#epidemiology time
Be wary of biases in how our brains interpret rates.
For example, a 100% increase (100 to 200) is undone by a 50% decrease (200 to 100).
I see this a lot in suicide reporting, where increases are breathlessly reported, but decreases are downplayed.
/1
Be wary of biases in how our brains interpret rates.
For example, a 100% increase (100 to 200) is undone by a 50% decrease (200 to 100).
I see this a lot in suicide reporting, where increases are breathlessly reported, but decreases are downplayed.
/1
Curious why that is?
The number set for decreases is 1 to 0. The number set for increases is 1 to ∞. So the increase "stretch" while decrease "compress"
100% up = 50% down
200% up = 66% down
300% up = 75% down
400% up = 80% down
500% up = 83% down
600% up = 85.7% down
/2
The number set for decreases is 1 to 0. The number set for increases is 1 to ∞. So the increase "stretch" while decrease "compress"
100% up = 50% down
200% up = 66% down
300% up = 75% down
400% up = 80% down
500% up = 83% down
600% up = 85.7% down
/2
So the next time you see "suicides up 8%!!!" remember that:
a) there is a moral panic such that increases be headlines
b) there is a numbers bias to increases over decreases
c) year-to-year comparisons are relatively useless without trend analysis and error measurement
/3
a) there is a moral panic such that increases be headlines
b) there is a numbers bias to increases over decreases
c) year-to-year comparisons are relatively useless without trend analysis and error measurement
/3
For example, here is the truth about California pediatric suicides during the pandemic (2020-2021). They are *up* by about 29% from 2019.
2019: 1.2 per 100,000
2020-21: 1.6 per 100,000
These numbers have been touted by media and people without experience in suicide #'s.
/4
2019: 1.2 per 100,000
2020-21: 1.6 per 100,000
These numbers have been touted by media and people without experience in suicide #'s.
/4
However, this requires context. First, trend analysis and error measurements tell us that 2020-2021 are not different than expected trends or variability in the pediatric suicide rate for California.
/5
/5
As well, that 29% increase has occurred multiple times in California, like in 2004 (+34%), 2009 (+26%), 2013 (+29%), 2015&2016 (+15% and +27%, for a combined whopping +45%)
/6
/6
One of the best ways around most of the biases, is instead of comparing year to year, get a baseline year (say, 1999 like in this graph), and calculate the odds ratio (in suicide, similar to relative risk) of each subsequent year, with error measurement, on a log scale.
/7
/7
With this representation (one of the best ways to visualize rate changes), we can see not only the long-term and short term trends, but we can also see the error measurement when rare occurrences, in this case something that happens 1.6 per 100,000 times.
/8
/8
So the next time media/politician/some twitter troll says "suicides were UP THIS YEAR!!!!", remember to ask the following:
a) what is the variability of that over time
b) what is the rate over time and trend
c) do I see an error measurement
a-b-c necessary to say what they say.
a) what is the variability of that over time
b) what is the rate over time and trend
c) do I see an error measurement
a-b-c necessary to say what they say.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
