Shortly after praising a news release by The BMJ earlier today for emphasizing the limitations of an observational study, another news release for another journal published by BMJ is at the other end of the spectrum.
“Fatherhood at young age linked to greater likelihood of mid-life death,” is the headline of a news release about a study in the Journal of Epidemiology & Community Health.
It is another big observational study, for which the published conclusion is: “The findings suggest a causal effect of young fatherhood on mortality and highlight the need to support young fathers in their family life to improve health behaviours and health.”
ScienceMediaCentre.org published this reaction from a professor of Applied Statistics:
“One problem is that it is an observational study and it’s always hard to work out what is causing what from observational studies. The authors are careful in their wording, and don’t go further than saying that the findings suggest the association between young fatherhood and midlife mortality is likely to be causal. The basic problem is that we can’t be sure whether it is the early fatherhood causing the increased mortality, in that if the young fathers had started their families later in life but nothing before that changed, then fewer of them would have died in middle age. Maybe there is something else that caused them to have children when they were young, and independently caused more of them to die in middle age.”
I don’t think the researchers were so careful in their wording – “suggest a causal effect”? What does that mean?
And the wording of the news release was also not so careful, allowing a researcher quote to be published, unchallenged:
“The findings of our study suggest that the association between young fatherhood and mid life mortality is likely to be causal.”
Again, what does “likely to be causal” mean?
And, whereas in the earlier example of study authors being clear about the limits of observational data – and then the journal news release re-emphasizing that – quality news stories picked up on the clues, in this case just the opposite happened.
The published manuscript used this confusing language suggesting a causal effect, which was echoed by the news release, which was echoed or at least inadequately addressed in much news coverage. Examples:
So – two different studies and two different news releases treated very differently by the same organization on the same day – with predictably quite different results in ensuing news coverage.
Addendum: On Twitter, Dr. C. Michael Gibson, Harvard prof and founder of Wikidoc.org, referred to all of this with the hashtag #LOSER, standing for Limited Observational Study Exercise Restraint.
— C. Michael Gibson MD (@CMichaelGibson) August 5, 2015