[R] what statistical method should i use?

[Jonathan Baron]

On 05/16/04 00:23, vinkwai wong wrote:
>in order to know which production the custumer most like,i design a question as follow
>:
>
>Q:there are six production listed below.according to your preference,the production
>you like most is_____,the production you secondly like is ____,and the third is_____.
>productionA      productionB      productionC      productionD     productionE     
>productionF
>
>when the data is collected. i type in a stata in such format:
>
>firstlike          secondlike          thirdlike
>A                      C                   D
>E                      A                   E
>¡­¡­¡­¡­¡­¡­¡­¡­¡­¡­¡­¡­¡­¡­
>
>if i want to make a decision what production should i choose as my main production
>according to the survey,what statistical method should i use to analysis my data ?
>
>my aim is to let the analysis result support my descision.

This isn't an R question yet, so it is off topic, but let me
answer in that spirit.  (It may become an R topic when it gets to
the point of converting the ranks into numbers, but you didn't
ask about that.)

I don't think this is a statistical question so much as a
question about decision making (my field, sort of).  And my
answer would be that you need to think about your goals (what you
are trying to achieve) and also about what could be producing
these rankings.

You mention customers.  If you are trying to produce something
that competes in a market, you need to think about market share.
If there are competitors, you need to think about them.  It could
be that the best option is not the one with the highest ranking
but rather the one that fills a gap (niche) in the market that
nobody else is filling very well.

If, on the other hand, these judgments are simply expert opinions
about the answer to the same question, then your task is
simpler.  You might make some simplifying assumptions, which
would allow you to base your decision on the average rank of each
option.

There are other models you could apply, such as the Rasch model,
or various models in the same spirit that often go under the name
of "random utility" models, although most of them derive in some
way from Thurstone scaling.

If the judgments are votes, then it would depend on what you told
the voters.  There are various methods for getting winners out of
rankings, but they make sense only if the voters know in advance
which method would be used.  Two of these are the Borda count
(which essentially adds the ranks) and the "single transferable
vote" or "instant runoff," which looks for a majority winner from
the first ranks, then, if it doesn't get one, takes the biggest
loser's second-place votes and adds them, and so on.

Finally, you could do some sort of cluster or classifcation
analysis on your subjects, to see if they really fall into
distinct groups with very different opinions.

I have used R for many of the things I just listed (but not all),
so I might be able to provide examples.

Jon
-- 
Jonathan Baron, Professor of Psychology, University of Pennsylvania
Home page:            http://www.sas.upenn.edu/~baron
R page:               http://finzi.psych.upenn.edu/