The scores aren't a normal distribution and should be changed to a box plot

[adamwho]

I understand people might find this obscure, but I teach statistics and this drives me nuts.

When you finish a game, it shows your time on a graph that looks like a normal distribution that is broke into quintiles (5 equal parts).

1. This cannot be a normal distribution, because no one can have a time less than zero. The max time can be very high. A Poisson distribution would be a better model

2. The Quintiles cannot be of equal sizes. I understand that the graph is a static display and changing it based on the game was ruled-out as overly complicated and unnecessary. However a box plot where you are calculating the actual MEDIAN and quintiles is much less difficult computationally.

3. Having an accurate display would be useful to players. and provide meaningful data on cheaters/bot. The information can also be used to characterize players so you can serve ads more effectively $$.

[tomohawk52]

" The information can also be used to characterize players so you can serve ads more effectively $$."

Interesting. How would this work, i.e. how would knowing the solving success / speed (or something else?) of any particular user allow the site to serve ads more effectively?

[contrary]

I imagine you have a few demographics here. Someone who sits down every morning and solves a leisurely puzzle might get ads for something like NYT or a brand of coffee; someone who is new and fumbling their way through puzzles might skew toward "train your brain" or "learn how to X" ads; someone who sits down and powers through puzzles for long sessions might get ads for the next addictive game. I'm sure there are other demographics just based on playstyle; when you combine that with other demographic information being collected by the big ad companies I'm sure you could get some well-tailored ads.

[singzeon]

I understand people might find this obscure, but I teach statistics and this drives me nuts.

When you finish a game, it shows your time on a graph that looks like a normal distribution that is broke into quintiles (5 equal parts).

1. This cannot be a normal distribution, because no one can have a time less than zero. The max time can be very high. A Poisson distribution would be a better model

2. The Quintiles cannot be of equal sizes. I understand that the graph is a static display and changing it based on the game was ruled-out as overly complicated and unnecessary. However a box plot where you are calculating the actual MEDIAN and quintiles is much less difficult computationally.

3. Having an accurate display would be useful to players. and provide meaningful data on cheaters/bot. The information can also be used to characterize players so you can serve ads more effectively $$.

Sounds like you found yourself another lesson plan with a demonstration of how statistics is used in real life