| RotoGuru Baseball Forum View the Forum Registry |
| |
| 0 |
Subject: Pitching Statistics, Part II Posted by: Madman - [146191423] Tue, Mar 06, 04:08 Warning: here comes some intense stuff. What I'm going to do is to provide empirical justification for why rotating cheap pitchers didn't work in 2000, and why it probably won't work again in 2001. Although picking pitcher money trains isn't that important in Smallworld, you'll be asked to rotate into pitchers more than you did before. This information should help you greatly with that endeavor. Further, even though all the evidence is in terms of SWP, this analysis generally applies to Swirve, as well. You'll definitely be rotating pitchers in that game. So at the risk of giving away my competitive edge, I present to you "Pitching Statistics, Part II: Why Rotating Cheapies Will Get you Nowhere". ----------------------------------------------- Below you will find a rather complicated table. But it's also incredibly useful, IMO. All statistics are for the 2000 season. I analyzed the 2000 season game by game in an attempt to find out how pitching works. The magnitudes of some of these effects I'm going to discuss really astounded me. The first row in the table indicates what ALL starting pitchers did in 2000. There were 4858 starts during the year. In 2504 of those 4858 starts, the pitcher scored more SWP than what the pitcher's average would be at the end of the year. This occured 51.5% of the time. The next column tabulates the average deviation for all of those starts. A deviation is just the points scored by the pitcher in the individual game minus the pitcher's seasonal average. Obviously, if you're averaging the deviations over all starts over a full season, the average deviation will be zero, which it is in the table. The "Quality" column represents the average of all the seasonal averages for the pitchers who made the starts in the given row. In this case, this is the same as the average point total per start -- 29.9. Now, to the good stuff. What did pitchers do after a "better than average" outing? For simplicity, I'll call such an outing a "good" (G) outing. The reverse -- a below average outing -- will be called a "bad" outing (B). Thus if Pedro scores 88 SWP, that's a "bad" outing, whereas if Dan Reichert scored that same total, that's a "good" outing. I want to see what makes a pitcher pitch above his/her ability level. Thus my comparison to a pitcher's average. It turns out that Good outings were followed by another start 2409 times. In only 1191 of those starts did the pitcher duplicate his feat of a "good" start -- 49.4% of the time. In those starts, he averaged 3.08 WORSE than his average. The next row represents the starts that followed a single Bad outing. In this case, that occurred 2147 times, and 1162 of those resulted in a Good outing -- a whopping 54.1% of the time!!! Dun-dun-dun (melodramatic organ music, please). In other words, pitchers did better after bad starts than after good ones. The difference in SWP is only about 5, however -- 1 out. Perhaps not enough to make any real decisions with . . . But, let's break it down farther. The next 4 rows of the table represent starts that followed two outings -- either two good outings, a good followed by bad, a bad followed by good, or a bad followed by bad. To make a long story less long, the probability of a subsequent start being good rises with the number of times the starter had a BAD outing in the previous two times!!! Dun-dun-dun . . . The next 7 rows represent all starts that followed THREE consecutive outings . . . Notice that pitchers after three consecutive above average starts were a whopping -10.7 points below their seasonal averages. Pitchers after three bad starts were a whopping 10.08 above their seasonal averages! Dun-dun-dun . . .
So, what should we conclude from all of this? What could be causing this? First note that the quality of starters in each of the categories is roughly the same (the first row has a lower average probably because starters who only made one start tended to be of lower quality). So, the quality of starters isn't causing this phenomenon (my focus on deviations was to try to prevent this sort of interference, but I checked to make sure). Secondly, it's possible that there's some attrition from the data set that is rather critical. If a starter gets injured and has three awful performances in a row before he goes under the knife, I'm NOT going to capture that in my data. In fact, the falling numbers of starts in each category supports this notion -- namely that if you have 3 good starts in a row, you're likely to get a 4th. If you have 3 bad starts in a row, you're likely to get benched. I can't prove this with the numbers above, but it seems quite plausible. Therefore, you still have to be very careful when picking up pitchers after their third awful start. . . Thirdly, notice that I'm using some information no manager has available at the beginning of the year -- the pitcher's end-season point average. You can't really use a pitcher's average through 4 weeks of a season as a substitute, either. Therefore, caution should be used when applying this data. But with that said, the above data indicates that there is no real reason to avoid pitchers after their bad outings, as long as you think they are healthy and will continue to get chances. In fact, managers who jump on bandwagons after a pitcher has 3 consecutive good starts are likely to suffer serious reductions in subsequent SWP accumulation! Notice that the reduction in monetary swings in SW in 2001 has now freed you, the manager, from being "forced" to jump on the cheap pitcher bandwagons. There are no longer any rewards to do so! Pitchers aren't likely to become money trains. And, it's likely that a cheap early season pitching bargain will remain a cheap pitching bargain throughout the season, since their price increases are likely 1/5th of what they would have been last year (and even last year, cheap pitchers tended to stay cheap). Therefore, you can exercise patience in 2001. Only buy cheap pitchers who you think will have high season SWP/G averages. Don't jump on those guys who you think over-performed once or twice or even three times! Further, if you starter bites the big one once, twice, or even three times, as long as you think it's healthy, it's time for a gut-check . . . he might be ready to turn it around! Of course, he might also be hurt :) Clearly, therefore, I think pitcher management in 2001 Smallworld has much nicer incentives than anything we've seen before. By effectively outlawing pitcher money trains, you can concentrate on point production from pitchers, rather than being forced to grit your teeth and cross your fingers to snag the latest train . . . This information is powerful. Use it with caution! See you in the Smallworld, Madman (errors in calculations are possible, of course - - I wish I had the energy left to beautify that table and put a divider between the different rows . . . but I don't.) | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 1 | JKaye ID: 4711592917 Tue, Mar 06, 06:39 |
Nice work. I think many people had suspected this to be the case, just based on experience, and the numbers clearly prove it. My strategy from the winter, before SW even launched the site this year, was not to go after any $$ from pitching, and build RV from hitters. Several things have occured since then that have given me even more confidence in the plan. 1. SW has made stud pitchers at mid range prices allowing a reasonable opening day roster, and therefore trade conservation. 2. SW basically got rid of pitching price changes, making any rotating that is not geared towards SWP alone, usless. 3. Madman's data regarding the consequences of rotating cheap pitchers. All have contributed to what I feel will be the best strategy for 2001. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 2 | Pond Scum ID: 54420321 Tue, Mar 06, 08:07 |
Another thought-provoking post and I agree with the conclusions. However, isn't it true that because the season end averages are being used, that the chances of the game following a string of below average ("B") games is most likely going to be above average ("G")? In the extreme, if every game for the season before the final game were B, then the last game would have to be G and in a very big way! So it is quite to be expected given the methodology, that the odds of a G game (so defined) go up the more B games there are in a row. Am I missing something? | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 3 | Myboyjack ID: 4443038 Tue, Mar 06, 08:28 |
Madamn, Did you really feel the need to use the "his/her" phrase when talking about major league pitcher? ;) Nice work. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 4 | biliruben ID: 231045110 Tue, Mar 06, 10:34 |
Madman- I think Pond Scum is on to something. I think your results are at least partially result of looking "prospectively" at retrospective data. You are assuming that final start isn't correlated with the previous 3, but when you key off the final season average in order to determine to quality of the start, it is correlated. Great work, none-the-less! Sorry I took so long to read it - couldn't see the gems through all the drafts. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 5 | Madman ID: 146191423 Tue, Mar 06, 14:28 |
Pond Scum, biliruben Ah, I see I can't slip anything past you guys :) . I had been thinking of ways to analyze this data, and this was the best that I could come up with. Feel free to believe me or not, but I was very concerned about this issue (one of the reasons I mentioned the seasonal averages and their limitations in the post). But I don't know an easy way around it at this point. Since you guys insist on picking my work apart :) (if no one ever picked apart my work, I'd cease to post, BTW), consider the following two tables. The first table is the same as above, but now restricting the sample to pitchers with 16 or more starts.
Now, let's take this to the extreme, and consider only pitchers who had more 30+ starts.
If the relationship between the finite sample size and the use of the seasonal average were the primary driving force behind my numbers, you'd expect this bad-start/good-start effect that I'm talking about to shrink as I increase the minimum number of starts. This didn't occur. The point here is that this is a fairly general phenomenon, it would appear, striking both stud starters and the guys getting a few appearances. If anything, the guys with just a few appearances in the year appear to be a bit more consistent . . . at least if they have a few bad outings in a row, they really are that bad . . . Finally, consider some back of the envelop math. Let's say that there's a 50% chance of any start being good, and that a pitcher has 32 starts in a year. Furthermore, let's assume that the distribution of his starts is symmetric, meaning that 50% of his starts were less than his average. This would mean that you'd expect him to have 16 good starts and 16 bad starts. If he also had 3 bad starts in a row, this means that 16/29 of the remaining starts were "good". This would represent 55% of his starts. Notice that I'm finding, on average, 60%+ positive outings after three bad outings. Also notice that empirically the sample is not necessarily symmetric. It's quite possible for 1 good outing to outweigh 3 bad ones, or vice versa. Therefore, there is no mathematical requirement that there be any such finite sample size effect, although I grant that it is plausible. Therefore, I think there's something going on here besides this sort of finite sample size phenomenon. I honestly wish I could have figured out a way to analyze this sort of thing without this problem. But, all other approaches seemed fraught with even bigger empirical issues. And it does not seem plausible to me that the finite sample size issue is responsible for the striking numbers above. Programming error, maybe :). myboyjack LOL! Just habit. There have been women pitchers in professional baseball, but from what I've seen, there's no possibility of a woman in MLB in the immediate future, if ever. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 6 | Madman ID: 146191423 Tue, Mar 06, 14:50 |
One more thing -- when I said that there was no mathematical necessity for the small sample to be a problem, I meant in theory. If you make various extremely reasonable assumptions about the distribution of starts and such, you'd expect it to play some role -- especially with those pitchers making few starts. I was mainly trying to emphasize that my simplistic example which bumped a 50% good-start chance to a 55% chance probably over-stated the case, since in practice those 3 bad starts could be outweighed by any number of good starts. Perhaps I could look at 3 bad starts that were relatively close the person's average or something . . . nah. I think the point is made. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 7 | biliruben ID: 3502218 Tue, Mar 06, 14:58 |
Hmmm... Your theoretical 55% chance vs. your actual 60% finding is fairly convincing that there is something here. Maybe road vs. home? I can't remember. Is there any established general trends in quality starts on the road vs. at home? I can see a pitcher getting shelled during a road trip, and coming home and putting in a good start in front of the home crowd. I have definitely seen this for individual pitchers in individual years, but I am not sure if it is more than chance. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 8 | walk ID: 212358 Tue, Mar 06, 15:00 |
Thanks mucho, madman. Your analysis has not led to any paralysis, and it is clearly mucho beneficial in terms of pitching management. - walk | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 9 | John G ID: 50216615 Tue, Mar 06, 15:32 |
Instead of using full-year averages, have you considered using year-to-date averages or a rolling 5-previous start average? That's information we have when we make decisions. Also, have you considered grouping similar pitchers such as aces and 50 SWP/start pitchers? | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 10 | Pond Scum ID: 54420321 Tue, Mar 06, 15:53 |
I think John G has a good suggestion, would that I had the means to do the analysis. Stepping back from the detailed math, the conclusions suggested by Madman make great sense to me (that sounds a bit like an oxymoron, but oh well...). It has occurred to me in the past that if your strategy is to trade on bad news (right after a bad outing) and buy on good, that you are upping the chances of increasing the bad's to the good's. For example, by trading away on bad news, you have absorbed the negative impact (in points) of the poor outing but foregone the benefit of any recovery the next time. By buying on a good outing, you just missed that one and are exposed to whatever the odds are for a bad outing next time. On average, you keep the bad and miss the good. I think what this says first and foremost is that you should buy premium pitchers that on average will do well and stick with them through the occasional bad outing, however tempting fright then flight may be. Secondly, if you do venture into cheapie territory, do your research very well and believe in it until you have the funds to promote the player to a more premium member of your staff. Preserve the trade for later. Both cases assume that a structural change in your pitcher's condition has not occurred. If you do have to trade defensively, does this mean that trading TO the pitcher with the good last two or three games does NOT make sense? While the analysis above may imply that, I am not ready to buy that yet. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 11 | Madman ID: 146191423 Tue, Mar 06, 16:28 |
biliruben Dang the Colorado factor. Honestly, I totally over-looked that issue. Geez. For most normal parks, I doubt this is a significant issue. But Colorado could really be playing with my data -- at least in the "after 1 good start" stuff. . . Don't have the energy to do re-programming necessary on this score . . . An eyeballing of the data doesn't really make it look like Col. is driving everything. There are most definitely Col. starts in here. But road teams never have 2 COL starts in a row. If anything, the awful COL. starting pitcher stats may be HIDING the trends. Bohanon, for example, crops up a billion times in the 3-B category . . . back to back to back to back to back bad outings . . . But he's not in the 30+ start category (wonder why!). Astacio is, but he isn't really messing up the data set. It's kind of remarkable the number of better than average outings he put together in Coors. Arrojo is early in the season. It looks like Colorado pitchers are susceptible to major funks. They get shelled a few times in COL, go on the road, the trend continues for a bit. More evidence that this effect may get bigger taking COL out of the dataset. John G Good points. 1) year-to-date averages. I thought about this. It was going to be a bit more work. I was also worried about the first few starts. You could have a "bad" outing that's 136 points. Like Pedro last year. 148-136-107-104-120-126-200-118-46-157-170-56-67-45-85-136-175-etc. Your question is really "which of those outings should be called 'bad'", right? My method says anything greater than 112 points. A rolling average differs from this in the classification of the 2nd start and 8th start. In both cases, I'm calling those starts "good" and a rolling average would call them "bad". Even if I didn't know Pedro's average would eventually be 112, I would probably count both starts as good. Actually, I might invent a third category, called "average" for the 8th start. Which brings me to a potential compromise solution -- not directly addressing your question, but kind of. . . Here's a repeat of the table, this time only looking at starts that deviate from the mean by more than 14 points -- something you could probably eyeball at any point in the season, simply based off last year's stats, or even using a rolling-average approach (also for pitchers with 30+ starts).
The BBB category likely suffers from the finite-sample size problem -- and the fact that there aren't that many observations (an important fact in and of itself!). Those cases are like Wolf in August (4 bad outings in a row, 2 of which were -80 or worse, and 3 of which exceeded the 15 point issue. He came back to post a net +140 over the next three starts.), Rogers in Sept. (he actually had 4 awful games in a row, but then turned it around to post 2 40+ deviations in a row). So, I'm not really answering your question (that would take more work besides just "replacing" some >'s and <'s). But I think you could rely on just "obviously" good and bad outings and get similar sorts of predictions. In other words, keep an eye out on all the "ace" pitchers. If they post 3 really bad outings in a row, consider jumping on them. This would have netted you Al Leiter's +160 (total) string from 5/5 to 5/17 (note: after a string of 3 great outings like that, consider dumping). This even worked for Albie Lopez in Sept. -- although he had a 3-B followed by a net 7 performance, his next two after that were 93 and 43 to end the season. BjJones late in May would have burned you -- 3-B followed by a -54 outing (YIKES). But his next 3 were a net +170! Aha! Found a counter-example. Brad Radke . . . August. 5 negative deviations in a row. The 4th was -136 by itself, a -100 SWP performance. So this method isn't perfect. . . Schilling in early June bombed 6 in a row . . . (Schilling himself may not fit this general pattern. He bombed 5 in a row later in the year). I could go on and on with specific examples. There is a lot of risk involved. But in terms of playing the odds, it looks like there really might be something to this method. I guess it can all be summed up -- Think a Bit Contrarian with pitchers. . . ---------- Regarding your second question -- I actually did do this in some preliminary work. When I changed approaches and conditioned on the number of starts, I didn't bother. Notice the "average quality" of start going up between the first three tables? I know that there are still a few bad pitchers in there that made 30 starts. But these are all definite MLB-calibre pitchers. So I think you can roughly take the 30+ start tables as indicative of "aces". Limiting it solely to really, really good pitchers would give me even smaller sample sizes than the table in this post . . . Keep the ideas coming! If anyone wants to do some work with the data, let me know. I have all the starts in an Excel file . . . have to work on figuring out a way to make it generally accessible if there's demand. --------------------- P.S. One additional strategy idea -- notice the BBG row also has a very high expectation for the subsequent outing. You could wait until a pitcher turns a bad streak around and has a good start . . . might hedge your bets against arm trouble this way. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 12 | Madman ID: 146191423 Tue, Mar 06, 16:33 |
Oops. The first row of the table in post 12 is incorrect -- I keyed those tabulations from a different part of the data set . . . relatively unimportant, so I won't bother to fix. . . Just an FYI. Pond Scum I'll work on posting the file somehow. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 13 | Rolodex ID: 1723021 Tue, Mar 06, 16:45 |
I think I'd find it hard to drop a pitcher after 3 G's, even after reading your convincing data. Just opinion of one, though. peace yo. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 14 | Madman ID: 146191423 Tue, Mar 06, 18:20 | It was hard for the successful managers to get away from Randro last year, too :) . It's all about playing the odds. . . | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 15 | Madman ID: 146191423 Tue, Mar 06, 18:27 |
Actually, let me say that I probably won't be selling 3G pitchers in Smallworld, either. If in doubt, always keep a trade rather than spend it. But I probably won't be picking them up too often (depends on opponent and pitch counts and stuff). The 3G issue is probably more important for Swirve where you rotate into and out of guys so often. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 16 | Free Gamer ID: 211461812 Tue, Mar 06, 19:03 | A very interesting analysis Madman! I'm new to RotoGuru.com and am amazed by the effort and intelligence with which you play these games. What I assume your research is demonstrating (and you may have written without my noticing) is that player performances tend to revert to the mean -- something all sports fans are somewhat intuitively aware of. Just as a player's sw price can't go up forever, player performances eventually "come back down to earth." | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 17 | Guru ID: 330592710 Wed, Mar 07, 10:00 | Madman - now if you can just tell us what each pitcher's end-of-season average will be, then I really think you're onto something! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 18 | egl ID: 52230716 Wed, Mar 07, 16:53 | SO THIS MEANS LIMA SHOULD BE READY TO GO ON FIRE! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 19 | Madman ID: 146191423 Wed, Mar 07, 17:58 |
egl If you believe that Lima's real ability is higher than his last few outings, you're darned right! This is where the "art" of scientific application comes in. You have to distinguish between the cases in which you can manipulate a known phenomenon to your advantage (i.e., a counter-cyclical trend in starter's points) versus a fundamental change in a starter's overall ability. Obviously, knowing these fundamental ability levels is the first step. For the record, I developed this counter-cyclical intuition last year and still didn't pick up Lima. Although I was tempted a couple of times. You can't blindly apply information. Information is power only if you use it powerfully. Thus the Guru's scathing yet pithy remark that the bigger question is to predict this overall ability level . . . Nevertheless, if you feel confident that a guy like Pedro is healthy, and you see him with 4 sub 90 point outings (like last year, May), I'd jump on him in a heart-beat. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 20 | Catfish ID: 71158811 Thu, Mar 08, 11:06 |
Madman, at the risk of wearing out your pencil, is there any correlation between patterns like GGG and BBB and cumulative innings pitched? I recall last season some good observations about the game after a high pitch count. The next game sometimes showed a decline in performance and fewer innnings than usual for the pitcher. I guess the reverse could hold -- three bad starts in a row and your manager is going to be pulling you early. Leaving you just full of juice for the next start and ready to pitch until your arm falls off 199 pitches later. I'm speculating that you may be proving that a rested pitcher, even if he/she is in the doghouse, is likely to do better than an exhausted one. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 21 | Madman ID: 146191423 Thu, Mar 08, 14:07 |
Catfish -- good points. I was going to do a pitch count analysis and an IP analysis as well. Just haven't gotten around to it yet. No promises, but I would indeed like to. If this is a causal relationship, you caught on to exactly why it is likely happening, IMO. To tell you the truth, I think it's not only plausible but inevitable. One thing by looking through the data, quite often a pitcher will come off a bad streak of games with 80-90 pitchers each. Suddenly break out with a whopper by throwing 110-120 pitches, and still be ok for another start or two, but at lower pitch counts. And then tail off a bit again . . . Similarly, if they have been throwing a ton and THEN throw the 110-120 pitch outing, I think they're much more apt to see a serious decline in productivity (short-run). But that's just by glancing at the data. And by looking at SWP to gauge effectiveness, I'm clouding some things, because Wins aren't really under a pitcher's control. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 22 | walk ID: 212358 Thu, Mar 08, 14:11 |
..."scathing yet pithy remark..." Madman, you are on a roll of unequaled proportions, both with analysis and prose. Keep it up man, and if you are playing this year, fret not, I won't strain my neck too much from watching you from behind (non-perverted metaphor). ;-) walk | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 23 | Madman ID: 146191423 Thu, Mar 08, 16:05 |
walk Glad to know you're enjoying it! There's one thing that I'd like to point out to everyone. And that is I actually enjoy figuring out strategy and calculating optimal decision making a lot more than I enjoy playing the game. I know that this makes me exceptionally weird. But it's true, nonetheless. I appreciate all the flattery I've heard either here or in the chatroom, but it's largely misplaced. For example, in football pickoff, I worked like a dog for a few days on Sludge's optimal strategy calculations, trying to improve his algorithms. Once I accomplished that task, the actual season was a real let-down. Suddenly, the whole thing wasn't nearly as interesting anymore. I ended up getting distracted by other little math problems and such (a good thing for my real life, BTW). So, although I strongly believe I won't get too upset with any SW service issues this year (as long as they are remotely truthful), I'm also not sure that I'm going to be able to really go all-out. And in a game as stacked with great competition from the likes of yourself, the gurupies, and all the unknown lurkers and others out there, anything less than all-out from anyone is a sign of trouble. What I really want to do is to make a positive difference in the strategies of whoever ends up winning, or even better, the top 10 teams. I'd also really like to have the Gurupies sweep the top 100. Or as close as we can come. This is a very hard task, since there are a number of great players who've never seen this site. That would give me more kicks that doing it myself, actually. With all this said, I'm still going to try to make everyone eat my dust :) . | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 24 | HooeyPooey ID: 41115208 Thu, Mar 08, 16:44 | Any reason why the charts are missing BGB stats? | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 25 | Madman ID: 146191423 Thu, Mar 08, 17:02 |
HooeyPooey Yeah, because I'm an idiot. D'OH!!!!!!!!!!!!!!!!! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 26 | HooeyPooey ID: 41115208 Thu, Mar 08, 17:50 | I wouldn't go that far. I'm still trying to work out a theory why we see the 60% trend, but I haven't come up with anything yet. Since the overall G% is somewhere between 51.5% and 52.8%, this seems to be helping the BBB to G ratio a little. Altering your example above, say a pitcher is expected 16 good starts and 15 bad (51.6%) and you take three consecutive bad starts away, the remaining starts is 16 good of 28 total, or 57% which still isn't 60% but it seems to tighten the gap on why your observation is as such, especially the more you go above 50% win ratio. Maybe I'll have to look at the data for more ideas. Nice analysis though. It's a lot easier to critique your work than to put it together in the first place, I'm sure. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 27 | Madman ID: 146191423 Thu, Mar 08, 18:11 |
HooeyPooey Yep. As an expert dam-buster, I know first hand that it's easier to poke holes in 'em than in building them in the first place. Actually, the BGB slipped through because I was really not too interested in all the combinations -- mainly the extremes. I still should have caught it. Regarding your point about the %'s, you have a valid issue -- and you might be able to construct an alternative theory for the 60% figure. But don't forget about my post with only extreme GGG or BBB starts. Although those are estimated with less precision (note: I didn't give standard errors for the %'s, but you can calculate them based on the table if you want), the %'s are more extreme there. Of course, there are still holes you can poke -- namely that you're a long ways below the mean with 3B's that are all worse than 15 below your average, therefore, it's more likely that you'll experience another fairly high draw in the remaining starts, etc., etc. But why would the next one be 17 points above average? This seems pretty extreme to me, and counter to the commonly believed "streak" theory . . . Did you notice that you're already implicitly giving the "streak" theory up? With the streak theory, the next start would be *more* likely to resemble past starts, even with my measurement method. Yet, you're comparing it to no relationship at all (pure random draw)! So, even if you can convince me that the 60% is coming from pure random chance, you'll still have to go a long way to telling me to not pick up Pedro after 4 less than 90 point outings. I guess in my excitement at having found this counter-cyclical trend, I forgot this more basic point that there is no reason to believe streaks exist with pitchers. Actually, that's a bit of an overstatement. Streaks do clearly exist. In my data there are pitchers with 5-7 bad or good outings in a row. That's more than a month of good or bad outings. However, there is some evidence that (so far unproven) forces exist and on average outweigh the streak factor, making them impossible to actually predict with any certainty. Further, these alternative forces (like arm fatigue, manager over-use, luck, etc.) are so great that, on average, the reverse appears to be true -- that there's a negative relationship between points in the immediate past and points in the immediate future. Oh well. I'll be calculating pitch count numbers and *maybe* IP later today sometime. If they end up inconsistent with the above story, I'll weigh that appropriately. But my hunch is that they'll support it, based on some preliminary work that I did in July. Although you never know. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 28 | HooeyPooey ID: 41115208 Thu, Mar 08, 19:26 |
I'm trying to figure what the value in the Quality column represents. One, I'm not sure how the Quality for All starts in the very first table (29.90) is lower than the quality for every breakdown of starts below it. On the most recent table, the quality for BBB is significantly lower than most of the other entries. I'm not sure if that possibly has something to do with the extreme 17 point difference or not. I'm also wondering how much of an effect points for wins and losses has, though I doubt that really plays into it. I just remember it does tend to skew the data a bit. As much as I love playing with numbers, stats, and probability, I am more or less looking at this passively as I don't really have the time to delve into this as much as I would like. Sorry if I am overlooking some obvious information. This is very interesting. Thanks for your hard work! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 29 | Madman ID: 146191423 Thu, Mar 08, 22:37 |
The Quality column represents the average SWP of all starts that occur in the given category. Essentially the weighted mean of the individual pitcher means that were used to get the individual pitcher deviations that were used throughout the rest of the analysis . . . (like that makes a lot of sense when said like that!) The reason why the 29.9 is lower than all the other numbers is that there are starters in the "ALL" category that never pitched a second game. Guys like Borkowski (-122), Beime (17), etc. If you've never heard of those guys, that's because they spent between 1 and 5 days in the majors. Typically, those guys stunk it up big time, thus a 2-3-4 point lower average for EVERYONE from just including the relatively small number of 1-start dudes. But this should make a lot of intuitive sense . . . In other words, the group of pitchers in the "ALL" column is of a lower "Quality" than the groups in the other situations. For example, Pedro is in both . . . The quality by itself wouldn't affect the 17 point difference. This because all the other numbers are deviations from each pitcher's mean. In other words, the 17 is saying that you are 17 points better than your (individual pitcher average). In other words, pitchers with lower overall average (i.e., lower quality) were more likely to have 3-B's in a row (either that or their 3-B's were worse than average). However, the average of all pitchers who had 3-B's was 17 points higher than what you would expect their average to be. Therefore, the only way the lower overall quality would be affecting the 17 point figure would be indirectly -- if lower quality pitchers tended to be more erratic; specifically more erratic in a negative way in which 3 very bad starts would suddenly result in one great one . . . | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 30 | Madman ID: 146191423 Thu, Mar 08, 23:16 |
If anyone wants a look at the data, I've put a comma-delimited text file up on my new website for such endeavors . . . Madman's New Home Page. In addition to this file, I put both Draft Analyzers up there, as well as the beta version of the SW Roster Manager. I anticipate that I'll release an updated version of the SW Roster manager this weekend. At some point, I also plan to release a version of the program that I used to generate all this cool pitcher data. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 31 | James K Polk ID: 4211362123 Thu, Mar 08, 23:23 | Roller Coaster Tycoon. Who knew? :) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 32 | Madman ID: 146191423 Fri, Mar 09, 01:39 | Yeah. It's a great little game. My brother-in-law got me hooked. It works for all ages and genders, IMO. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 33 | Pond Scum ID: 54420321 Fri, Mar 09, 08:15 | Just looked at the Draft Analyzer, very cool. Thanks, Madman!!! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 34 | Madman ID: 29246911 Fri, Mar 09, 17:39 |
OK. The Guru happened to mention that it would be nice to know the end of season pitcher averages to put this information to good use. In post 11, I showed that this 3B versus 3G effect existed even in more obvious cases -- namely when a pitcher was clearly below par. Here is some additional evidence. Technically, it still suffers a bit from some of the same sample size issues, etc. But it should start to help illustrate that it is actually not impossible to identify pitchers in the different categories. What follows here is a table with information based on INNINGS PITCHED . . .
This is coming from the same data that was used in post 5, table 2. Namely, pitchers with 30+ starts during the season. I similarly define a G outing as one in which the pitcher had above average SWP. However, here I'm tabulating IP. Notice that an average "G" occurs with 7.17 IP. Pitchers were able to replicate a G 52.1% of the time. No big deal so far. But as they try to follow 2G's with a third, this probability drops to 48.4, and then to 43.6 after a 3rd successful G. This data is so far mostly a repeat. What I found interesting, however, was that you can basically identify these folks, since a "G" outing is 0.50 inning more than a regular outing. For just one start, this isn't a big deal. But if a pitcher consistently throws 7+ innings, there's a darned good chance that he's had 3 G's in a row. Even 1 bad outing in the last 3 drops the average IP for the last 3 to a high of 6.6. Further, this logic works in reverse. Take a decent pitcher who's been under-worked over the last 3 starts. If a pitcher is averaging under 6 IP for three starts, there's a good chance he's in the BBB category. Of course, you can supplement the IP information with the SWP information. But I think what this does is support the idea of pitcher fatigue. An average pitcher in this data set goes for 6.6 IP. The more outings in a row that he sustains a higher average than this, the higher the probability of a set back in the near future. Yes, I'm still using the "3 under the average" SWP calculation to determine the G's. But the correlation between SWP is obvious in the scoring formula. And it's obvious here in the data set. Further, it's also easier to really observe a pitcher's recent workload. I dunno. There are a zillion permutations of this that you can do. BTW, I also checked the Pitch Counts. The results were almost identical, as you would expect. Pitch count averages ranged from 98.9 (3B) to 107 (3G). Interestingly, the increase in pitch count after 3B's was only 3.28. Pitchers are getting a bit of rest with bad outings, and worked a bit harder in good ones. More interesting, the increase in pitches is rather small relative to the number of extra outs that they are getting. Therefore, it's the pitch count per IP figure that is changing dramatically. I say that this is interesting, since it was what I used to track pitcher fatigue and predict changes in fortune for pitchers last year. If the pitcher had a good outing, but a bad pitch to out ratio, I stayed away from him. If the pitcher had a bad outing, but relatively low pitch counts and decent IP totals, I tended to buy. Technically, this is another testable hypothesis, I suppose. But I think the results are so inter-related to all the work I've already done, I don't anticipate many more insights, so I'm probably not going to do it. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 35 | HooeyPooey ID: 41115208 Sat, Mar 10, 21:23 |
Upon further review, it seems to me that if "you're a long ways below the mean with 3B's that are all worse than 15 below your average, therefore, it's more likely that you'll experience another fairly high draw in the remaining starts" is why the numbers show as they do. With each consecutive bad start, in order to attain the determined year end average, the probability of a good start increases. The greater the deviation is negatively, the more likely it will be that a greater positive deviation occurs. If not, then the mean would be lower and therefore, the negative deviation would also be lower, which has the potential to starts to invert from bad to good. I was somewhat surprised to find that the average deviation of a good start was about 46, and a bad start about -49. I would have guessed it would not quite be that much at first glance. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 36 | Madman ID: 29246911 Sun, Mar 11, 00:07 |
HooeyPooey Yes, there's a higher probability that one of your remaining 27 starts will be above average. But why the next one????? | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 37 | HooeyPooey ID: 41115208 Sun, Mar 11, 08:43 | I am wondering how you calculated the data in post 11. When I calculated what happened in the next start after three consecutive <-14 bad starts, I came up with different results, namely 124 of 202 good (61.4%), with the average being 12.25. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 38 | Madman ID: 29246911 Sun, Mar 11, 20:47 |
Hooey Pooey Note first that your argument in 35 is inherently contradictory. 3 Bad starts put a pitcher -150 or so down from "average". If spread out over the other 27 starts, this means you would roughly expect the next start to be +6. Instead, the next start is (by your caculations) +12. By mine, +17. -------------- In post 37, you are getting 202 good starts because you are looking at all pitchers. You would expect your figure to be more biased by the finite sample size problem you are concerned about. My post 11 dealt exclusively with pitchers who had 30+ starts (note the parenthetical immediately prior to the table). Counter to what you would expect if this phenomenon was driven by the finite sample size issue, this problem gets worse as pitchers have more starts. Yet more evidence against the hypothesis that each start is a random draw. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 39 | HooeyPooey ID: 362451122 Sun, Mar 11, 22:48 | I do think the data does indeed show that the next start after three bad is typically higher than what the probability for this scenario should be. I didn't go to the trouble of calculating what the average 'should' be in this scenario as a whole for all pitchers, but I did for fun, run a few random samples, in that each pitcher still had the same start results, but I randomized the order for each pitcher. I don't have the data in front of me because it is at home, but I found in the three samples I ran that the G after 3B % was about 57% (which is about what we figured it should be) and that the average was about the 6-8 range. (This also close to what we would expect.) Excluding bad starts less than -14 did not seem to affect the average. Therefore, I do concur with your reasoning. The occurance of a good start and higher average after three bad are both higher than probable that that of a completely random draw. Now I'm off to read your Part III. :) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 40 | HooeyPooey ID: 362451122 Sun, Mar 11, 22:53 | BTW, don't think that I was trying to rip apart your work. It was just nice to know what the expected probability was in the described scenarios, so that the difference in that and what actually occurs could be measured. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 41 | Madman ID: 29246911 Sun, Mar 11, 23:40 |
Feel free to criticize the numbers. I think I probably was getting a bit agitated, however. Primarily, I haven't seen you post much around here (at least I didn't recall it right off), and was suspicious that your name, HooeyPooey was chosen to reflect a criticism of all this. I can take an argument about this number being wrong or backward. But I don't take well to irrational name-calling. I thought about doing a wise-crack or two about what's really hooey-pooey :) Because I realized I wasn't sure whether or not you were indeed trying to make a subtly juxtaposed commentary between your posts and your name, I tried to take that stuff out, but some of it probably seeped in, anyway. Your criticism that I didn't 'prove' the degree to which this was a 'statistically significant' result is a valid point. Sometimes I look at a result and am so shocked that I think it speaks for itself. Plus, despite the long-windedness which is apparent, I do try to limit my post-length! | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 42 | HooeyPooey ID: 362451122 Mon, Mar 12, 00:20 | I'm primarily a lurker though I've frequented the boards for over a year. But it's hard to pass up a good statistical analysis. It's like you said somewhere, this is the enjoyable part. The competing in the actual event isn't as thrilling. (I don't even like playing SW that much.) I suppose a few things could be said about Madman but you've already established your reputation through your posts. :) They are enjoyable to read. HooeyPooey is probably what my teams will be by the end of the season. It also parallels my apathetic attitude towards most things, but not so of a good discussion of baseball stats. I have better things to do than criticize just for the sake of stirring up arguments. :) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 43 | Madman ID: 29246911 Mon, Mar 12, 06:26 |
LOL. Glad to know I misread your intentions. A lot could be said about "Madman". I was ambivalent about using the name, actually. But had to come up with something. Kind of stuck with it now. Always feel free to pipe up. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 44 | F Gump ID: 25213 Sat, Jun 16, 16:54 | BUTT in re to next chapter of this info being posted today | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Rate this thread: | If you wish, you may rate this thread on scale of 1-5. Ratings should indicate how valuable or interesting you believe this thread would be to other users of this forum. A '5' means that this thread is a 'must read'. A '1' means that this is a complete waste of time. If you have previously rated this thread, rating it again will delete your previous rating. If you do not want to rate this thread, but want to see how others have rated it, then click the button without entering a rating, or else click here. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Post a reply to this message: (But first, how about checking out this sponsor?) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||