How London cycle hire usage varies with weather | Hacker News Reader

How London cycle hire usage varies with weather

reasonableapproximation.net

3 pointsphilh5 comments

I like graphics, but this has too many graphics.

If I understood correctly, the bike usage almost don't depend on the rain. This is very strange and unintuitive. Did I misread the article?

I agree - if I hadn't been writing this to fit udacity's project evaluation specs, I would probably have included fewer. (Even as it is, I guess many of them should have been smaller.)

Bike usage depends noticeably on the rain, which is the only weather feature I discover that it does depend on. Rain causes about a 20% drop. But many caveats: abs(d.num.bikes) doesn't seem to be an especially good measure of bike usage; rain is binary, not quantitative; and I'm not sure how much to trust the rain variable anyway.

Also consider that lots of people use bikes for commuting, and may not have a good alternative. (I have reasonable alternatives, but the rain is basically never at a level that makes me take them instead of cycling.) I don't think I looked at how rain interacted with weekends, but I just had a look and rain causes about a 30% drop on weekends and 15% on weekdays.

      is.weekday  rain mean(abs(d.num.bikes))
    1      FALSE FALSE              0.4567856
    2      FALSE  TRUE              0.3115902
    3       TRUE FALSE              0.5343955
    4       TRUE  TRUE              0.4595437

Interesting. You may copy the graphic of the main result to the beginning of the page with these data, something like a long tl;dr with two or three paragraphs.

Reformatting all the article and graphs is a lot of work, specially if the main objective is other. But a nice abstract with the main result helps the reader to understand what is the interesting part and find it with all the details later.

Out of interest, what if anything would you say the main result is? I don't have one in mind, which I think harms the post - there's no narrative, just a bunch of questions and answers. The three final plots are there because the spec required them, and I chose them more because the plots themselves seemed interesting than for the questions they answered.

But you're right that an abstract would have helped, and I feel a little embarrassed that I didn't think to include one myself. Even without a single main result, I could have included a short list of interesting findings.

(For a blog post) You can declare whatever you like as the main result. I think that the % of the reduction of the bike use by the rain is good candidate. Choose a nice small fact that you can tell to your friend during lunch. Think about the linkbait version: "You can't believe how much the bike use is reduced by rain ..." but please use a non linkbait title, because I hate them.

Usually the biggest change is a good candidate. For example "Moon phase reduce the bike use a 0.027% with p<.99" is a very bad selection.

When you have more time, I think that classifying the rainy days as "small drizzle" and "heavy thunderstorm" will make the changes bigger. I've never been there, but London if not famous for being sunny, so I guess people just ignore the drizzle.

Reply on news.ycombinator.com