classification

[Jon Oh]

If we want to classify survey respondents into several groups based on the
score they obtain from three seperate scales. Is there a way to determine
the cutoff scores for grouping?
I know I can decide the cutoff myself and classify respondnets into groups
first and then run discriminant analysis or multinomial logistic regression
to see how accurate my classfication is. But, if the classification turns
out to be bad, how do I find the better cutoff?

thank you

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[mpirritano]

Jon,

You could use Cluster Analysis. Using K-Means or Hierarchical Cluster
Analysis groups are determined by relative locations on multiple scales.
If you request cluster membership be save to the dataset a new variable
will be created that indicates the group, or cluster, individuals belong
to. You would have to determine, somewhat subjectively, what the groups
created by the cluster analysis mean. You would also have to determine
what number of clusters makes the most sense from a theoretical
standpoint. It's similar to the subjective steps involved in doing
exploratory factor analysis. It is in fact akin to factor analyzing
individuals instead of questionnaire items. You can then include cluster
membership as a variable in subsequent analyses and see how well it
works to distinguish people on theoretically meaningful dimensions.

One other important thing to keep in mind here, I think, is since there
is some degree of subjectivity involved, you want to use convergent
evidence from multiple methods of clustering. If you start to look at
the cluster analysis you'll come across some of the standard methods
such as K-Means and Hierarchical Cluster Analysis. Just a google search
of these terms should provide some useful info. It's also worthwhile
noting that some of these methods are good at clustering your data but
theoretically do not create clusters that represent true population
level classes of individuals.

Matt

Matthew Pirritano, Ph.D.
Research Analyst IV
County of Orange
Medical Services Initiative (MSI)
[hidden email]
[phone]


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Jon Oh
Sent: Monday, July 14, 2008 6:57 PM
To: [hidden email]
Subject: classification

If we want to classify survey respondents into several groups based on
the
score they obtain from three seperate scales. Is there a way to
determine
the cutoff scores for grouping?
I know I can decide the cutoff myself and classify respondnets into
groups
first and then run discriminant analysis or multinomial logistic
regression
to see how accurate my classfication is. But, if the classification
turns
out to be bad, how do I find the better cutoff?

thank you

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD