3 ms·

> Because of slowly but steadily improving equipment, training, nutrition, medical treatment. The simplest predictor is "how many people practice a sport?" In

> Because of slowly but steadily improving equipment, training, nutrition, medical treatment. The simplest predictor is "how many people practice a sport?" Insofar as fitness for a particular athletic trial follows the central limit theorem, larger samples will open the door to more spectacular outliers. The relationship between total participants and the time it takes for a record to fall shows up in sport after sport. The other factor is: new PEDs that are forbidden but difficult to detect at a given point in time. These affect some sports more than others. Periodically records in these sports are "reset" through administrative twiddling. For example, every 20 years or so Olympic Weightlifting weight classes get changed, meaning all the old records are essentially obsoleted.

majormajor 6y ago
Even if the number of new entrants to a sport per year stay constant, I think you'd expect records to continually fall as the cumulative sample over time gets larger and larger. That said, yeah, them falling more rapidly is also helped by the training/equipment/nutrition/drugs/etc improvements.
- jacques_chester 6y ago
  The training, equipment etc improvements themselves follow population. More natural and designed experiments, more alternatives available, more accumulated know-how and theory. But any given combination of factors is itself part of the athletic trial. The Soviet Russian system of weightlifting carefully husbanded the outstanding weightlifters they had to work with and so selected for athletes who responded well to that training. The Bulgarian system was based on masses of warm bodies, enormous drug regimens and ruthless elimination upon injury. It selected for athletes who responded well to drugs and weren't injury prone. Afficionados of the sport argue constantly about which is "better", since both turned out world record upon world record. But such a debate was beside the point: an athletic system selects whatever it is set up to select for. The rest of the effect is then down to numbers.
- gwern 6y ago
  Yes, you'd expect records to always keep falling given a fixed distribution, but it will slow down fast. The expected maximum of _n_ random samples is roughly log(n) in size (formulas: https://www.gwern.net/Order-statistics https://www.gwern.net/Order-statistics ), so you need to increase dramatically to noticeably increase the maximum. Or to put it another way, since a record is defined as 'beating all previous samples', the chance of the next sample setting a record is always 1/n. So if you have m entrants per year adding to that total n, the probability is always positive, but it quickly becomes very small.

majormajor 6y ago

Even if the number of new entrants to a sport per year stay constant, I think you'd expect records to continually fall as the cumulative sample over time gets larger and larger. That said, yeah, them falling more rapidly is also helped by the training/equipment/nutrition/drugs/etc improvements.

jacques_chester 6y ago
The training, equipment etc improvements themselves follow population. More natural and designed experiments, more alternatives available, more accumulated know-how and theory. But any given combination of factors is itself part of the athletic trial. The Soviet Russian system of weightlifting carefully husbanded the outstanding weightlifters they had to work with and so selected for athletes who responded well to that training. The Bulgarian system was based on masses of warm bodies, enormous drug regimens and ruthless elimination upon injury. It selected for athletes who responded well to drugs and weren't injury prone. Afficionados of the sport argue constantly about which is "better", since both turned out world record upon world record. But such a debate was beside the point: an athletic system selects whatever it is set up to select for. The rest of the effect is then down to numbers.
gwern 6y ago
Yes, you'd expect records to always keep falling given a fixed distribution, but it will slow down fast. The expected maximum of _n_ random samples is roughly log(n) in size (formulas: https://www.gwern.net/Order-statistics https://www.gwern.net/Order-statistics ), so you need to increase dramatically to noticeably increase the maximum. Or to put it another way, since a record is defined as 'beating all previous samples', the chance of the next sample setting a record is always 1/n. So if you have m entrants per year adding to that total n, the probability is always positive, but it quickly becomes very small.