Johnny Mostil

09-17-2005, 12:37 AM

. . . short, quick, and dirty, that is . . . and the short story is it doesn’t matter . . .

Given some of the posts I’ve seen about “momentum” recently, I wondered how “momentum” has affected teams in recent postseason play. I decided to test this with the results of the nine 162-game seasons that have featured three divisional winners and “wild cards” in postseason play from each league. That yields 72 observations (four teams in each of two leagues over nine seasons, or 4x2x9). I beg the indulgence of any sabermetricians among WSI’s 5,858 members (and counting!) for what I know is a flawed analysis (more on that below, but I’ll note here I haven’t done this type of work in a while, and have never before dabbled in sabermetrics specifically), and welcome suggestions and comments.

I decided to look at each playoff team’s record over its final 27 games (more on that number below as well) and the number of games it won in the postseason. If “momentum” is important, then we should expect teams winning more games in their final 27 to win more postseason games.

This doesn’t seem to be the case. I ran both a correlation and a linear equation and found statistically insignificant coefficients. In fact, the coefficients were negative—meaning teams winning more games in the final 27 actually lost more in the postseason—but, as noted, the numbers weren’t statistically different from zero. (And even if they were significant it still doesn’t mean it’s better to lose games at the end of the season, of course—just that there’s no evidence for the effects of “momentum” here.) I’d be glad to share the data and my calculations with anybody wanting to look at this further.

Non-statistical analysis of some teams also found no effects for “momentum.” Only once in the past nine years (2003) did a team having (or tying for) the most wins (among playoff teams in both leagues) in the final 27 games win the World Series. Only five times (Yankees in ’98, Braves in ’99, Yankees and Marlins in ’03, Red Sox in ’04) did a team having the most wins (among playoff teams) in the final 27 games in its league win the LCS. Seven of the seventy-two teams won at least 20 of their last 27; none won an LCS. Five of the 72 teams had losing records in their last 27 games; three of these (Braves in ’96, Florida in ’97, and Yankees in ’00) won an LCS, and two won the World Series.

Problems with the statistical analysis: (1) There are only 72 observations. (2) Correlation or, especially, linear regression, may not be appropriate for such data regardless of the number of observations, given there is more variation in the number of postseason wins than in the number of wins in the final 27 games. (3) The “samples” drawn for each team here aren’t independent. (E.g., these teams played each other in the final 27 games.) (4) Other variables may need to be included. (I did a quick google search and didn’t see any sabermetric analysis of season-end “momentum” and postseason play, but I didn’t look long.) (5) As noted, I haven’t done this type of thing in a while, so who knows what else I’m forgetting to check?

Why 27 games? Well, if I can figure out how to do this right, then it might be interesting to look at whether the timing of a team’s wins affect its post-season chances. I doubt that it does, but analyzing six segments of 27 games (6x27=162) over the full season would help pinpoint where, if anywhere, wins are more important. (E.g., how does a team whose wins are concentrated toward the beginning of the season compare with a team whose wins are more frequent at the end?) My guess—they’re all important, but there’s no reason to fret over them more at one time of the year than another.

Sleep soundly, Sox fans . . .

Given some of the posts I’ve seen about “momentum” recently, I wondered how “momentum” has affected teams in recent postseason play. I decided to test this with the results of the nine 162-game seasons that have featured three divisional winners and “wild cards” in postseason play from each league. That yields 72 observations (four teams in each of two leagues over nine seasons, or 4x2x9). I beg the indulgence of any sabermetricians among WSI’s 5,858 members (and counting!) for what I know is a flawed analysis (more on that below, but I’ll note here I haven’t done this type of work in a while, and have never before dabbled in sabermetrics specifically), and welcome suggestions and comments.

I decided to look at each playoff team’s record over its final 27 games (more on that number below as well) and the number of games it won in the postseason. If “momentum” is important, then we should expect teams winning more games in their final 27 to win more postseason games.

This doesn’t seem to be the case. I ran both a correlation and a linear equation and found statistically insignificant coefficients. In fact, the coefficients were negative—meaning teams winning more games in the final 27 actually lost more in the postseason—but, as noted, the numbers weren’t statistically different from zero. (And even if they were significant it still doesn’t mean it’s better to lose games at the end of the season, of course—just that there’s no evidence for the effects of “momentum” here.) I’d be glad to share the data and my calculations with anybody wanting to look at this further.

Non-statistical analysis of some teams also found no effects for “momentum.” Only once in the past nine years (2003) did a team having (or tying for) the most wins (among playoff teams in both leagues) in the final 27 games win the World Series. Only five times (Yankees in ’98, Braves in ’99, Yankees and Marlins in ’03, Red Sox in ’04) did a team having the most wins (among playoff teams) in the final 27 games in its league win the LCS. Seven of the seventy-two teams won at least 20 of their last 27; none won an LCS. Five of the 72 teams had losing records in their last 27 games; three of these (Braves in ’96, Florida in ’97, and Yankees in ’00) won an LCS, and two won the World Series.

Problems with the statistical analysis: (1) There are only 72 observations. (2) Correlation or, especially, linear regression, may not be appropriate for such data regardless of the number of observations, given there is more variation in the number of postseason wins than in the number of wins in the final 27 games. (3) The “samples” drawn for each team here aren’t independent. (E.g., these teams played each other in the final 27 games.) (4) Other variables may need to be included. (I did a quick google search and didn’t see any sabermetric analysis of season-end “momentum” and postseason play, but I didn’t look long.) (5) As noted, I haven’t done this type of thing in a while, so who knows what else I’m forgetting to check?

Why 27 games? Well, if I can figure out how to do this right, then it might be interesting to look at whether the timing of a team’s wins affect its post-season chances. I doubt that it does, but analyzing six segments of 27 games (6x27=162) over the full season would help pinpoint where, if anywhere, wins are more important. (E.g., how does a team whose wins are concentrated toward the beginning of the season compare with a team whose wins are more frequent at the end?) My guess—they’re all important, but there’s no reason to fret over them more at one time of the year than another.

Sleep soundly, Sox fans . . .