Several news sources have discussed a new study published in Nature claiming that the maximum human life expectancy from now on is 115 years, and will not increase, through an analysis of existing data. The study found that the last generation of people who were born with a considerable improvement in longevity were born in the 1980’s, and the data has plateaued since then. The scientists took this shift as evidence for a larger study on humanity’s natural limit, even though the regression lines found were far from statistically significant.
The paper’s authors plotted the maximum reported age at death (MRAD) for 534 “supercentenarians”, people who have lived to be older than 110 years old, from the US, UK, Japan, and France between 1968 and 2006. They created linear regression lines for two subsections of the data, 1968–1994 and 1995–2006, and found that while the MRAD increased by .15 years per year on average before 1995, after 1995 it decreased by .28 years per year until 2006. However, the scientists only found a weak correlation (r=-.35) for the decreasing MRAD, with a p-value (P=.27) too high to be significant, which they do not explain in the paper. Upon analyzing another set of MRAD records, they found similar results; MRAD values increased before 1995 and decreased from 1995-2015. Although they include r- and p-values in the study, again the authors fail to explain that the correlation for 1995-2015 (r=-.36, P=.70) is very weak.
After observing that the human MRAD has levelled off at 114.9, the authors said nothing about their statistical shortcomings, instead citing other studies which tried to find an age limit and failed to do so, blaming the lack of data available at the time. The article ends with musings on a biological cause of fixed longevity:
“…what appears to be a ‘natural limit’ is
an inadvertent byproduct of fixed genetic programs for early life events,
such as development, growth and reproduction.” (
Overall, there seems to be a lack of actual evidence beyond a weak trend line that their is an upper limit to the human lifespan.
While a statistical approach to longevity is a promising concept, the execution of statistical analysis in this study was poor. While the data does seem to show a levelling off of ages at death, the regression lines which the paper claims show a decrease in MRAD values are not statistically significant. Since the scientists mainly used these models to come up with the “115 years” conjecture, I am not confident in the results of this study as presented.
After some great discussion across the web, I’d like to clarify a few points. Several people have brought up the idea that the scientists were observing a positive increase in MRAD (and a significant one at that) but stopped and suddenly observed a decrease, and although it’s not significant, it’s still something, right?
I see it differently. Take a look at the raw data points from the second database:
It seems to me like the involved scientists broke up the data at an arbitrary point to be able to see a decrease instead of a weaker positive correlation. Break the data at a different point, 1980 for example, and you can make two lines that show a positive correlation. Or don’t break it at all; it seems a regression line for the whole data would fit.
So did someone make the glaring oversight that a p-value of .70 isn’t significant? No, but the creation of the line itself appears to be flawed.