Rating people on their looks: How to normalize user ratings between 1 - 10?

P

PiMakesMeHungry

New member
Joined
Feb 11, 2012
Messages
4
If you have a Database of users who have rated each other on looks from 1 - 10, how should you normalize that kind of data?

I'm thinking of just doing it in percentiles as per: http://en.wikipedia.org/wiki/Percentile_rank

But that doesn't seem intuitive to me as there are probably more "average" looking people than in either extreme.

Any ideas?

Thanks a bunch.
 
You already have your answer ...

If the data fits well to a normal distribution, as you suspect, and as would make sense, then further normalization would not be required. If it fits a log-normal distribution, which is probably more accurate, again, you have your answer. There is no reason to distort your data.
 
Will57 said:
If the data fits well to a normal distribution, as you suspect, and as would make sense, then further normalization would not be required. If it fits a log-normal distribution, which is probably more accurate, again, you have your answer. There is no reason to distort your data.
Click to expand...
Thanks for your reply, but I do not think this is correct.

For example, the lowest rating for men is a 1, the highest is a 9, but the average is 3.6. For women, the lowest rating is also a 1, the highest is 9.3, but the average is 4.4.

In both cases, I would like to normalize the ratings so that the average is a 5, and the rest of the ratings fit on a normal curve around 5. How would I do that?

It does seem that normalizing ratings is pretty standard for sites that rate people. This is from the FAQ of one of the more popular ones:

My rating and my votes don't seem to match. What's wrong?

Official HOT or NOT scores are normalized to account for things such as the fact that different people have different voting styles. For instance, some people, for reasons unknown to us, only vote 1-5. Should a 5 from these people count as a 5?Many standard scoring systems use normalization techniques, including the Olympics and the tests people take to get into college in the US. We still give you the chart, which represents raw data, just for fun.
Click to expand...


Thanks again for your help.
 
PiMakesMeHungry said:
Thanks for your reply, but I do not think this is correct.

For example, the lowest rating for men is a 1, the highest is a 9, but the average is 3.6. For women, the lowest rating is also a 1, the highest is 9.3, but the average is 4.4.

In both cases, I would like to normalize the ratings so that the average is a 5, and the rest of the ratings fit on a normal curve around 5. How would I do that?

It does seem that normalizing ratings is pretty standard for sites that rate people. This is from the FAQ of one of the more popular ones:



Thanks again for your help.
Click to expand...

Standard way of normalizing data is to transform each data point by multiplying with [(x - m)/s] where

x = given data point

m = average of the group

s = standard deviation of the group.

This will make transformed average = 0 and transformed s = 1
 
JeffM said:
Not my joke. All politicians lie and cast iron sinks, but not all politicians lie in cast iron sinks.
Click to expand...

Time flies like an arrow - fruit flies like a banana ...... Groucho Marx
 
You must log in or register to reply here.
Top