[simoncion]

> ... I was conflating Tokyo with Japan. ... These data for Tokyo are interesting:

> http://nbakki.hatenablog.com/entry/2014/02/20/124942

Those data for all of Japan (not Tokyo(!)) put the median commute time somewhere around 44 minutes, the max commute time at 90 minutes, and the average commute time at 57 minutes. That is totally in line with what I reported in my comment.

Did you maybe mean to link to [0]? If you did, then that survey tells us that of the 583 Tokyo respondents:

* 2.9% have a 120+ minute commute.

* 17% have a commute between 90 and 119 minutes.

* 66% have a commute shorter than 70 minutes.

* 20% have a commute between 60 and 69 minutes. This is the most frequent commute bucket. Second is 50->59 (14.9%). Third is 40->49 (13.89%). Fourth is 70->79 (13.03%).

> Really, it's not unusual to meet a salaryman with a 2-hour commute.

If only 3% of a population has a property, that property is actually pretty unusual.

> ...and this one for Japan paints a more vague but differently interesting picture:

From that page:

"About half of the Japanese respondents indicated that they need less than 30 minutes to go to work/school. On the other hand, one fourth of the respondents need more than one hour."

To break it down:

~50% < 30 minutes

~25% >= 30 minutes but <= 60 minutes

~25% > 60 minutes

That doesn't mesh with the official stats for the country.

> But you know, normal kinda also means "within 3 sigma" as well as "mean".

It's a pity that the standard deviation of the reported figures was not reported. I gather that it's hard to determine what is within 3 SD of the mean without that information.

[0] http://nbakki.hatenablog.com/entry/2014/08/05/231455

[ectoplasm]

I didn't link to the wrong thing. If you read that article the top cities excluding Nara are all in greater Tokyo. The max time for Kanagawa is not reported, the average time for Kanagawa is 90 minutes (for men with kids under 6). Your calculations are median / max / average of averages.

+/- 3 sigma = 99.7% of the sample, and we know 3% is > 2 hours (for all of Japan, not Tokyo). There's near the same percentage of LGBT adults in the US (~4%).

All we're arguing over is the use of the word normal. Would you be happier with "uncommon but not an extreme outlier"?

[simoncion]

> ...the top cities excluding Nara are all in greater Tokyo.

Oh. You meant "The Tokyo Metropolitan Area", rather than Tokyo. Gotcha. :)

> Your calculations are median / max / average of averages.

Mmhmm. Given that that's the only data we have to work with, I don't see the problem with it (other than the violence it should have done to the phrasing in my previous comment). I'd rather have the more detailed data, but alas.

> ...and we know 3% is > 2 hours (for all of Japan, not Tokyo).

How do we know that? The data that my 3% came from was -apparently- from Tokyo, Kanagawa, Chiba, or Saitama respondents.

> There's near the same percentage of LGBT adults in the US (~4%).

1) If that's L, G, B, and T adults in the US, I don't believe that number for a second. That stat has to suffer from under-reporting.

2) I would absolutely say that a property shared by only 4% of a population is not common.

> All we're arguing over is the use of the word normal. Would you be happier with "uncommon but not an extreme outlier"?

Based on the data, a 120+ minute commute appears to be rather uncommon. I'm uncomfortable about speaking about the nature of the outliers without knowing more about the individual points in the data set. That last bucket is potentially a very large one; who knows what its contents look like? [0]

"Normal" is... not the best term to use when trying to speak precisely.

If I have a system that only fails in a particular way 3% of the time, I could reasonably say "That failure mode is not normal.". On the other hand, if I know that it fails in that particular way 3% of the time, I can reasonably say, "Oh, that's infrequent, but normal behavior of the system.".

See the problem?

[0] I mean, obviously, we could have a few reasonably good guesses at its highest possible upper bound, but other than that...

[ectoplasm]

My 3% came from this article (which I linked, but which is fine to have ignored):

http://www.japan-guide.com/topic/0011.html

The LGBT stat came from Wikipedia. For controversial stuff like that in the US, WP is pretty good. I used to think it was 10-15%, 1 in 7 was the number I learned growing up, I guess it's 1 in 25.

https://en.wikipedia.org/wiki/LGBT_demographics_of_the_Unite...

For me, "abnormal" generally means "what is wrong with you" (the response given here), whereas "normal" can even include eye-rolling and "oh so you're one of those". The difference between 1 in 33 (3%) and 1 in 666 (0.015%, from the tail end past +3 sigma which accounts for 50% - 99.7% / 2) is really quite palpable.

But again, let's face it: it's not as common as I thought, and it's not as uncommon as you thought.

[simoncion]

> My 3% came from this article (which I linked, but which is fine to have ignored):

...really? From my reply:

> From that page:

> "About half of the Japanese respondents indicated that they need less than 30 minutes to go to work/school. On the other hand, one fourth of the respondents need more than one hour."

I read and commented on everything that you linked to. The only time 3% appears in that article is "3% take the motorcycle, while 7% indicated not to commute at all." The bar graph that is titled "International comparison of commuting times (one way, in minutes)" has no Y-axis label or grid lines, and is small and low-resolution so determining what percentage is represented by the 120+ minute bucket is tricky at best.

> For me, "abnormal" generally means "what is wrong with you" (the response given here)...

Folks feel that people who are spending four hours every weekday on the road are doing something strange and aberrant because that's an enormous amount of unpaid time to spend doing something required by work. It doesn't matter how much of a population does it, it's aberrant and -to a degree- self-destructive behavior. [0]

> For controversial stuff like [LGBT "membership"] in the US, WP is pretty good.

Sure. I'm making the argument that the studies are suffering from under-reporting. This is something that you have to ask others to disclose, rather than something you can test for.

> But again, let's face it: it's not as common as I thought, and it's not as uncommon as you thought.

No, the Japanese commute time stats sound about right to me. The numbers for Japan were more like what I expected the US numbers to look like, actually.

[0] As always, remember that if I were assigning blame, I would do so explicitly.