Take a look at these articles from various media outlets:
– People over 6ft have double the risk of coronavirus, study suggests
– Does being tall raise the risk of getting Covid-19? Men over 6ft are TWICE as likely to get infected, study claims
– How your HEIGHT could double your risk of catching coronavirus, scientists discover
– People who are over 6ft tall are twice as likely to catch coronavirus, study claims
– Being over six foot tall doubles Covid-19 risk
– Studie: Flere høje mennesker fik coronavirus (for my Danish readers)
Well, a new study claims that people (or men?) over 6ft (183 cm) tall are more likely to catch coronavirus. That’s interesting, I guess. I can see how tall people (or tall men?) face different environments that might make them more likely to catch the virus. Maybe. I am over 6ft tall and maybe I am twice as likely to get infected than my counterfactual shorter me. However, as I noticed in a previous post, there are profound reason to be skeptical towards a lot of social science research conducted in relation to COVID-19.
Accordingly, I decided to look at the working paper that found the specific relationship between height and COVID-19 risk. It is available here. Alas, it didn’t take me many minutes to see that there are significant problems with the manuscript in question.
Take a look at the first paragraph in the abstract: “The paper provides new evidence from a survey of 2000 individuals in the US and UK related to predictors of Covid-19 transmission.” That’s impressive, yes? No. Among the 2000 individuals, only a small proportion had the virus in both countries (and all the data is self-reported!). For example, only 43 men in the UK and 26 men in the US reported having had the virus. Those are rookie numbers in this racket. There is simple no to little data to actually explore variation in whether men had the virus or not.
For the actual results, take a look at this paragraph (also from the abstract): “Thirdly and finally, there is some, often weaker, evidence that income, car-owership [sic.], use of a shared kitchen, university degree type, risk-aversion, extraversion and height are predictors of transmission.” (my emphasis). Height is listed as a predictor among a series of other explanatory variables (and presented as ‘some, often weaker’ evidence). The problem here is the good old kitchen sink approach to social science. Throw in a series of variables in your model and see what shows up significant. It goes without saying that this is not good science.
And now take a look at this paragraph (also from the abstract): “The evidence about height is discussed in the context of the aerosol transmission debate.” Of course it is. This can explain why the height finding is receiving so much media coverage. We can think about a sexy theoretical mechanism explaining why tall people are more likely to get COVID-19. Here is the “mechanism” as outlined in the paper: “The fact that height is a significant predictor for men suggests that downward droplet transmission may be less important than aerosol transmission (particularly prior to lockdown) in which case the use of specifically designed air purifiers should be further explored.”
Noteworthy, the problem here is not that the work is not peer-reviewed. A lot of good research is not peer-reviewed and a lot of bad research is peer-reviewed. Also, I don’t blame the media for not picking up on any of the significant issues with the paper. The researchers doing the work are affiliated with institutions such as LSE, Oxford, Columbia and UCL, and I totally understand why journalists find it newsworthy when such “experts” conduct interesting “research”.
With that being said, let’s dig a little deeper into the paper. The first thing to keep track of is that the study is only making a point about the risk of being tall as a man. The problem here is potential p-hacking, i.e. that the researchers tested the relationship for the full sample – and then for men and women and decided that it was most interesting to go with the findings for men (the research was – from what I can see – not pre-registered). However, the choice is sensible if you take the measure of height into account. Normally, surveys interested in the height of the respondent asks for a precise height (e.g. in cm). That’s not the case here. The researchers use a weird binary measure of height and simply ask “Are you TALLER than 6 ft (1.83 m) in height?”. This is not the best way to explore how height relates to anything (including the risk of COVID-19 transmission).
We can begin to see how severe the problems are. Here is a simple cross-tabulation between height and transmission of COVID-19 (medical diagnosis/positive test for) among men in the US. We only have 11 men in the US sample that are tall (i.e. above 183 cm) and had the virus (again, according to the self-reported data). I don’t see any possible way the researchers can use this data to make any robust conclusions about how height matter for COVID-19 transmission.
The small cell sizes show that it is limited how much we can actually learn from any statistical analysis of the data, especially if we do multiple regressions. For the record, if we run a simple bivariate OLS regression, there is a significant effect of being tall on COVID-19 transmission in the UK (p = .04) but not in the US (p = .13). Again, these are results from bivariate regressions and we do not have sufficient data to do anything more advanced. Unsurprisingly, these results do not get any better if we begin to control for obvious confounders such as age (UK: p = .09; US: p = .15). In other words, the results are not convincing.
So, does being tall raise the risk of getting COVID-19? Maybe, but there is no evidence for this in the working paper. A working paper that, alas, led to news articles that are now shared thousands of times.