Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> And I rolled it over 30 days to be able to show any recent changes while not getting lost in the noise of daily change.

small thing, but learned at work that often there's a weekly pattern (and I'd bet there is for airplane travel) so you ought use a rolling 7 day average instead of 30 because there are different numbers of weekends in each day's 30 day number.

This is why there's a slight zigzag in the line charts in the article.




Or use 28 days ? If we need to average both weekly and monthly trends


yep typically rolling 28 days is used for “monthly” metrics for this reason


I've been following work metrics quite closely this year that has this problem. I don't know why I didn't think of this!


One fun one is that in a lot of year on year change reporting for China specifically, Jan-Feb is considered a single month and then all the other 10 months are seperate. Reason is:

[spoiler]

*

*

*

*

*

The dates of Chinese New Year move around across the year and they have such a big effect on the stats that it's easier just to lump the two months together.

[/spoiler]


Or just seasonally adjust your time series. Just sayin.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: