I have also suggested (in tandem with greywolf) that numerical ratings are subjective and should not be used as anything but a personal measure of the experience.  The "menu" is already given, so why should the numerical ratings include that as well.I also was asking for a way in which one could see how often each "menu" item was performed, because some providers only do certain things with certain customers, and people reading the review should know whether or not they can expect that service.

I can see where ratings creep could be a real problem.  The statistics I have done indicate that the mean of the ratings is about 7 for both appearance and performance.   This is well above the 5.5 that is the middle of the scale.  I don't have statistics on the distribution, but my guess from some spot checks is that the distribution is approximately normal (bell-shaped).  I would guess that the number of 4's is about equal to the number of 10's and the number of 5's is about equal to the number of 9's.  Perhaps I will get the time to check this impression, but not tonight.This means that 1, 2, and 3 are essentially unused at the present time.  I think the only time someone uses these three numbers is to "get" a provider they are mad at.  Even if that isn't true, that is the effect since the numbers are so little used.  One could almost change the scale to go from 4 to 10, except that would probably shift the average upward even further.I like the idea of using weights.  Probably the easiest way to do this is to allow the user to assign a weight (w) in the range 0.0 through 1.0 to performance and automatically assign the weight (1 - w) to appearance.  For example if a provider has an average rating of 5.92 for appearance and 8.83 for performance and I assign a value of 0.667 to w the rating calculation would be1.0 - 0.667) * 5.92 + 0.667 * 8.83 = 7.86097 or 7.86 (rounded)Another feature I would like to see is the ability to exclude older reviews by setting the first month to be used in calculating the individual averages.  For example if I select "Sep 2000" only the reviews from that month or later would be used.  If I set "Sep 2000" for Luscious Laurel I would use her latest 12 reviews (as of last night) and calculate the averages I gave above.  I think the ability to exclude old reviews is important.The problem caused by different providers having greatly different numbers of reviews is one for which I have not found a satisfactory solution.  Perhaps there is someone who knows more statistics than I do who can make a suggestion.  I have found nothing better than displaying the number of reviews that a rating is based on as an indicator of confidence.  One might want to eliminate all results for providers having fewer than a user supplied value.  I typically pay little attention to profiles having fewer than 4 reviews in the time interval (normally 3 to 4 months) for which I am computing the averages.--modified by Mathesar at Wed, Dec 13, 2000, 22:40:06

I don't think you should apply weightings because some people value looks more than performance and vice versa.  Why do you need to combine them?  To sort?  I've suggested that you offer to sort by giving two options "sort by looks" and "sort by performance".  If you really want, you could add "sort by overall", and apply a 50/50 weighting.I don't share the concern about inflated ratings.  People will give 9's and 10's if they really like the girl, and less if they don't.  That's good enough for me.  It's much worse if they give 9's and 10's because of extras that I don't want or can't even tell which were applied.  If it's the best FBSM you ever had with no extras, you should be able to give her a 10!