Say I have a dataset that contains Users & all the sites that each user visited & how often each user visited it. For example:

User1 -> (Site1, 5 times), (Site5, 3 times)
User2 -> (Site2, 10 times) & so on.. 

Now based on how may times a user visited each site, I need to come up with a rating in the scale of 1 thru 5. For example:

User1's ratings:

Site1 ->  5 (rating)  very interested
Site2 ->  1 (rating)  least interested
& so on...

What's the best statistical technique to compute these ratings? One easy way I can think of is to calculate the range (max - min). Divide the range in 5 buckets & see which bucket each Site falls under. Is this a good way to calculate ratings?

Thanks for your help.

  • $\begingroup$ Are there exactly five sites? $\endgroup$ Mar 16 '12 at 9:23
  • $\begingroup$ Do you have any measure of how long the user stayed on the site? Someone who went once or twice but lasted for more than 30 minutes (say) could be more interested than someone who visited multiple times for <30 seconds a time. $\endgroup$
    – Michelle
    Mar 16 '12 at 11:37
  • $\begingroup$ Peter - No there are quite a few sites. Michelle - Great point. I don't have 'time spent' at this time. I only have 'frequency'. $\endgroup$
    – StatsFan
    Mar 16 '12 at 20:02

Actually in most cases it's best not to convert the usage numbers into ratings, but to work with the original numbers instead. For one thing, you lose information in categorizing, e.g. you can no longer make the distinction between someone 'very interested' and someone 'extremely interested.' More than that, the categorization process will involve some arbitrary decisions: where exactly should the line be between one category and another? And any arbitrariness in your data means their reliability will be undercut.

If you still feel you need to create these categories, there is no single best method for doing so. You just need to come up with one that seems sound, logical, and consistent to those who will be working with the data and those who will be viewing the results of any analyses.

There's some good information here that relates to this question, although it deals with the specific topic of categorizing into just 2 levels.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.