Book: Stephen Jay Gould's Full House

Last month we did some book triage, looking for books that we could bear to give away to save storage space, and ran across a copy of Stephen Jay Gould's Full House. Apparently my mother gave it to me for Christmas 25 years ago, it wasn't at the top of the pile of books to read on Christmas afternoon, and I somehow never got to it. So I'm reading it now.

Gould's target in this book is the idea of "progress" in evolution: the idea that evolution through natural selection produces successively "better", or at least larger and more complex, organisms, because large size and complexity (however you define that) are generally pro-survival traits. Even today, and more so in the 1990's when he wrote the book, the standard popular image of evolution is of a ladder from bacteria to algae to jellyfish to trilobites to reptiles to mammals to primates to humans. And we see the same within lineages: we've all seen the picture of the evolution of horses from the dog-sized Eohippus to modern horses, donkeys, and zebras, and of primates from lemurs to monkeys to apes to humans, all illustrating the trend towards greater size and greater "complexity" or "sophistication".

And Gould says, in a nutshell, there is no such trend -- or if it is, it indicates a statistical artifact, rather than any advantage inherent in large size or "complexity". (In particular, modern horses represent not a pinnacle of evolutionary perfection but the pitiful remnants of a failed evolutionary branch: 5-10 million years ago there were dozens of genera, in a wide variety of sizes, then they all died out except Equus, and then a million years ago Equus died out over 90% of its geographical range, spreading a few species again only with the help of human domestication.)

What we perceive as a trend appears for several reasons.

1. We tend to pay attention to the largest or "best" of any group. If we're looking at batting averages in Major League Baseball (as Gould does for about a third of the book, to illustrate his point), it's easy to look up the best averages in the league, but much harder to find the worst averages in the league, because nobody's interested in those. If, hypothetically, the mean batting average were to stay the same but the standard deviation increased or decreased, we would perceive the maximum increasing or decreasing, not because batters as a whole are getting better or worse, but because they are getting more varied or more uniform. When Gould analyzed not only the best but the worst batting averages in MLB from year to year, he found them moving symmetrically: on rare occasions, either or both might move away from the mean, but in most years, they both move towards the mean (precisely what one would expect from a sport as it matures and standardizes on better ways of doing things).
   
   But why would the mean be staying the same? Wouldn't you expect overall batting skill to get better over the decades? Yes, but pitching and fielding skill also get better over the decades. And whenever some technological or technical breakthrough has given an advantage to one over the other, the rules have been changed to "restore the balance"; as a result, the mean batting average in MLB has been remarkably constant for a hundred years.
   
   Gould concludes that the "extinction of the .400 hitter" (nobody's hit .400 in a season since, IIRC, 1941) indicates not a degradation in play, but rather of an improvement (and, more importantly, a uniformization) of play: as the bell curve of performance gets narrower, there are fewer and fewer examples at any given distance away from its peak.
   
   The reverse happens in biology: the real trend in evolution is increased variation among individuals of a species, among species, among genera, etc. (Until there's a mass-extinction event, at which point most of the groups disappear completely, and the survivors start over, increasing in variation again from their new starting point.) If we pay attention preferentially to the largest or "most complex" members of any group, as seems to be the human tendency, we'll perceive a trend towards increasing size or complexity when in fact there's only a trend towards increasing variability.
   



2. If a drunk is walking along a sidewalk with a wall on his left and a gutter on his right, randomly staggering to left or right, he'll eventually end up in the gutter -- not because his stagger is biased towards the gutter, but because when he's close to the wall, he can't move any closer to it; the only direction he can go is away from it. The "bell curve" of his possible paths is truncated on one side, so it's necessarily skewed and asymmetrical. Of course, how long it takes him to fall into the gutter is an exponentially decaying function of the distance from wall to gutter, so if it's a hundred yards, he may die of old age first.
   
   Similarly, there are probably physical limits to how good a human baseball player can be. The maximum fast-ball has been between 100 and 105 MPH for a hundred years; IIUC, nobody has ever thrown 110 MPH, much less 140. As a sport grows and matures, you expect any given performance measure to get steadily better, but as it approaches the physical limits, it will get better slower and slower. And those farthest away from the limits will get better faster than those closer to the limits. For example, when women were first allowed to run marathons with men, their times were much longer, but their rate of improvement from year to year was much faster -- exactly as one would expect because their sport was less mature and they were initially farther from their limits. As a result, one expects to see the best in any field with a hard upper limit get microscopically better while the worst get substantially better, thus drawing the curve tighter and narrower (while remaining skewed away from the wall).
   
   In biology, we know that life developed "from the bottom up", from simple organic molecules, to more complex organic molecules, to self-replicating molecules, to prokaryotic cells, to eukaryotic cells. This isn't necessarily because life "wants" to be more complex, but simply because for any given definition of "life", there's a lower bound of complexity below which something can't meet that definition. If organisms have the option of evolving greater and lesser complexity, some will probably do each, but the ones that fall below the lower bound will die (or at least no longer qualify as "life" for our purposes). So one would predict that a purely random stagger in biological complexity would produce gradually more complex (and exponentially rarer) organisms.
   
   One particularly revealing example involves foraminiferans, a group of microscopic organisms that (fortunately for biologists) produces silica shells that survive nicely in fossil records. The standard way to find them is to take a bunch of silt or sand and filter it through successively finer sieves, recording how many you find at each size. But the finest sieve available in most biology labs is about 150 μm, which means any foraminiferan below that size will never appear in a sieve and therefore in a biology paper. This serves as a "wall", or lower bound, on the size of foraminiferans, not in the wild but in our observations. And, just as one would expect, the data indicate that the maximum and mean size of foraminiferans gradually trend upwards over time, not necessarily because they're actually getting bigger, but because we're seeing the large end of the distribution and ignoring the small end (which may well be getting smaller as the biggest get bigger -- at least, there are still plenty of them near the 150 μm lower bound).
   

Fortunately, there are several ways to tell the difference between a measure increasing as a statistical artifact, and a measure increasing because of selection pressure.

1. You can look at not only the maximum but also the minimum: if maxima and minima move symmetrically around the mean, that suggests it's just a change in standard deviation, while if they both move in the same direction, that suggests an actual selection pressure.
   


2. You can look at the mode rather than the mean or the median. Means are notoriously susceptible to asymmetric skew: if my brother and I are in a room with Jeff Bezos, the average per capita income in the room is billions, but that doesn't tell you much about my brother or me. Medians are less susceptible to asymmetric skew, but still somewhat. It appears that the mode of vertebrate evolution (whether you count species, individuals, or biomass) is still fish; the mode of multicellular evolution is still arthropods (especially beetles); the mode of life's evolution is still bacteria. If you surveyed all the life on earth objectively, you could be forgiven for not noticing the existence of two tiny evolutionary branches called "plants" and "animals"; if you surveyed all the animals objectively, you might not notice the vertebrates; if you surveyed all the vertebrates, you might not notice the mammals; and if you surveyed all the mammals, you might not notice the existence of humans. At each of these levels of grouping, the overwhelming majority of individuals, species, and biomass are small and "simple", suggesting no evolutionary pressure towards greater size and complexity.
   


3. You can look at ancestor/descendant pairs in speciation events. If a "parent" species that's not up against a wall is equally likely to produce "child" species larger or smaller than itself, that suggests a random stagger; if, on the other hand, "child" species consistently tend to be larger than their immediate predecessor species, that suggests an evolutionary pressure for larger size. This is difficult research to do, as it requires identifying lots of ancestor/descendant pairs in an incomplete fossil record, but in most of the cases where it has been done (according to Gould, writing in the 1990's), the results have indicated a random stagger, not a consistent selection pressure. If anything, there's actually a tendency in the direction of simplicity, as formerly-independent organisms become parasites (or at best symbiotes) on other organisms and shed the now-redundant parts of their bodies and genomes.
   

Wild stuff. And Gould was such a good science writer.