Thank you!

[MadScientist]

There was a real stats guy around here a few years ago who posted some analysis that looked at players in clutch vs non clutch situations and found no significant difference.

 

I'm sure somewhere there are stats about bunt success rates, but perhaps the more critical thing is that in most cases where people used to regularly bunt, teams would be better off not bunting. To put it another way, the expected number of runs scored in a given situation is lower when a team bunts.

[logan82]

I've heard often that the bunt success rate is terrible. Heard many conflicting numbers. I hear constantly... well hes a bad bunter so bunting him is bad.

Bunting in general is bad. Sorry, giving up outs in general is bad. Being predictable about when you are trying to bunt is bad. It cuts down on making it safely to a base. Here is a link to expected runs depending on base state and how often at least one run will score based on the base state. This is obviously based on average but you are basically better off pinch hitting than bunting.

 

http://www.tangotiger.net/re24.html

[Ennder]

1. Yes analytics look at game situation but not every stat does. I'm assuming this is in regards to the bunt attempt discussion and in that specific case I was showing you numbers based on just scoring 1 run. There are also numbers that look at scoring multiple runs which would be more important earlier in the game. Most stats are context neutral though so you have to add the context.

 

2. I couldn't find this info with a quick search. I know the number of bunts are way down, don't know about the success rates.

 

3. I would assume it is but I haven't seen any actual proof ever.

 

Using context neutral stats a bunt is never good, ever. The context has to push the decision into being good. Whether it is a good bunter, bad hitter, great hitter up next, bad defense etc. If we had say Dee Gordon up and Christian Yelich was up next the bunt is almost certainly the right move in that game because Gordon is an elite bunter and Yelich is a really good hitter. I just can't imagine Pina is a good bunter and Broxton isn't a great hitter.

 

In the specific case of Broxton such a large percentage of his hits are extra bases that he doesn't need the runners moved in order to drive them in with most of his hits. Such a large percentage of his outs are via the popout or strike out that moving the runner to 3rd doesn't increase his chances to score a run on an out nearly as much as others. The context of having Broxton up next is a negative, not a positive.

 

I'm assuming Pina isn't a great bunter, it is really just a guess. Not very athletic catchers usually aren't. So his context is a negative. Tampa Bay is a good defensive team, Longoria being one of the better defensive 3B in the game. Sucre being a good defensive C. I'm assuming this context hurts the bunt.

 

In this specific case all of the context seemed to point away from a bunt.

 

I will also say this is only a major league baseball thing. In high school where the fielding isn't as good and the players don't hit as many extra bases bunting is right in a lot more cases. Even in college it is right more often than in the majors. 50 years ago it was right more often as well because there just wasn't as much power in the game and there were way fewer strike outs.

[bjkrautk]

Using context neutral stats a bunt is never good, ever. The context has to push the decision into being good. Whether it is a good bunter, bad hitter, great hitter up next, bad defense etc. If we had say Dee Gordon up and Christian Yelich was up next the bunt is almost certainly the right move in that game because Gordon is an elite bunter and Yelich is a really good hitter.

 

The other aspect of the sacrifice bunt which you have to take into account (assuming the hitter doesn't bunt for a base hit) is that it opens up first base....meaning your really good hitter may just be given a free pass to set up the double-play if the hitter behind him is prone to hitting ground balls, or if there are already two outs.

[LouisEly]

I think what TJseven7 is getting at is the chances of scoring one run, if that is what is needed late in a game.

 

Expected runs is a weighted sum of: 0 runs x (probability of scoring zero) + 1 run x (probability of scoring 1 run) + 2 runs x (probability of scoring 2 runs)... etc.

 

The bunt reduces the probability of scoring 2+ runs, which decreases expected runs. But what is the probability of scoring 1 run when bunting? If by bunting it reduces the probability of scoring 0 runs or increases the probability of scoring one run, and one run is what you need late in a game, then the analytics would justify that.

 

I do argue the merit of used and discarded variables.

Agree with this whole-heartedly. There was quite a debate here many years ago about the risk/return/value of stolen bases, and/or the value of a "fast" player.

 

What was never included in the discussion is the OPS of the players batting after said fast player when said fast player is on base. To evaluate the value of a player such as Juan Pierre, yes his career OPS was only .704, but I want to know what the delta in OPS is for players batting after him when he is on base versus not on base. Does speed cause a pitcher to be more predictable or make more mistakes? That is the variable that I want to know.

[torts]

is bunting even tracked, publicly? I know internally, teams track that, but do fangraph or BR keep a log? So unless you or I had access to that data, I think its extremely difficult for us to really "analyze" its success outside of what we remember or track ourselves.

[BuckyBrewer61]

But what is the probability of scoring 1 run when bunting? If by bunting it reduces the probability of scoring 0 runs or increases the probability of scoring one run, and one run is what you need late in a game, then the analytics would justify that

 

It's right there in the link at the start of the thread.

 

The odds of scoring at least one run with a man on 2nd with no out, or a man on 1st and 2nd with no outs is about 61%.

 

The odds of scoring at least one run with a man on 3rd with one out, or a man on 2nd and 3rd with one out, is about 66-68%.

 

So a successful sacrifice bunt increases your odds of scoring at least one run by about 6-8%...not nearly as much as some would like to claim, but yeah in the bottom of the 9th in a tie game you rather have the man on 3rd with one out than on 2nd with no outs.

 

BUT, we all know quite well that sacrifice bunting is not a 100% sure thing, and in this case you'd have to successfully sacrifice at least 90% of the time to increase your odds of scoring at least one run. I don't know the stats, but I'd imagine today's major leaguers aren't nearly that successful.

 

So you can make all the subjective arguments you want about more ways to score, scoring without a hit, etc etc. But the objective evidence is pretty straightforward, bunting decreases your expected number of runs while not increasing your chance of scoring at least one run unless you have a hall of fame sacrifice bunting specialist up there at the plate.

[FVBrewerFan]

The other issue is how do you count/track an AB where a bunt attempt is made on strike 1 and 2, then the hitter swings away and Ks are makes an out. Trying to bunt puts hitter in the hole 0-2, greatly decreasing odds of getting a hit or even advancing the runner for the rest of that AB. Bottom line, basically an unsuccessful bunt attempt but stats wouldn't count it that way.

 

So, if you include that in any analysis, I would bet a sacrifice is successful at 70% at the very most and maybe close to 60%. But that's all we have to go on, opinion.

 

But no matter how you slice it, bunting (generally speaking) is not the best choice outside of pitchers, of course. There are times where it's closer to a 50/50 call, but as others have said it comes down to game situation, who's up (and their ability to bunt) who's behind them in the order, etc. For example, the Pina non-bunt the other day was pretty close to 50/50 in my opinion. Close enough that I don't think either choice would have been "stupid" or "crazy."

[BuckyBrewer61]

A cursory search shows the rate of successful MLB sac bunt attempts is 70-80%, honestly seems high to me. But let's give the benefit of the doubt on the high range of success given our data:

 

From the link in this thread, we know the odds of scoring at least one run with a man on 2nd (+/- 1st as well) with no one out = 61%

 

Next, the odds of scoring at least one run when bunting with a man on 2nd (+\- 1st as well) and no one out = Odds of a successful sac bunt attempt x Odds of scoring at least one run with a man on 3rd (+/- 2nd as well) and 1 out = 80% x 68% = 54%

 

This obviously does not take into account game situation, matchups, batter at the plate and those due up, etc. But I know which odds I'd rather have.

[Bombers]

Great Post TJSeven7

 

I work in the analytics space and have a few thoughts I can offer up - I apologize for the length

 

Data Science is growing into a top degree and career path based on these studies. Analytics / Big Data is not just a baseball thing anymore as businesses everywhere are interested in data.

https://www.indeed.com/jobtrends?q=Big-Data%2C+data-science&l=&relative=1

http://www.teachthought.com/everything-else/a-new-social-science-statistics-outgrowing-other-stem-fields/

 

I had received a powerpoint from a Cubs Season Ticket Holder that was given by Jed Hoyer and had a graph showing that Analysts in MLB have grown from 33 in 2009 to 142 in 2016 for a 330% increase. I would imagine that number will continue to skyrocket as teams are building their secret sauce through data. So what does all of this have to do with the Pina vs Broxton to bunt vs not bunt? It points to teams are building these big data systems and these systems will pull in many factors to consider. But in the end, these systems have variability as there is more than 1 way to build the house. The op has a link to the run scoring as a hole. 68 % when runners at 2nd/3rd with 1 out and 61% when 1st/2nd with 0 out. One can ask is that enough or what else can we consider?

 

Splits of Pina vs Broxton vs players on the bench. Splits for 2017 vs their career.

https://www.baseball-reference.com/players/split.fcgi?id=broxtke01&year=2017&t=b

https://www.baseball-reference.com/players/split.fcgi?id=pinama01&year=2017&t=b

 

How about the current pitcher Tommy Hunter is he a fly ball pitcher or a ground ball pitcher

https://www.baseball-reference.com/players/split.fcgi?id=hunteto02&year=2017&t=p

 

As the OP stated "statistics often times are interpretive" and I agree with that. Teams are building their framework around what they believe to be important factors and leaving others out. Some teams may look at position player bunting or maybe bunting with a person on 1 / 2. How fast is the person at 2nd base? Can he get thrown out at 3rd? What is the chance for a double play with 1st/2nd with current hitter? Is there a better player on the bench to deliver a fly ball? Will they walk that person to load the bases up and then what is the double play rate for the next person due up? So many variables

 

As many have stated - statistics don't give you absolutes - they paint pictures. However different artists paint different pictures. One analyst may rely more on one set of data and another on a different set. In the end, teams are building trust into their analytics machine. While fans don't sit there with an ipad connected to terabytes of data, teams do and their scientists are in place to make sure it is readily available during the game to help them make in game decisions. So I would lean towards answering the OP's question / comment around assumption that analytics people assume there are no quantifiable variations to be I bet you they do it's just not readily available on baseball-reference.com and sometimes not statistics based. (Pitcher looks tired, maybe Pina or Broxton has hit the ball well all day, etc, etc)

 

My assumption is these teams are taking gobs of data and presenting multiple scenarios for the manager to consider and then Counsel ultimately makes the final call. I would love to get a peak into these systems to see the data being given to the coaching staff these days.

[torts]

you looking for work, Bombers? LOL, were struggling to fill an analytics position on my team at work

[Baldkin]

Some quick links:

 

Win Expectancy List: http://www.tangotiger.net/welist.html

 

This article contains my single best argument against bunting ever: Half the time you're giving up a strike.

http://www.fangraphs.com/blogs/the-truth-about-bunting/

I collected data from between 2008-2013. I classified things simply: a good bunt is a fair ball in play, and a bad bunt is a foul bunt or a missed bunt. Of course, not all fair bunts are good bunts. Of course, this misses out on bunts that were pulled back at the last second. This tracks only bunts that were committed to. Over the six years, there’s a sample of more than 36,000.

 

The breakdown:

 

Overall: 49.7% fair bunts

Pitchers: 49.9%

Non-Pitchers: 49.6%

 

The sample for pitchers is about 10,000. The sample for non-pitchers is about 26,000. There’s basically no difference. About half the time they’ve committed to a bunt, they’ve bunted the ball in play. That means that, half the time, they’ve messed up.

[BuckyBrewer61]

I work in the analytics space and have a few thoughts I can offer up - I apologize for the length

 

Data Science is growing into a top degree and career path based on these studies. Analytics / Big Data is not just a baseball thing anymore as businesses everywhere are interested in data.

https://www.indeed.com/jobtrends?q=Big-Data%2C+data-science&l=&relative=1

http://www.teachthought.com/everything-else/a-new-social-science-statistics-outgrowing-other-stem-fields/

 

I had received a powerpoint from a Cubs Season Ticket Holder that was given by Jed Hoyer and had a graph showing that Analysts in MLB have grown from 33 in 2009 to 142 in 2016 for a 330% increase. I would imagine that number will continue to skyrocket as teams are building their secret sauce through data. So what does all of this have to do with the Pina vs Broxton to bunt vs not bunt? It points to teams are building these big data systems and these systems will pull in many factors to consider. But in the end, these systems have variability as there is more than 1 way to build the house. The op has a link to the run scoring as a hole. 68 % when runners at 2nd/3rd with 1 out and 61% when 1st/2nd with 0 out. One can ask is that enough or what else can we consider?

 

Splits of Pina vs Broxton vs players on the bench. Splits for 2017 vs their career.

https://www.baseball-reference.com/players/split.fcgi?id=broxtke01&year=2017&t=b

https://www.baseball-reference.com/players/split.fcgi?id=pinama01&year=2017&t=b

 

How about the current pitcher Tommy Hunter is he a fly ball pitcher or a ground ball pitcher

https://www.baseball-reference.com/players/split.fcgi?id=hunteto02&year=2017&t=p

 

As the OP stated "statistics often times are interpretive" and I agree with that. Teams are building their framework around what they believe to be important factors and leaving others out. Some teams may look at position player bunting or maybe bunting with a person on 1 / 2. How fast is the person at 2nd base? Can he get thrown out at 3rd? What is the chance for a double play with 1st/2nd with current hitter? Is there a better player on the bench to deliver a fly ball? Will they walk that person to load the bases up and then what is the double play rate for the next person due up? So many variables

 

As many have stated - statistics don't give you absolutes - they paint pictures. However different artists paint different pictures. One analyst may rely more on one set of data and another on a different set. In the end, teams are building trust into their analytics machine. While fans don't sit there with an ipad connected to terabytes of data, teams do and their scientists are in place to make sure it is readily available during the game to help them make in game decisions. So I would lean towards answering the OP's question / comment around assumption that analytics people assume there are no quantifiable variations to be I bet you they do it's just not readily available on baseball-reference.com and sometimes not statistics based. (Pitcher looks tired, maybe Pina or Broxton has hit the ball well all day, etc, etc)

 

My assumption is these teams are taking gobs of data and presenting multiple scenarios for the manager to consider and then Counsel ultimately makes the final call. I would love to get a peak into these systems to see the data being given to the coaching staff these days.

 

Excellent post.

 

Gets to the likely true answer of to bunt or not to bunt, which is...it depends.

 

Compared to what MLB managers have at their disposal, we have very limited information. Are there scenarios where bunting is statistically favorable? I'm sure there are, but I'd guess based on limited research of what is readily available (and MLB managerial trends) that those scenarios are statistically few and far between.

 

In the end, in this day of endless data, the manager has to decide whether to go with the odds or go with his gut based on the situation. We often say it's the manager's job to put his team in the best position to win. Generally speaking, I'd argue to go against objective data would be to not do so.

[TJseven7]

Ok so trying to piece this all together:

 

1: The stats we are looking at in regards to run expectancy and the likelihood of scoring does not factor in game score and inning. No variation assumed. I don't see how that can be assumed. We discuss leverage situations but don't think it effects scoring. Seems strange.

 

2: Baldkin thank you! Please understand the math I'm about to do will be crude. 49.6 fair bunts. As you stated that gives away a strike. Agreed and understood. But you can say by those numbers you'd expect 2 strikes to net you 74% chance of a fair bunt. I'll zero the potential of strike 3 for ease. Claims of 80% success sound asinine. Claims of 70% success sounds very optimistic. Pop bunts play in... bunts that get the lead runner mowed down exist. So i would suspect we are looking at numbers closer to 60% chance of moving the runner to 3rd... slightly lower chance of moving both runners to 2nd and 3rd. We don't know but that seems realistic to me. Fair? 60% seems awful... SEEMS awful.

 

3. Inconclusive at our level of knowledge. We have no idea if the pitcher has a greater effect on bunting success than the pitcher has on hitting success. We can assume... or not.. some might assume the elite pitcher has a greater downward effect on hitting than bunting. We simply dont know.

 

I can safely say that giving up an out lowers run expectancy. On the chart supplied by logan that holds true in ever single field across all recorded time. That isn't debatable. The only alteration to that is in the example BJ posted of the successful bunt and subsequent walk to set up the force. That gives back expectancy (and then some .104 runs increase over 1st 2nd no outs) but cuts the successful bunt gain from a 6.6% to a 4.7% increase in odds to score.

 

But in this case I don't care about run expectancy. 1.6 era... assuming all scores are solo (its not so numbers are better than this) he allows TB to tie the game 5 of 27 times. Win expectancy jumps from (roughly) 45% to above 81% (possibly as high as 85%) moving from zero runs to 1 run. That's significant enough that I dont care about runs beyond 1.

 

Bunting or giving up an out is always bad... for run expectancy. Fact... proven.

Bunting or giving up an out is always bad... in regards to likeihood to score... problem.

 

I've seen it multiple times. 60% x 67.6% = 40.5% (notice I'm using my lower successful bunt number)

 

That's easy to do. I also think moving one option a move down the road while leaving another option alone is disingenuous, not comparable and an oversight. Aren't we dealing with diminished returns? This is baseball afterall. 60% chance of a successful bunt. 60% of successful anything in baseball in regards to hitting is a gigantic number. With pina hitting you certainly have a chance he plates a run. You also leave a larger chance for a gidp on the table vs the bunt. (I dont know the % offset) That chance turns 61% chance to score into 25.7% in a flash. Is it likely... no(I assume, but an elite GO pitcher could certainly inflence it)... but its a viable percentage that has to be calculated in. Ks, shallow pop outs, force at 3rd are the same result as a failed bunt. Force at 1 same as successful bunt. Force at 2nd same as some bunt outcomes.

 

A 6.6% increase in the odds of scoring is making it 11% more likely that you score. A 4.7% increase (post ibb) makes it 7.7% more likely that you score. While swinging away gives you a chance to plate a run in that ab.. it also increases the likelihood that you will depreciate your chances of scoring in that inning. You simply can't adjust for the black and white outcome of the bunt without adjusting the outcome of the ab.

 

I believe the success rate of the AB will fail to match the success rate of the bunt. One has more upside and downside... one has more security. Once you move on to the broxton ab (or arcia ab post ibb) there are new and more plentiful options that plate a run which is reflected in the increase of likelihood to score despite the out with runners at 2nd and 3rd.

 

So I'm questioning the breakdown of the pina broxton arcia swing away abs... vs the breakdown of the pina bunt broxton and/or arcia (and batter in the hole) swing away ab.

 

Pina bunt is around 55% at 67.6%, 5%(?) at 63.4% (force at 2nd) 40% at 40.6%

Pina swing has some 100% some 67.6%... 63.4%...40.6% and 25.7%

 

Question 2... Does the swing away ab truly put the team in better scoring position or is it a push or near push with higher upside and drawbacks.

 

Until you can properly calculate that you can't say things like giving up an out is always wrong.

 

I do believe it is usually a bad idea to bunt. Run expectancy matters too much to give up outs in most cases. (Especially early) But I also believe this definitive talk of never is wrong. I also believe talk of only elite bunters followed by great hitters can push it into being a good choice is wrong. I believe remaining outs and the arms you can turn to can also flip that scenerio due to killing (kind of) the need for multiple runs. When mn bunted in the 7th with two bad bp arms in the holster a play for a 1 run lead didnt seem worth giving up the out. With knebel in the holster in a series where 7 runs were scored in 26 innings and 1 run in 26 innings by TB... 1 run looked far more imposing.

 

I'm certain our remedial look cant come close to it being a certainty... despite swimming in mountains of data i doubt that even the pro analytic groups can prove a statistical advantage. Too many factors variables filters to specify. Too many ways to paint the painting. Would love to see how deep down the rabbit hole they go though.

 

So the question is... bat by bat... where do the numbers land. 1st 2nd no out... pina bunt vs pina swing. Compare. You cant make an honest comparison without advancing both outcomes that ab.

[Oxy]

This is a great thread that I somehow missed years ago.

 

I've asked multiple times if those Tango run expectancy (and win expectancy) charts have data from years that were not 1999-2002.

 

Obviously the more runs on average that will be scored decreases the importance of each individual run, and all "small ball" strategies are focused on scoring single runs.

 

1999-2002 was probably the greatest run producing times in history, so maybe those charts are...wrong?!?!? Maybe they are RIGHT, I don't know but using them as gospel truth doesn't seem quite right to me. Nobody has provided me or this site in recent years with better numbers though.