Friday, February 26, 2021

The CDC gets life expectancy wildly wrong

According to a CDC spokesman, U.S. life expectancy has fallen by a year as a result of Covid. A little arithmetic shows that that cannot be close to correct. 

Total deaths so far are about 500,000 out of a population of about 330,000,000. The average death cost 12 years of life. Multiply that out and the average person lost not one year but .018 year of life.That's an error of almost two orders of magnitude. Including deaths indirectly caused and additional deaths over the next few months might increase it a little, but there is no way it can be one year or even close.

Dr. Peter Bach explains the error on his blog. What the CDC apparently did was to calculate what the effect on life expectancy would be if mortality rates stayed at their 2020 level,  how much Covid would reduce life expectancy if the pandemic was repeated every year forever.

After an error of that magnitude, it is difficult to take seriously any factual claim they make. It will be interesting to see if they admit the error.

P.S. To be fair, I have not found the claim on the CDC web site, only in media reports that attribute it to "Robert Anderson, who oversees the numbers for the CDC." 

P.P.S. I have now found it on the CDC web site in the form of an interview with Elizabeth Arias, who apparently produced the number:
I posted a version of this to FaceBook. Some people defended the CDC on the grounds that the usual way of calculating life expectancy is by measuring the current mortality rate as a function of age and projecting it it into the future. The problem is that, as you can see by the interview, Elizabeth Arias makes no attempt to explain that what she is describing is a measure that she knows, in this case, badly misrepresents what it is supposed to be measuring. 


Anonymous said...

How do you feel about Richard Epstein's repeated errors in light of the attention among decision-makers given to it?

David Friedman said...

I am not aware of repeated errors by Richard Epstein. What are they?

Brandon Berg said...

This isn't really an error; it's the usual way life expectancy is calculated. It's not ideal, but what we really want is to use future mortality rates to calculate life expectancy at birth, and we can only speculate about what those will be.

This method can fail rather spectacularly when mortality rates are unusually high or low in one particular year (maybe next year life expectancy will jump up due to COVID-19 disproportionately culling the frail). On a few occasions, life expectancy in some countries has briefly dropped to single digits due to extraordinarily high mortality rates. But this is arguably the best we can do without making data up; any other method will have other drawbacks.

Another way this method fails is that it makes no attempt to account for future reductions in mortality rates. On average, we should expect babies born in 2019 to live at least a few years longer, and perhaps much longer, than the 2019 life expectancy.

It's best to think of life expectancy as a summary of age-specific mortality rates in a single year, rather than an actual prediction about how long people will live.

By the way, I heard secondhand that you had said that Cade Metz misrepresented what you said to him, but was not able to find where you had said that. I'm inclined to believe it, since it didn't sound like something you would have said; do you remember where you might have commented on this?

David Friedman said...


It is the usual calculation, but it gives a wildly wrong answer if you use mortality rates that you know are a poor predictor for future mortality rates, and the people doing the calculation have to know that.

"Life expectancy is a statistical measure of the average (see below) time an organism is expected to live, based on the year of its birth, its current age, and other demographic factors including gender." (Wikipedia)

There is an interview with Elizabeth Arias, who apparently did the calculation, on the CDC site, and nowhere does she explain that this is what life expectancy would be if we had another pandemic every year forever.

Metz did not misquote me. What he did was to insert the statement "The voices also included white supremacists and neo-fascists" between two quotes from me, in a way that makes it sound as though it is a paraphrase of something I said, which of course it isn't. I don't think it's even true — at least, I don't remember any posts that fit either description.

DinoNerd said...

Thank you. This was useful information, which I signal boosted with a tag about the innumeracy of the press.

Vpn خرید said...