Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.2 9/5/84; site philabs.UUCP Path: utzoo!linus!philabs!dpb From: dpb@philabs.UUCP (Paul Benjamin) Newsgroups: net.sport.baseball Subject: NL catchers, AI, and philosophy Message-ID: <409@philabs.UUCP> Date: Sun, 11-Aug-85 16:18:35 EDT Article-I.D.: philabs.409 Posted: Sun Aug 11 16:18:35 1985 Date-Received: Tue, 13-Aug-85 02:03:56 EDT Distribution: na Organization: Philips Labs, Briarcliff Manor, NY Lines: 205 Now, why don't we expand this whole discussion so that others can take part, too? After all, no one else has made any contributions to this, and we might as well conduct this by mail, rather than over the net. So since the disagreement is one of underlying philosophy, let's argue on that ground. It may provoke more response. I make some relevant responses to your statistical epic later on, but the important stuff is this: You feel that by computing statistics based upon the raw numbers which are available in various publications, you can attain an understanding of the inner workings of the game, to the point that you can describe the strengths and weaknesses of players, discuss strategies, etc. I feel that this is not true, that instead, statistics lead to a superficial appreciation of correlation, but no definite understanding of cause-and-effect. This can be compared to the type of knowlegdge which current "Artificial Intelligence" systems employ. Such systems, e.g. MYCIN and its descendants, use associational knowledge to correlate stimuli and desired responses, e.g., "If symptom A and symptom B are present, then prescribe remedy X." This can be effective in many cases. But there is a great deal of knowledge which cannot be captured this way - the structural knowledge, i.e., WHY A and B tend to respond to X. Even adding probabilities to this knowledge does not change this limitation. In the same way, baseball stats can reveal many correlations, and can thus lead to a greater appreciation of the game and its intricacies. But the structural knowledge, e.g., "Is player A better defensively than player B?","Will player A contribute more to my team than B?", is subjective, and hence dependent upon the knowledge that a person has acquired about the game. Thus, the only people who are qualified to make these judgements are baseball professionals, not statisticians. Even though Davey Johnson uses a compendium of stats to help him manage the Mets, HE still makes the decisions, and can easily decide to ignore a stat. (After all, he initially decides which stats to put into his machine, and which to leave out.) Also, these baseball pros have access to three sources of info that you and I do not: 1) They keep charts of every pitch, and every hit and play in the field. Thus, they have MUCH better data to rate the players; 2) They see many more games, and from a better vantage point than we do (they have access to the tapes of the games, too, so they miss nothing that we see); 3) They have the accumulated experience from their careers, which enables them to interpret what they see in ways that can be very different from the way we see things. It is precisely this lack of understanding that prompts people like us to revert to statistics. (Yes, I used to love stats, too. I really used to get off on dreaming up new measures, and reading all the numbers I could, until I became convinced of the lack of content of the stats.) So, I leave the decisions to the pros, and hope that my hometown pros make good decisions. The pros said last year that Pena was better then Carter defensively. There was no outcry. Frankly, I was surprised that Pena won it, and expected an outcry. But, when there was none, I realized that Pena might actually have passed Carter defensively. My amateur perceptions confirm this. Your stats fail to convince me, for the reasons above. Send them to the people who vote for these awards, or else it seems to me that you don't really believe them yourself (or do you really think you know more than they all do?) (I will be away to conferences and vacation for two weeks. Postings during this period may not be saved by my site.) ------------------------------------------------------------------ Responses to statistical verbiage: > Also, we really don't want to consider years > before the principles established themselves in their respective teams > starting lineups, so we consider Carter in 1975 and 1977-1984 (in 1975, > Carter spent most of his playing time in the outfield (as the starting > left fielder, generally) while serving as the #2 catcher; in 1976, he > was just the back-up catcher) and Pena from 1981-1984. Why consider any years before 1984? The original question was which should start the 1985 all-star game, not which was better in 1975. I knew Carter was better than Pena before 1984 even without reading any stats :-) > It is widely recognized that the purpose of the offense is run > production, and there are two distinct ways in which a hitter may > contribute to it. The first is to score runs, the second to drive > them in. Thus, traditionally, fans have placed great store in the > most obvious measures of that production, runs scored and runs batted > in. Unfortunately, those traditional measures are heavily dependent > on circumstances beyond the hitter's control: how well his teammates > fare in doing THEIR job. You can't score if no one drives you in, and > you can't drive some one in if no one is on base. If we are to > evaluate individual performance, we must look at statistics that are > NOT dependent on the action of anyone save the individual in question. EXACTLY!!! If you can find a statistic that is truly independent of teammate's contributions, I'd love to see it. All the stats you list below (Putouts, %thrown out, DP's, BA, OBA, HR, R, RBI, slugging, etc.) are dependent on: teammates' seasons, manager's tactics, place in the lineup, ballparks, and others. Apparently one of your favorite stats for pitchers is Earned Runs Prevented. You love to post a detailed list to the net every so often. This stat is no more independent than the old ERA. For example, if a pitcher pitches for a team which scores fewer runs for him, then he may be lifted earlier, on the average. This leads to fewer innings for him, and a lower score on ERP. This effect can also be achieved bu playing for a manager who loves to go to the bullpen early (e.g. Chuck Tanner.) These stats can be revealing at times (Gooden leads by a large margin no matter how you measure things) but using them to make finer distinctions is meaningless. > Thus, we look at on On-Base Percentage (a.k.a. > Average) to evaluate how well a player performs this function. You really love this stat. Fine. This is a free country. I think it is just another meaningless stat. Hits are better than walks any day. Since you love to compute, why not analyze how often runs are scored with only walks, versus how often runs are scored with only hits? Or how often runs are scored with no hits at all, versus how often runs are scored without walks? This is baseball, not the on-base derby. >(1) As we are limiting ourselves to statistics upon which the > performance of one's teammates has no direct bearing (OB > and SA), and for which there is no empirical evidence of a > substantial indirect bearing, it is irrelevant. As stated above, this is an unproven assumption. >(2) Pittsburgh was not a substantially less capable offensive > team than Montreal. Pittsburgh's worst year was 1984 We are voting for the 1985 starting all-star catcher, remember? > Both Carter and Pena batted in the middle of their respective orders > (generally fifth for Carter, sixth for Pena), and probably have about > 45 such opportunities in a season. Assuming that the opportunities > are uniformly distributed among out counts, 30 of these occurred with > none or one out (actually, this probably overestimates the number of > such opportunities, as outs accumulate as batters bat, thus implying > that more runners are on base, on average, with two out than with none > out). Pena has good speed for a catcher, average for all runners, and > would probably advance to third about 33% of the time (choosing the > median value from Texas regulars); for Carter, my best guess is 20% > (he's not as hopeless on the bases as Sundberg). The difference, > then, is probably about 30/3 - 30/5 = 4. It does not make up for > Pena's negative contribution in his stolen base attempts. By your argument, speed is a negligent factor in baseball, at least offensively. You apparently feel that breaking up double-plays at second base, or avoiding them at first, or taking the extra base, or causing an errant throw, are small factors. You like HRs and walks. Well, why not take up this argument some time with a professional baseball person (which neither of us is) and tell him that speed is negligable offensively? From everything I have heard and read, this is not so. Again, we are fans. I trust what I hear from pros more than your amateur judgement (or my own). FIELDING > As always, I advocate ignoring the number of errors and the fielding > percentage, as lending substantial credence to those figures favors > the sure handed man who doesn't cover much ground. Total chances may > be ignored, as we will treat assists and put outs separately. Double > plays should also be ignored, as they are more a function of > opportunity (pitchers who tend to get grounders, pitchers who tend to > load bases) than of skill. Although I tend to agree with you here, can't you see that these assumptions are subjective? How can you prove that the statistics you favor are the "right" ones, and that the others can be ignored? > If we assume that those strikeouts were as likely > to occur at one time as another, > Catcher PO Est. K Est. (PO-K) > Carter 772 .83*861 = 715 57 > Pena 895 .86*992 = 853 42 Why make this assumption? It is quite possible that the Ks are not exactly uniformly distributed. For example, it is possible that Tanner used Pena's backup only with experienced pitchers. Now, two of the three main K pitchers on Pitt in 84 were experienced, but only one of the remaining two starters was a big K pitcher. This would invalidate your assumption. Are you going to analyze all 162 Pitt and 162 Montreal games? This is especially true since your final numbers (57 and 42) are so small relative to the size of the Ks. A few here and there could change things drastically. > Thus we find that, in 1984, Carter was more successful in hunting down > foul balls and getting putouts at the plate in somewhat less time catching. Awfully unstable conclusion. Particularly in view of your own statement that Montreal's park is larger than Pitt's. Even 10 or so more foul pops caught by Carter change the stats significantly. > The results: inconclusive, but not consistent with Paul's > claims of clear Pirate supremacy. These are not just my claims. As I have said before, and will say again and AGAIN, argue this with the voters for the Gold Glove. They have more knowledge about the players and their abilities, not just numbers. HOW DO YOU KNOW THAT YOU EVEN HAVE THE RIGHT RAW DATA WITH WHICH TO COMPUTE STATISTICS? Couldn't some raw numbers be not even available to you?