Thomas Basbøll and I have discussed how we learn from the immutable aspect of stories: like that famous rock of Dr Johnson, they push back when you kick them.
This comes up with statistics examples too.
Yesterday I wrote a post on the Jesus principle of regression coefficients: if a predictor takes on values far from zero, then the intercept can be difficult to interpret. The name of the principle comes up because, if your predictor is “year,” then the intercept corresponds to the predicted outcome in the year Jesus was born.
The idea is simple enough, but I always like to give an example, so I searched my directories and found the series of world record times for the mile run. This led to a lively discussion in comments, with almost nothing about the subject of the post (“What does Jesus have to do with linear regression?”) but lots of interesting stuff on the mile run, for example this from Jerseg:
Just noting as a track fan that there’s a widespread belief that the fast times in the late 1990s were largely a function of doping, specifically with erythropoietin (EPO) and related drugs. There was little testing for the drugs back then, especially compared with testing levels today, although there still is some amount of doping in the sport (although doping with EPO is probably relatively rare these days, since it’s easy to test for). The major technological improvement in recent years has been in shoe technology, specifically the use of carbon-plated spikes and road-racing flats. Although the men’s outdoor 1500m/mile world records were set decades ago, there are more 1500m/mile performers under 3:30 each year than ever before (3:30 is a world class time), so depth has improved substantially (see: https://www.alltime-athletics.com/m_1500ok.htm). This depth is also reflected in the high school and college ranks. Furthermore, outdoor men’s 5000m, 10000m, half marathon, and marathon world records have also recently been set, as have all the women’s outdoor world records from the 1500-marathon (despite most of the previous women’s world records also thought to be a product of doping). It’s somewhat likely Jakob Ingebrigtsen or one of his competitors finally breaks the outdoor 1500m/mile record this year.
It’s fun when the comments go in an unexpected direction. This also shows a benefit of bringing in real examples–not just “real data” like some canned dataset in R or whatever, but a real example with real interest. God is in every leaf of every tree.
In general we should subtract off a number in the middle of the time series or other measurement which represents a typical value for that measurement in the data and divide by a typical value to make things dimensionless.
If youre looking at obesity among adult males between 1970 and 2025, then maybe if your variables are w for weight and d for date… You’d do
wbar = mean(w)
(w – wbar) /(0.1 * wbar) ~ (d-2000)/20
Now your intercept represents the mean weight in the year 2000 and your slope represents how many multiples of 10% of the mean weight are added on average every 20 years.
The expectation of the coefficient is that it should be in the range of around 1 and the choice of 20 years and 0.1 of the mean weight as scales are based on our background info. The coefficient being around 1 means the prior is easy to specify and that the sampler will need to take steps that are neither too tiny nor too large and its easy to initialize the sampler (much more important when you have 500 parameters)
For a full Bayesian version you can make the weight subtracted a parameter, and set the intercept to 0 in the formula. Then the posterior for wbar tells you the mean weight of men in the year 2000
Hypoxia-inducible factor prolyl hydroxylase inhibitors have similar effects to EPO, are oral, and there aren’t tests for them. The current performances in distance running and road cycling today are similar to those of the EPO era in the ‘naughties’ (ie the ‘90s). Such cynicism saddens me, but should likely be considered in any analysis of professional endurance athletes.
The most well known hypoxia-inducible factor prolyl hydroxylase inhibitor (HIF-prolyl-hydroxylase inhibitor) is Roxadustat (Evrenzo) and that one can be tested for. It was the drug that Simona Halep was banned for after a routine drug test (she was banned in March 2024 even though it was apparently an unintended ingestion from a contaminated supplement). So there is a test for that one.
Anyway having just enjoyed a fantastic Tour de France I’d rather not have to think about the possibility of doping! While the race was overshadowed by suspicion in the 2000s (Armstrong/Contador/Landis) and some of the 2010’s (Team Sky) there hasn’t been much of that in recent years. Perhaps I’ve been naive but have to say it was great watching the Tour last month without suspicions of doping.
There are several others for which there is no test at all.
The times up the big cols in the TdF are on par or faster than the 90’s and faster than the Armstrong era, and far faster than the Froome era (look how much faster Yates time up the Finestre in the Giro was than in Froome’s famous 2018 ride). Froome would be left in the dust lol. The speeds are comparable to Pantani and Riis and the like.
Is that right? (i.e. HIF-prolyl-hydroxylase inhibitor’s lacking tests). I would assume that any drug/treatment on the banned list would have a test associated with it – is that an incorrect assumption or are there other of these drugs not on the list?
Apparently the percentage of TdF riders violating anti-doping regs (testing positive, admitting doping or having been sanctioned at least once in their career) was 55% in 2000 and drifted downwards over the last approx quarter century so that it was 2.3% in 2024 [Statistica.com using data from cyclisme-dopage.com – can Google “share-of-tour-de-france-riders-convicted-of-doping” to find this].
So either doping levels have dropped significantly and TdF cycling is way cleaner than before or doping is so sophisticated that it’s largely beyond detection! I can’t answer that though my perception is that it’s now cleaner, tho maybe some mixture of both.
There are some reasonable (non-druggy) reasons why TdF race speeds have drifted up again to surpass the 2000’s and one thing that was noticable in this years race with a ridiculously fast average speed, was the style of racing – the commentators repeatedly said that many of the stages were being ridden as if they were one day classics.
No that is not correct. They put methods and drugs on the list for which there is not yet a test. They are working on these but they are quite structurally diverse.
Certainly it seems to have cleaned a bit prior to 2020, but this was also reflected as slower speeds on the big climbs compared to the 90s or 00s.
Yes, there is much talk of more carbs, aero equipment, heat training and the like, but none of these things will transform a rider into doing almost 7w/kg for 45min in week two or three of a grand tour on the final col of a big mountain stage. The effect of aero equipment is minimal at the lower speeds on the big cols.
Cycling is unique in that we have data points from known dopers, and these performances are once again being matched after a couple of decades. These athletes were highly trained, had lightweight bikes by the mid to late 90s, trained at altitude, and were very talented + doping. That should raise some eyebrows. Froome and others took a lot of heat for performances that were far less impressive, following the fallout of the doping scandals in the 00s. It’s odd that today everyone is so credulous.
OK thanks, let’s see how next year’s Tour pans out…
This line caught my eye
“It’s somewhat likely Jakob Ingebrigtsen or one of his competitors finally breaks the outdoor 1500m/mile record this year.”
because I have lived in Norway a couple of times and “Ingebrigtsen” is the name of a well-known store in Minneapolis. The Wikipedia article about Jakob Ingebrigtsen
https://en.wikipedia.org/wiki/Jakob_Ingebrigtsen
refers to this chilling domestic abuse just concluded trial of his father, Gjert Ingebrigtsen
https://en.wikipedia.org/wiki/Ingebrigtsen_family#Domestic_abuse_trial
“The trial concluded on [2025] 15 May.[7] The verdict was announced on 16 June; Gjert was convicted of abuse against his daughter Ingrid, but acquitted of the charges against Jakob.[8] In the verdict, district court judge Arlid Dommersnes wrote that “even though Jakob and his brothers and spouses have given credible statements, the total weight of the statements and other evidence reviewed means that the court must conclude that there is reasonable doubt about the defendant’s guilt.”[28] Gjert was sentenced to a 15-day suspended prison sentence and pay NOK 10,000 ($1,010) in damages.[8][28]”
Yes, it has been (too) heavily covered in national media here in Norway. The family has been divided in their statements, and both sides of the family has stuck to their stories, I think. It’s all weird and sad – they’ve even had their own reality/documentary tv show, which was quite popular.
Whatever the truth, training for world champion level athletics and being a family may not combine so well. As has been seen with musical careers and so on as well.
Back when I was still working at PhRMA, Major League Baseball invited me to participate in a symposium on the use of human growth hormone as a performance enhancement drug. It was a fun experience, all expenses paid and two nights at the Beverly Hills Hotel (way above my expense account😊). got to meet the head of the players’ union and a couple of players who were team reps.
There was a professor from University of Southampton who gave a nice presentation discussing what is and isn’t known. If there are effects they may be modest and not nearly of the magnitude of anabolic steroids. Papers from the meeting were published in Drug Testing and Analysis volume 1 Issue #9-10 (2009)