Gould, Trends as Changes in Variance: A New Slant on Progress and Directionality in Evolution

326
JOURNAL OF PALEONTOLOGY, V. 62, NO. 3, 1988

an example from my own research, not on fossils but on baseball. (I don't mean to be either facetious or flaky; neither, to use a metaphor appropriate to the subject, am I grandstanding. Baseball is a remarkable and unmatched source for the study of trends, for it offers nearly 100 years of fully comprehensive data on a system operating with no major changes of rules or dimensions. The last major alteration, the retreat of the pitcher's mound to its current distance of 60 feet 6 inches from home plate, occurred in 1893.)

The disappearance of .400 hitting is the most widely discussed and puzzling trend in the history of baseball. Fans lament this supposedly lost excellence, and I doubt that any other "decrease trend" has received so much discussion and speculation in our culture. The facts are impressive: only one player has exceeded .400 in the last half century (Ted Williams at .406 in 1941), but eight players hit higher than .410 between 1890 and 1930. Where did this excellence go?

Traditional explanations are legion, but all rely upon a mistake in categories and a confusion between movement of an entity and reduction in variance of a system. The traditional arguments range from the nostalgic and sentimental ("they don't make 'em as tough as they usta"), to the analytic (more grueling schedules, too many night games, bigger gloves for fielders, invention of the slider, introduction of relief pitching specialists), but all share the fallacious assumption that .400 hitting is a "thing"--an entity that used to exist, has since disappeared, and therefore has moved away. But if we view .400 hitting as an extreme value in a coherent system of variation, then we have the key to a correct understanding.

The simplest hypothesis would hold that batting averages, in general, have declined, bringing down the extremes in this general fall. But this easiest potential explanation is false. Not only have league averages for full-time players remained approximately constant (at roughly .260) for more than 100 years, but baseball's guardians have always intervened to change the subtle balance between pitchers and hitters, and to restore the .260 equilibrium, whenever any trend or innovation gave temporary advantage to one side or the other (decreasing the strike zone and lowering the mound when pitching became dominant in the mid 1960's; adopting the foul strike rule in the 1890's when batting improved drastically after the pitcher's mound moved back to its current distance). The disappearance of .400 hitting has therefore occurred within a system that maintains an unvarying mean.

This situation led me, several years ago (Gould, 1983, 1986), to conjecture that the disappearance of .400 hitting might best be viewed as a consequence of decreasing variance about this unchanging mean. Since the reasons for declining variance are so different from the causes for removal of an entity, I realized that such a finding would revise interpretations of this trend. In the immodesty of a "hot idea," I did expect to find such a symmetrical decline of variance in both tails of the distribution for batting averages. But I was not prepared for the uncanny, exceptionless regularity of the data.

Figure 7 shows the symmetrical shrinkage of extreme values for both low and high batting averages for means of the five highest and five lowest averages in each season (by decade). Figure 8 is the full and proper calculation of standard deviations for all regular players (defined as two at-bats or more per game, giving N between 100 and 300 for each season) year by year from the foundation of the National League in 1876 to 1980. The steadily declining variance is uncanny in its regularity (and cannot be attributed to such artifacts as short seasons, and therefore fewer at bats with greater variance, early in baseball's history--for seasons of more than 100 games were already established in the 19th century). For example: all four beginning years of the 1870's exceed 0.5 in standard deviation, but the last value larger than 0.5 occurs in 1886. Other 19th century values are all 0.4--0.5 (with three just below at 0.38--0.40), while the last reading in excess of 0.4 occurred in 1911. Even small details of later decline in the 0.3--0.4 range show regularity: the last value as high as 0.37 occurred in 1937; 0.35 was exceeded for the last time in 1941. Since 1957, only two years have topped 0.34. Between 1942 and 1980 all values remain in the restricted range of 0.285--0.348. All standard deviations from 1906 back to the beginning of major-league baseball in 1876 are higher than every value from 1938 on. There is no overlap at all.

I conclude, therefore, that .400 hitting has disappeared as an automatic consequence of symmetrically shrinking variation around a constant mean. This new depiction of an old observation implies a reversed interpretation as well. The old explanations wept and wailed, because they assumed that something precious had been lost--the obvious interpretation for removal of an entity. But I hold that the trend reflects increasing general excellence of play--and that symmetrically shrinking variance should occur in systems that stabilize as they improve. Pitching and hitting have both become substantially better as training of athletes intensifies, and as opportunities open for players of all races and nations. But the balance between hitting and pitching has been maintained as both improve--and we define that balance by the unchanged mean batting average of .260. As everyone gets better, the discrepancy between average and best must decrease (leading to the disappearance of .400 hitters), while poor batters once tolerated for excellence in fielding no longer make the grade as the pool of players who can both hit and field grows (leading to shrinkage of the left tail). The game has become more precise and unfailingly correct in execution -- as 100 years of trial and error distill the optimal procedures in all situations of fielding, hitting, and pitching. The best can no longer take advantage of sloppiness in a young system still regulating its subparts. Wee Willie Keeler could "hit 'em where they ain't," and bat .432 in 1897, because fielders didn't yet know where they should be. Now every pitch and every hit is charted; the weaknesses and propensities of every batter are assessed in detail. Boggs and Carew were surely better hitters than Keeler, but neither has reached .400 in modem baseball. Increasing general excellence of play has eliminated .400 hitting, but we must first picture the phenomenon as a symmetrical reduction in variance before we can grasp this explanation.

Potentially random plucking.--Anagenetic bias also lies behind our conventional (and probably incorrect) interpretation of what may be the most important of all trends mediated by reduction in variance--patterns of removal and replacement surrounding major phases of extinction. Consider our two modes of diversity and disparity, both properly described as reduction in variance among entities. First, mass extinction reduces variance in an entire system because large-scale episodes completely eliminate many clades, and therefore reduce the number of Baupläne available for recruitment of new taxa. Second, consider the extinction of the 20-or-so Burgess Shale phylum-level Baupläne (Whittington, 1985; Briggs. and Conway Morris, 1986). We have not resolved the actual pattern in this case, because we cannot trace the Burgess creatures through later soft-bodied faunas, and cannot tell whether Burgess "oddballs" succumbed by droves in mass extinctions, or petered out. But suppose that the latter alternative holds, and that diversity remained constant while disparity declined as the likes of Opabinia, Hallucigenia, and Anomalocaris disappeared.

In either case--provided that diversity is reestablished at or above former levels following the episode of extinction--traditional interpretation has favored the anagenetic mode of survival and change under competition. That is, we have assumed

148