With the 2012 Olympics in full swing in London, it’s a good time for statisticians to reveal the power and glory of their discipline. For example, given the way Olympic records have fallen in the past, how many are likely to fall at these games? 

Today, Elliott Hollield at the University of North Carolina at Asheville and a couple of pals publish a fascinating analysis of this question. In short, they make predictions for records in 51 events and say that between 20 and 31 of these will topple.   

More interesting and subtle is their analysis of the factors that determine how long records last. 

As with many statistical questions, this one is deceptively tricky. Hollield and co collected data on 63 events going back to 1896. In this time, a total of 693 records were broken. It’s then straightforward to plot a time series for each record showing how long each one lasted. 

But why do some records last longer than others? To provide insight into this question, Hollield and pals looked at the influence of various factors on how long a record lasted using various different statistical approaches. 

In total, these guys studied 17 so-called covariates including age and gender, whether the record setter was from the host country, whether the setter was already an Olympic medalist, the growth rate of the GDP per capita of the setter’s home country and so on. The results make for interesting reading. 

First, the factors that have little or no influence. Gender makes no difference; men’s records do not tend to last longer than women’s or vice versa. Neither does age. 

More surprisingly, the host country does not have a significant influence wither. It’s easy to imagine that home support might spur athletes to greater heights making them more likely to break Olympic records. Not so, say Hollield and co. 

It’s also easy to imagine that a country’s wealth or a change in population size can make a difference but they don’t. 

So what does make a difference? It turns out that if the current holder also set the record in the past, the record is more likely to be broken at the next games. 

If the current Olympic record is also the world record, it is less likely to be broken in the next games. 

A change in the number of countries competing in an event  is also an important indicator of whether the record will fall. 

And most surprising of all, the percentage by which the existing record improved on the first Olympic record, is also a significant indicator  

That’s interesting stuff. However, Hollield and co could have gone further. There are some 300 Olympic events but they don’t bother to mention which 63 events they analysed.  

Neither do they have any fun with their results. They could, for example, have pointed to events where the Olympic record looks most vulnerable. But instead, they keep their heads firmly in their shells.

No doubt, these guys will argue that they are unable to make statistically significant predictions for specific events.

On the other hand, it is possible to publish results, even if they are statistically insignificant, and let the rest of us have some fun with them. There’s no harm in clearly labelled speculation, right? 

Ref: arxiv.org/abs/1207.6133: A Survival Analysis of the Duration of Olympic Records