Statistics Filter: Finding the top 10% from thousands of data points. Bonus question: inflection point value.
August 7, 2008 9:49 AM   Subscribe

Statistics Filter: Finding the top 10% from thousands of data points. Bonus question: inflection point value.

Statistics Filter: I have thousands of data points (ranging in value from 1 to 0, 8 decimal points) representing a score. I graphed the data set and it's a classic long-tail distribution--looks like a hockey stick. How do I find the value of the "inflection point" and also the value/position of top 10%, in other words, which scores are in the top 10%. I've googled my little heart out and think I don't know the right questions to ask. Thanks, wise ones!
posted by lrivers to Computers & Internet (4 answers total)
 
What program are you using/do you have available to you?
posted by desjardins at 9:59 AM on August 7, 2008


Best answer: In R, put your data points into a vector called x, then type quantile(x,probs=0.9).

Programmatically, you can find a point of inflection by taking the absolute value of the rate of change from one point to the next. Then take the minimum value across those values.
posted by Blazecock Pileon at 10:11 AM on August 7, 2008


Even the most basic program (e.g. Excel) will tell you the highest value, which should be your "inflection point," unless I'm misunderstanding. So just sort your column Z>A and take the top value.

Then enter this formula into another column (C2 in the next step) and it will give you the cutoff value for the top 10%: =PERCENTILE([enter range here],0.9)

Enter =IF(A2<C$2,TRUE) into B2 and fill it down column B. All your values with TRUE are in the top 10%.
posted by desjardins at 10:18 AM on August 7, 2008


An inflection point of a curve is where the second derivative is zero. You can modify BP's suggestion to find it.
posted by Horselover Fat at 12:58 PM on August 7, 2008 [1 favorite]


« Older Can you tunnel via SSH to access Gmail via IMAP in...   |   Mapgeek looking for a career Newer »
This thread is closed to new comments.