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

The parent is correct, the claimed relationship being y ~ rank^a for some a<0. You would want to see a linear relationship on the log log plot. Another choice would be to examine the complmentary CDF P(X > x) and look for power law tails.


> The parent is correct, the claimed relationship being y ~ rank^a for some a<0

You clearly don't know the difference between exponential and polynomial relationship.

    rank^a is polynomial.

    a^rank is exponential.
The data is neither (unless you can demonstrate it with the actual formula).


I don't really have time to do this because I'm supposed to be working on a paper about, wait for it, power laws, but I went ahead and re-plotted the data using log-log axes. Sure enough, it's linear as would be expected for power law type behavior:

http://imgur.com/qdWxr

I suggest you start with http://www.hpl.hp.com/research/idl/papers/ranking/ranking.ht....


Well of course, pretty much any data looks linear on a log-log chart.

The real question here is what the formula is between the rank and the number of views. You absolutely cannot derive it from the log-log chart.

My point from the beginning is that it's not a power law (y=c*a^bx, where y = views, x = rank and the rest is constants).


> Well of course, pretty much any data looks linear on a log-log chart.

Hah! If you seriously believe that, I've got some stock options I'd like to sell you...




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

Search: