Another way to confirm this is to compare with the morning rush hour. Way more crowded as most people start work at around the same time (most companies have a starting time of 9 or 10 am).
I can confirm. I used to live in Kawasaki in 2010 and I regularly visit Japan from California for vacation and business. The morning rush hour is relatively short (roughly 7am to past 9am) but intensely congested due to the fact that most companies start work around the same time, and lateness is verboten. However, the evening commute is not as sharply congested, although it can still be standing room-only on the trains. The evening commute begins as early as 5pm but could last until the last train, typically around midnight. This is not only due to people working overtime, but also due to coworkers and colleagues going out at night for drinking and other activities. This was a regular occurrence for my colleagues and me back when I was interning at a Japanese company a decade ago in Kawasaki.
One additional challenge that will need to be addressed in Japan is unreported overtime (サービス残業). Often my friends, upon preparing for a day off, will work about 8 hours of overtime beforehand and not report it on the time and attendance systems at their companies. From my understanding, unreported overtime is not always reflected in most statistics; even when self reporting I bet many people forget all the unreported overtime they do. Another example is rounding down: work 2.5 hours of overtime but only report 2.
I’m not sure what the solution is, but one idea is to lock out work laptops after a time limit, or at least record overtime when a work laptop is being used.
Until we can begin to accurately track and measure the unreported overtime, we won’t be able to improve it.
Finally, I think working in multicultural teams helps. When you see your peers have diverse beliefs and actions about work and life, you may be more willing to take more time off, and more accurately report the overtime you do take.
In Tokyo, when I have the choice between an above ground train or a below ground option, I tend to choose the above ground choice if I have time and it’s not too much slower. This way I can enjoy the view and avoid the subway tunnel noise. I’d love software that would factor these things in too.
In addition to Differential Privacy, Secure Multiparty Computation is another way to maintain privacy, while allowing computation across multiple users.
The benefit of this is that you can get an exact computation, whereas with differential privacy the output is rougher.
The benefit of differential privacy is that it does not rely on the trust of a majority of other users; you can theoretically verify that a certain percent of the time your device sends out a wrong answer.
MPC works only hand-in-hand with DP. MPC ensures that you don't have access to the "remote" dataset, but does nothing to mitigate model "memorising" specific private information (if we are talking about advanced analytics that is)
Isn't this distorted by the amount of unreported overtime? Anecdotally many of my friends in Japan work overtime but don't report it, or report a smaller overtime amount.
Having worked in Japan and USA, Japanese seem more dedicated to taking a holiday when they have one, also keeping working hours. All my jobs were academic though so I don't know for companies.
