Hello everyone,
I was wondering if someone could help me with the following.

I have a database with n observations. There are nine variables, all of
which are dichotomous.
I would like to bring back these nine variables to only one variable (I
guess this variable would then need to be categorical).

What is the best way to do this?

There are 2^9 = 512 possible combinations of the nine variables. I was
thinking of a categorizal variable with, let's say, 6 categories.
I was thinking of using a cluster analysis. Would you agree with this? The
problem is that there are not really any 'natural' clusters. It is kind of
evenly spread out over the 512 possible combinations. How to deal with this?

Cheers!

Hello everyone,
I was wondering if someone could help me with the following.

I have a database with n observations. There are nine variables, all of
which are dichotomous.
I would like to bring back these nine variables to only one variable (I
guess this variable would then need to be categorical).

What is the best way to do this?

There are 2^9 = 512 possible combinations of the nine variables. I was
thinking of a categorizal variable with, let's say, 6 categories.
I was thinking of using a cluster analysis. Would you agree with this? The
problem is that there are not really any 'natural' clusters. It is kind of
evenly spread out over the 512 possible combinations. How to deal with this?

Cheers!

thinking of a categorizal variable with, let's say, 6 categories.
I was thinking of using a cluster analysis. Would you agree with this? The
problem is that there are not really any 'natural' clusters. It is kind of
evenly spread out over the 512 possible combinations. How to deal with this?

On Sat, 6 Jan 2007 19:23:15 +0100, "Nickk" <Nickk40@gmail.com> wrote:

One of my hypothesis is that the nine dichotomous variables,as a group,
affect "performance" of a person. Therefore, for each person I also have
a
performance-score.

If there is a natural good/bad direction for each variable,
the natural thing is to take an average score. Or total.

Factor analysis is more usual than cluster analysis.

If all the correlations are about the same, the factoring
will tend to segregate variables with similar skewness,
since they can have better correlations with each other.


--
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html

N is 150.
Unfortunately, there is not a natural good/bad direction for each variable.
They are traits (not good/bad but just different).
So there are 9 traits (which the cases either have (1) or have not (0)).
This trait-profile is a variable in a larger SEM model. To be able to make
sense of this model, it is necessary to reduce this trait-profile of 9
variables into a single variable.

My first thought was that there are 2^9 = 512 possible profiles. We could
bring the 9 variables back to a single categorical variable with 512
categories. 512 categories is of course too many (something like 6
categories seems more appropriate). So the question is then how can I reduce
the 512 categories into, for example, 6 categories (preferably with some
meaning) and subsequently classify each case into one of the 6 categories?

- Factor analysis / Principal components analysis: This technique reduces
the 9 variables into one or more continuous variables that are combinations
of the values of the 9 variables. I don't see how a continuous variable can
capture the profiles.

- Cluster analysis: Cluster analysis looks at similarities between cases in
their profile of the 9 variables. So, this seemed a good idea to use.
However, there are not really any 'natural' clusters. It is kind of evenly
spread out over the 512 possible combinations.
- Latent class analysis: I'm not familiar with this technique. But it seems
very interesting. Nevertheless, because it requires special software, I was
hoping the problem could also be tackled with another technique.

- Latent class analysis: I'm not familiar with this technique. But it seems
very interesting. Nevertheless, because it requires special software, I was
hoping the problem could also be tackled with another technique.