Fandom Data Viz

[redfiona99]

At work I get to play with data using SPSS and Prism.

One of the things we do with the data and programmes is Kaplan–Meier analyses.

But at home, I do not have such things, but I like to play about with stats, partly to understand how a measure works, partly because I find it calming.

I was originally going to look at survival times of US Presidents. But it struck me as morbid, and while I can just about do the maths for 2 categories, I had no idea what to do with the Whig presidents and George Washington. (Ideas about how to do deal with either of the above are gratefully accepted, I think I've got a hang on the maths now.)

Then I realised I had access to a whole swatch of data, which easily divides into two groups. Hits on my fics at AO3.

So I divided my fics into shippy and non shippy, and looked to see if any had more than 1000 hits. (1000 was chosen at random.)

Unfortunately, I can only easily get hold of data for the last 12 months and all of the fics that had reached 1000 were more than 12 months old at the beginning.

But I thought, what the hey, it's over time and it gives me a starting point so I plotted the data using the method outlined here - http://www.real-statistics.com/survival-analysis/kaplan-meier-procedure/survival-curve/

Last month, a new fic had achieved 1000 hits (Battalion of Worries, NC-17 rating shippy Torchwood fic).

The Kaplan-Meier curve now looks like this:



First thing to note, the difference is not significant (at a p<0.05 level. p value was 0.164776, thanks to ioplokon for noticing that I'd not put it in). (I cheated and used http://www.socscistatistics.com/tests/chisquare/Default.aspx to work out my X-squared values.)

I think the analysis might be clearer if I used the reciprocal to produce a curve that goes upwards not downwards.

I've also found another useful source (Survival analysis in clinical practice: analyze your own data using an Excel workbook by Lucijanić - https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4800329/) so there's lots more fun I can have.

Comments

[ioplokon]

So basically does this show the likelihood of the story to continue getting hits after x amount of time? I've got a really poor background in stats but I'm trying to develop a better intuition for these things! It's cool to see the behind-the-scenes process of modeling data!

[redfiona99]

I think it's more like "does thing x have an effect on y happening?" So in this case x is "shippy or not shippy" and y is "reach 1000 hits".

I am most a dabbler in stats, and it is way easier with various programmes but I like knowing what's going on under the hood.

[ioplokon]

Ah, okay and then... the shippy fic is more likely to reach 1000 hits, is that correct? I just wanna be at a point where I can understand & interpret graphs (cause also like, manipulating graphs is a really common way of massaging data in mass-audience publications so... it's good to be able to take a graph apart). Thanks for going into it a bit more!

[redfiona99]

Right now no, the shippy fic isn't more likely to reach 1000, or rather, it could just be random chance that makes it look like it's more likely to reach 1000. (Actually, thanks for the heads up, I forgot to put the actual p value in the original post.) Full explanation once I have the right figures to explain it.

[ioplokon]

That makes sense. Thanks for explaining!