Reformatting, only, of orginal post by Johnson Lau <johnsonlau@CUHK.EDU.HK>.

At 11:47 PM 3/27/2007, Johnson Lau wrote:

>I'm currently analysing some election results. The data format is as >follow:

(The following reformatted data brought to you by WRR and the editor Notetab, wondering why so much data does come through unrolled and nearly unreadable. I hope this may help.)

district districtclass age party sex percent A H 32 X M 35 A H 41 Y F 65 B L 21 Y M 70 B L 21 X M 10 B L 50 Z M 20 C L 46 X F 35 C L 35 Y F 45 C L 27 Z M 20 D H 49 Z M 40 D H 41 X M 60 E H 37 X M 30 E H 70 Y M 45 E H 61 Z F 25 F L 28 Y F 40 F L 30 Z F 60

>These mean the District A is a Higher class district. It has 2 >candidates. The first one is Male, 32, representing party X. He got >35% of total vote. The second one is Female, 41, representing party Y, >and got 65% of total vote. District B is a Lower class district. It >has 3 candidates, etc. Regardless the number of candidates, the sum of >"percent" for any "district" must be 100. > >I would like to predict the "percent" by age, party and sex. Since >"percent" is restricted for each "district", I'm not sure what model >should I use for such data structure. I have tried the followings: > >UNIANOVA > percent BY district party sex WITH age > /DESIGN = district party sex age . > >UNIANOVA > percent BY district party sex WITH age > /RANDOM = district > /DESIGN = district party sex age . > >UNIANOVA > percent BY party sex WITH age > /DESIGN = party sex age . > >Would anyone please to suggest which model should I use (and possbily >other model not listed)? In addition, what should I do if I want to >access the interaction effect between "districtclass" and "party"? >Thanks a lot!