Base Stealing

[rpriske]

I asked this once when I first starting playing and was told that I "shouldn't know" this.

That didn't make any sense to me then, but I was new so I let it slide.


My question is, what is the calculation to decide the odds of successfully stealing a base? This is something that we SHOULD know because that tells us what we are looking at when we look at the numbers.

Does a 15 STL rating mean a baseline of a 75% safe percentage? How is this affected by a catcher's arm? How is this affected by a pitcher's hold rating? How is this affected by a runner's speed?

I have been trying to get a good idea of this on my own, but the results I am getting don't make any sense.

For example, in a game last night I had two different base runners each try and steal twice each and get thrown out (collectively) all four times. Bother runner had stealing ratings and speed ratings of 15/15 and 15/14. The catcher's arm was a 9. The pitcher's hold rating was a 1. To me, these numbers look like an IDEAL time to run... but they were 0 for 4.

I know what percentage is worth it for me, but I have no idea what ratings turn into what percentage (and IRL, I certainly would).

Also, are those odds era-dependant? I find that in my 1902 league, despite the stats being heavily weighted in the runners favour, they are much more likely to get thrown out than in my modern league, where the ratings are not so lopsided.

[Charlie Hough]

I can't answer all of your questions, but you've got to be careful not to fall into bad reasoning, which we're all prone to do when looking at numbers like the ratings you cited.

Historically, I don't think we have any caught stealing numbers until 1913. But, I believe that OOTP uses real life percentages for successful steals to calculate the baseline probabilities. It also has year-by-year SB and CS numbers in its era stats file for historical play, so I would assume that it adjusts the baseline based on the real life totals for a given year.

So, like anything, I believe OOTP starts with league totals. So the 15 rating does not equate to a certain baseline percentage. It starts with the leaguewide percentage and then works from there to determine ratings and success rates.

Also, statistically, for the early years of baseball, steal attempts were far less successful than they are in the modern era. The difference is as much as 20%. But this is a point that few of us think about because the frequency of steals and steal attempts was so much greater, especially in the deadball era.

For example, here are some numbers for a few years in history:

1920: 1745 SB, 1560 CS, 53% success rate

1950: 683 SB, 568 CS, 55% success rate

1980: 3287 SB, 1602 CS, 67% success rate

2012: 3229 SB, 1136 CS, 74% success rate


Now, I don't know how OOTP would adjust these rates based on the player ratings. But, when you consider that base stealing has become quite rare in MLB, with teams stealing well under one base per game, and when you consider that managers rarely attempt to steal unless they're using a very good runner, the success rate would tell you that it can still be tough to steal.

If you're playing in a league based in the early decades of baseball, then I'd expect to see my players thrown out quite often based on the league totals. If you're playing in more recent years, then I'd expect the success rate to be higher, adjusted for player ratings, of course.

The MLB-wide success rate has gone up to around 73% to 74% in recent years, and a lot of that has to do with managers being very careful about stealing. In the preceding two or three decades, it was hovering around the mid-60% to upper 60% range. And, even then, teams were not trying to steal very often with weaker base stealers, and the best guys in the league were not stealing at rates that were vastly higher.

At the height of their base stealing, Rickey Henderson and Vince Coleman had success rates of around 82%. At other times, they were lower. And in Henderson's record-breaking 1982 season, he succeeded on 76% of his steal attempts. So even the greatest of all-time was thrown out once every four attempts that year. In fact, he was caught stealing 3 times in a game against the Angels and was caught stealing twice in each of three other games.

Now, in that 3 CS performance, he was gunned down on 3 out of 4 attempts by Mike Heath, who had a pretty good arm as a catcher at the time. He threw out 45% of would-be base stealers, but that means they still succeeded 55% of the time. The other three guys who nailed him a couple of times were Joe Nolan, Carlton Fisk, and Bob Boone. Fisk only gunned down 35% of runners that year, which was about average. Boone, who was one of the greatest defensive catchers to ever play the game, gunned down an MLB-leading 58% of base stealers that year. But Nolan only gunned down 32%, which again is average.

So it's a mixed bag, but it tells you that even an average catcher can nail the best in the game. And your guy with a 15 steal rating is no Rickey Henderson. And the catcher's arm rating is a 9, but that's right around average on a 20-point scale. So you've got runners with solidly above-average steal ratings going against an average catcher with a pitcher who is pretty poor at holding.

In the end, that means there's still a sizable percentage chance that they will get gunned down on any given steal. My guess is that the chance of success considering the league totals and the ratings would be somewhere around 60 to 75%. So, is it really all that far-fetched that they would have a bad day and get caught twice each? All it would take is falling into the 25 to 40% chance of getting caught on those four occasions, which can easily happen statistically.

[rpriske]

Thanks for this. This is good stuff.

What I am trying to determine is when to run. It has been established pretty strongly that attempting to steal with a success percentage of under about 75% is a negative strategy. You are giving up more than you gain.

I am just trying to figure out what comes out to 75%. Basically, I am running to liberally in 1902 and being a little too stingy in 2015.

The first mistake was treating the ratings like absolutes. I was making decisions based on ratings the same in both eras and based on your comments, that will always give me sub-par results.

[RchW]

What was each runners SB/CS numbers before the game in question? What was the catchers CS% before that game. How many games played so far? Sorry for all the questions but we have to look at the numbers in context. No single game (or 5 or 10) can be used to judge a fundamental stat output.

The other question is; were all 4 SBA in good order strategy wise? Tough one to answer but even the best base stealers have to pick their spots. I have a sneaky feeling that the OOTP game engine has an algorithm for that scenario.

Last but not least, were they straight steals or R&H or H&R. I don't think I've used "steal x base" since v11.

Edit

Oops forgot to ask. Does your league actual SB totals output and SB success rate match the expected output from the year/era in question?

[RchW]

See my post in an older thread regarding SB. Several hypotheses come to mind:

That OOTP cannot necessarily produce the same distribution of base/out states as real life that call for a steal, and that have a very good base stealer on base.

The exact combination of player attributes that produces real life SB leaders like Ellesbury, Bourn, Reyes, Gardner, Crisp et al may not work the same way in OOTP. We know that the AI may choose a better batter which will reduce the PA of a potentially superior base stealer.

Since league totals drive the output, stolen bases are distributed amongst more players with the best having slightly less than real life.

I'm not sure that this could be fixed. When you consider the number of variables involved it's astounding just how well OOTP does.

Originally Posted by RchW

Adjusting league wide stolen base attempts and success rates is very easy in OOTP. It's probably one of the most stable stat outputs from year to year. Just have your Commish look at the league total modifiers vs the previous year stat output.

Getting individual players rates up to the 90% mark is a little more challenging for OOTP as replicating the exact talent distribution as RL baseball must be difficult within the wide range of settings available in-game. The distribution of SB in OOTP is wider than RL ie more players have a few SB and the leaders have slightly less than RL. Over the last 9 seasons the average success rate of the SB leader is 84%. IRL there are players every year who get into the high 90% often 100% rate but they are opportunists like Jeter or Pedroia not leading base stealers like Ellsbury. It doesn't surprise me that a ratings based game has trouble with this.

OOTP may not produce the same set of base out states as real life baseball. It could be as simple as the the conditions for individual players to steal successfully do not occur at the same rate in each game in OOTP. That being said I get success rates in the 83%-88% for the top 5-6 in SB which matches the RL average above.

See the comparison below.

My fictional 2063 season.

2650 SB 1019 CS 72.8% success rate. Top individual SB% 83-88.

Projected 2013 MLB (20% left in season).

2666 SB 1010 CS 72.5% success rate. Top individual SB% for SB leaders 90-92.

Can't get much better than that.

[rpriske]

Oh, I am not trying to imply that OOTP is wrong. What I am saying is that I am trying to figure out how to optimize by base running decisions.

I don't want to run in situations where my odds are less than 75%... but I can't figure out who to calculate that.

In strat, you have a safe range that tells mw what I need to know. What I was trying to figure out was that same thing. Apparently there is no such thing... but that doesn't really make sense to me. If a scout tells me that a particular runner is a 15 stealer, what does that mean? How likely is he to be safe. If they tell me that a catcher is a 9 arm, how does that affect the odds of that runner being safe? What about the hold rating.

Apparently this ISN'T a equation that can be answered... and to me THAT is a failing of the game. We should be able to know what the ratings MEAN.

[Charlie Hough]

Originally Posted by rpriske

What I am trying to determine is when to run. It has been established pretty strongly that attempting to steal with a success percentage of under 75% is a negative strategy.

The problem with this is that there is no way to determine the statistical probability of a successful steal in real life. Sabermetricians can cite every conceivable stat from the past, but that says nothing about the present circumstance. No mathematical or statistical model can be applied to the present to predict the result. Not rationally. This is because the assumption that future results will resemble past outcomes is a logical fallacy.

There are numerous factors that you would have to control for anyway, most of which are never recorded or accounted for statistically. For example, what is the condition of the ball, the humidity, the level of perspiration in the catcher's hand, the wind speed and direction, the psychological state of all involved, the tagging ability of the player who will receive the ball, etc., etc., etc.

In real life, part of a manager's job is assessing some of these intangibles and not merely managing according to theoretical percentages based on past results. Honestly, if we want to manage everything according to sabermetric models, then we don't need managers. We can simply turn everything over to AI and have a notebook PC in each dugout that displays each decision.

So I throw all of that nonsense about playing the percentages out the window and make my judgments based on the ratings of the players involved, the game situation, and my inclinations of the moment.

With OOTP, at least you know that there is a baseline success rate based on real life tendencies. That will be adjusted based on the relative ratings of the players involved, so you have some sense of the chance of success. But you can't pinpoint whether a given circumstance will be 70% or 75% or 80%. But think about it: if it's 74%, then you never steal? If it's 75%, you steal every time? That's arbitrary, isn't it? And, even if you could know the percentage in the game, then you would have access to information that no real life manager could ever have. No real life manager can ever know what the chances of success are for a given steal attempt that he's considering. So this would be an artificial and unrealistic indicator.

Personally, I love to steal bases and will heavily use the options to steal, run and hit, and hit and run. I don't like relying on the next couple of guys to hopefully advance the runner over and then get a base hit to score a run. If I see a ratings match-up that I like, I'm going to send my runner. In reality, I'm probably playing to something like a 75% percentage, but I don't worry about that. Since I like to use the hit-and-run, I end up with quite a few steals by players who aren't necessarily great base stealers, especially on hit-and-run plays where the batter doesn't make contact. The percentages in those cases are supposedly lower, but that doesn't stop me.

If I have a batter who isn't strikeout prone but maybe isn't a great hitter for average, then I'm more likely to send the runner on a hit-and-run or run-and-hit, to at least move him over on contact or maybe get a steal. Again, I'm not going to trust that an average to below-average hitter is going to get a base hit.

If I have a great match-up but a batter who is highly prone to strikeouts and not putting the ball in play, then that changes things. I'm going to be more particular about whether I try a hit-and-run, but if the runner is really great, then I might steal or do the run-and-hit anyway. If the runner is good but there's a significant risk of him getting thrown out if the hitter doesn't put the ball in play, then I might be more cautious.

Things may depend on the game situation too. If I'm one run down in a tight game, but there are no outs and my runner is on first, then I probably won't try to steal. There, I don't want to risk wasting the tying run by getting my runner thrown out. But I might use a hit-and-run depending on the batter. If I'm pretty confident that the hitter will at least make contact, then I know that it's almost a certainty that my runner will at least end up at second, in scoring position with only one out. At best, he might end up at third on a single or he'll score on a double or bigger hit.

I've managed this way in computer sims for the past 30 years, and it never fails me. I'm able to rack up a lot of steals and score runs that I probably wouldn't have gotten otherwise. And generally my teams do extremely well and can outperform their base offensive ability.

[rpriske]

Originally Posted by Charlie Hough

The problem with this is that there is no way to determine the statistical probability of a successful steal in real life. Sabermetricians can cite every conceivable stat from the past, but that says nothing about the present circumstance. No mathematical or statistical model can be applied to the present to predict the result. Not rationally. This is because the assumption that future results will resemble past outcomes is a logical fallacy.

I like your post in general, but we disagree about this. You CAN have an idea about how likely someone will be safe when stealing. I would argue that EVERY manager who gives the steal sign DOES have an idea about how likely it is that the runner will be safe.

Now, it doesn't mean he is RIGHT and that is why we have 'fog of war' with the ratings. They aren't the REAL ratings. They are what our scout THINKS they are.

All I am asking is what that means. If a scout says he thinks a player is a 15 base stealer... what does he MEAN by that? How likely does he think it is that the player will be safe, on average?

To say that the scout doesn't know that is just wrong. This is no different than a scout saying that he thinks a player will be a .250 hitter or a .300 hitter.

So... what does he mean?

[Charlie Hough]

If that's all you want to know, then simply use 1-year recalc for a historical game, look at the real life SB vs. CS totals for the players with the best base stealing ratings, and then figure out how the game translates from percentage rates of success to the ratings.

Of course, keep in mind that the game is using league totals as its basis, so, strictly speaking, the ratings may not be based solely on the individual's stats. They may also be relative to how those stats compare to the overall league numbers.

It's no different than looking at historical players who hit .350 vs. .325 vs. .320 vs. .300 and figuring out how this computes to ratings for that given season relative to all the other players. Of course, you would have to use the editor and look at the ratings for the game engine's full scale to really refine this. 1-10 or 1-20 doesn't compare to how the game handles ratings behind the scenes, which is on a much larger scale of rating.

If you did this for enough eras in baseball history, including the latest MLB season, then you'd have a general idea of what the ratings mean relative to the rates of success.

[rpriske]

That makes sense. Thanks.