one assumption that is being made here which may NOT be as valid as one thinks is that ... it is better to have roommates who have similar interests ... than different
perhaps a more sensible assumption is to try to avoid pairing roommates with radically opposing "wants" or "attitudes", etc. for example, given the current global situation, pairing someone from Palestine with someone from Israel ... might not be a very good idea (of course, some might argue just the opposite) however, there is something to be said for trying (at least in some % of the assignments) to deliberately pair roommates with different interests ... that is one way to try to change attitudes and knowledge about lesser known things/characteristics/peoples in any case, using some statistical method to accomplish this seems to be trying to have a "statistic" help you solve what is essentially a human problem and probably won't work finally, what is the criterion here? how often roommates seek to change who their roommates are? again, the assumption with using something like that is that roommate changes are "bad" and, i am not so sure that is the case At 08:05 PM 4/4/02 -0800, Glen wrote: >[EMAIL PROTECTED] (Seymann, Richard) wrote in message >news:<[EMAIL PROTECTED]>... > > This isn't exactly a statistics question, but at least one possible > solution > > might employ a statistics tool. > > > > Here's the problem: > > > > In order to assign roommates, admitted freshmen are given a list of ten > > characteristics and asked to rank-order their preferences from 1 for most > > important to 10 for least important. The objective is to match roommates' > > interests and preferences as closely as possible. Whatever matching > > algorithm is used, it will be run independently for the four > combinations of > > male smokers, female smokers, male non-smokers, and female non-smokers. > > > > So, for each student we would expect a permutation of the integers from > 1 to > > 10, and if any two students submitted the same permutation, they'd be a > > perfect match. My initial reaction is to calculate pair-wise correlation > > coefficients and then sort through the correlation matrix to make roommate > > pairings based on the magnitude (positive) of r, but I'm sure there must be > > other approaches. Any comments, ideas, or suggestions would be welcome. > >Depends on how you want to weight matches/mismatches. There are other >measures/kinds of correlation you might find useful. > >For example, the spearman correlation corresponds to squaring the >differences in rating. Taking absolute values of difference in rating >would correspond to Spearman's footrule. > >Counting 1 for each exact match and 0.5 for each "difference of rank = >1" would be another measure. > >And so on. There's a myriad of possible similarity/distance measures >you might use. > >Consider also that close matching on some criteria might matter more >than on others. Further, consider that close matches on the top-rated >items might matter more than on lower rated items (e.g. I might insist >that my top rated item be in your top two, but not care much where my >middle-rated item is in your list). > >These sort of decisions aren't statistical. They come from what you >want to achieve. > >Glen >. >. >================================================================= >Instructions for joining and leaving this list, remarks about the >problem of INAPPROPRIATE MESSAGES, and archives are available at: >. http://jse.stat.ncsu.edu/ . >================================================================= Dennis Roberts, 208 Cedar Bldg., University Park PA 16802 <Emailto: [EMAIL PROTECTED]> WWW: http://roberts.ed.psu.edu/users/droberts/drober~1.htm AC 8148632401 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
