3 ms·
The underlying data here is, of course, a distribution (each with its own distinct shape). In this case, I suspect that the "average" statistic is not the best
by gen220 6y ago
The underlying data here is, of course, a distribution (each with its own distinct shape). In this case, I suspect that the "average" statistic is not the best representative of that distribution.
For instance, "children" is definitely a bimodal distribution. There are many people in life who never have children, and therefore drag down the average for all those who do have children.
I suspect that, in reality, there are many modes for each of these dimensions, at each point in time. I think seeing the modes would be more interesting than the averages.
If you believe in free will, you can probably choose which mode you'll end up closest to. You certainly aren't doomed to be in some statistically-unlikely valley in-between. There's hope you won't die alone, in spite of what the picture painted by the average might say.
- throwaway2245 6y agoYes - I also didn't found the averages hard to interpret and not super full of insight. Presumably a lot of people spend 6-10 hours with their coworkers, and a lot of people spend 0 hours; a lot of people spend 8+ hours with their children, and a lot of people spend 0 hours, and pretty much the same for all of those categories. I'd have found a few 'typical' people in each age group more interesting. This is harder to calculate but maybe k-means clustering would be a good solution.
- bcherny 6y agoThis is a great point. Can anyone find a link to the raw dataset? Happy to whip up a better visualization.
- gen220 6y agoLooks like the data source is here: https://www.bls.gov/tus/datafiles-0319.htm https://www.bls.gov/tus/datafiles-0319.htm It's not a super easy to use format, but it should be fun to extract some more useful data out of it. For what it's worth, the sample size here is very small. It is hard to believe that it's representative of the distribution. This link [1] suggests the data come from 210k interviews conducted over 16 years (13ish thousand per year). And the data is longitudinal (so over those 16 years, many of the interviewees are the same person). Even if every interviewee was unique, this would represent <0.0001 (or less than .01 percent) of the ~240m americans in 2003 who were over the age of 20. [1]: https://www.bls.gov/tus/atussummary.pdf https://www.bls.gov/tus/atussummary.pdf