Wanikani Users Distribution by Level

I’ve gathered all users information and created these graphs showing how many users are at what level. The number of all users was at the time of download around 28,000.


Number of people at each level.

Drawing the data in log scale and ignoring the first 3 levels since they are biased reveals something interesting.


The number of users drops exponentially which might be another case of Zpif’s law


Now for some some graphs how many users have given up meaning they are no longer paying. ![|345x207](upload://5F3dZIGReHDoVrKzWn4f1v34a3K.png) How many of the users are not paying for wanikani. All the people after level 3 gave up. It is interesting to the spikes on multiples of 5 like 45 and 50.

The spike on 51 are probably all the people that completed wanikani when it only had 50 levels.


Number of all users that are paying for wanikani. It is interesting to see that allot of users are paying even when they reached level 60.


Number of users that have that have bought lifetime subscription. The only interesting thing to see here is that it rises much slower than paid subscription.

All the data is available here in csv format: https://pastebin.com/eDRCb3rr
If anyone wants to play with “raw” downloaded json files, PM me and i will send them to you. The file size is around 3MB.

Now i have to do my 200 reviews that stacked up while i was doing this (x_x)

93 Likes

Very interesting information indeed. It amazes me how high the bar of people on level 1 is, probably meaning that many people signed on and never even started the course. I don’t know why, but I didn’t expect the difference between people on lvl 3 vs people on lvl 1 THAT high… In my mind I kind of assume you at least would give it a chance until you have to pay for it…

10 Likes

I don’t know if they still do this but it used to be that you get level 3 assigned to you if you stop subscription after level 3. So some of those who seem to have given up in level 3 were probably people who reached level 60, then end their subscription.

3 Likes

No this used to be in old forms where only your access level was displayed. Here in new forms even if you are no longer subscribed it still shows you your real level.

If that what you said was true, then the spike at level 3 should be much higher than it currently is.

4 Likes

Just want to add not all WaniKani users have a forum account. It only gets created when the user logs to the forums for the first time. And when we migrated from the old forums only those who have posted had an account made on the new forums :slight_smile:

8 Likes

How did you get that info?

2 Likes

That means that this graph is missing data from A LOT of users. But since these users never “showed up” I don’t think there is a reasonable approach to get their data.

Maybe you could share some of the data (^_~)

4 Likes

Yeah, but you did a pretty good analysis. I don’t think it should be surprising to anyone there is a drop off as they get deeper into the levels. Some people get burned out. Others can’t commit the time anymore. Many find the most effective way for them to study Japanese and we are no longer a good fit in their workflow. Getting halfway through the program sets up a pretty good foundation to go into other areas of Japanese learning.

We haven’t receive any comments of regret from those who make it to 60 though.

31 Likes

@viet, I must inform you of my regret at reaching level 60 on wanikani. Other than being able to

  1. Pick up a Japanese book and start reading with minimal fuss
  2. Recognize and read nearly all the kanji (except for those darned names!)
  3. Learn intermediate to advanced Japanese grammar from Japanese textbooks
  4. Actually believe I might be able to attempt and pass the N2 JLPT
  5. Get a golden level 60 icon for my profile
    I didn’t get anything out of it at all. Not even a lousy t-shirt. Or a sect na… Uh oh, no forget I mentioned that bit.
65 Likes

Thanks for running this analysis; I had always been curious about how these charts would look like.

1 Like

Tshirt may be incoming for 60s. I’ve been pushing it for a while and I think it’ll happen.

“I learned 2,000 kanji and all I got was this t-shirt”

119 Likes

How about underwear for men with the word “金玉” emblazoned in glorious gold.

90 Likes

Dude…

You are onto something.

36 Likes

another tshirt “Viet noticed me”

12 Likes

Thank you for running this data. It actually made me feel, as someone at level 4, that I’ve already jumped my first Wanikani hurdle.

11 Likes

Very interesting statistics! Thanks for taking the time to put this together.

Also: Just a heads up since you did this both in the main post and in this one.

A lot is spelled like “A lot”, not “Allot”. Not trying to be jerk, just thought you would like to know. :heart_exclamation::crabigator::heart_exclamation:

3 Likes

10/10 would buy :wink:

5 Likes

That is really interesting!
There is a small mistake: the distribution of levels does not follow the zipf law, since it is exponential (while zipf means power law!), in other words p(x) = lambda e^(-lambda x). And that makes indeed more sense, since the levels of the users are independent between each other.
A quick calculation from the data you provided, estimates that the “scale” lambda of the exponential decay is around 12 levels. Which I guess it means you can feel proud of yourself once you surpass that level (and not just level 4) :stuck_out_tongue:
Still 3 levels to go for me xD

Details for nerds

Exponential distributions are obtained in the limit of small p of geometric distributions, which have the form: p(1 - p)^k.
Geometric distributions arise when considering how many trials k you need before an event takes place (where the event is drawn from a bernoulli process ). In our case, the number of trials is the number of levels a user reached, and the event taking place is you retrieving the data.
From the data you showed, I estimated that the parameter p of the geometrical distribution is very small (around 0.077) so that we can take the limit of exponential distribution. Considering an exponential distribution, from the data one can estimate that the scale lambda of the decay is around 12 levels.

9 Likes

Is the small bump at 51 people who completed WK when it was 50 levels, left, and then got automatically leveled up when 51-60 came into existence?

4 Likes

Critical reading on this topic:

18 Likes