Generating a matrix of nominal correlation values in Crosstabs

[N Knight]

Dear all,

When I need to run a factor analysis on dichotomous variables I
generally create a Kendall's tau correlation matrix, copy that into
syntax, and run the factor analysis from matrix input.

I am now dealing with a set of 7 variables that are quasi-ordinal, but
which for the sake of honest statistical procedure I want to treat as
nominal.

Measures of association for nominal variables, such as Goodman-Kruskal's
Lambda and Cramer's V, are found under Crosstabs, rather than under
Correlate. So far, I have not found a way to generate matrix output from
Crosstabs, and I'm afraid that constructing a matrix from the piecemeal
output of Crosstabs will make it more likely for mistakes to creep in.

Does anyone know if it is possible to create a matrix of Lambda and/or
Cramer's V correlation values?

All the best,

Nicola Knight

--
Nicola Knight (Mr)

PhD candidate, Psychology/Anthropology/Culture & Cognition, University of Michigan
Visiting research associate, Department of Anthropology, LSE

http://sitemaker.umich.edu/nicola.knight

[N Knight]

In my previous email I forgot to mention that each of the 7 nominal
variables has 3 levels, so all measures of association for dichotomous
variables are unsuitable (hence my desire to use Lambda or V).

Cheers,

N Knight

N Knight wrote:

> Dear all,
>
> When I need to run a factor analysis on dichotomous variables I
> generally create a Kendall's tau correlation matrix, copy that into
> syntax, and run the factor analysis from matrix input.
>
> I am now dealing with a set of 7 variables that are quasi-ordinal, but
> which for the sake of honest statistical procedure I want to treat as
> nominal.
>
> Measures of association for nominal variables, such as
> Goodman-Kruskal's Lambda and Cramer's V, are found under Crosstabs,
> rather than under Correlate. So far, I have not found a way to
> generate matrix output from Crosstabs, and I'm afraid that
> constructing a matrix from the piecemeal output of Crosstabs will make
> it more likely for mistakes to creep in.
>
> Does anyone know if it is possible to create a matrix of Lambda and/or
> Cramer's V correlation values?
>
> All the best,
>
> Nicola Knight
>

[Art Kendall]

try CATPCA    CATegorical Principal Components Analysis

Art Kendall
Social Research Consultants

N Knight wrote:

> In my previous email I forgot to mention that each of the 7 nominal
> variables has 3 levels, so all measures of association for dichotomous
> variables are unsuitable (hence my desire to use Lambda or V).
>
> Cheers,
>
> N Knight
>
> N Knight wrote:
>> Dear all,
>>
>> When I need to run a factor analysis on dichotomous variables I
>> generally create a Kendall's tau correlation matrix, copy that into
>> syntax, and run the factor analysis from matrix input.
>>
>> I am now dealing with a set of 7 variables that are quasi-ordinal, but
>> which for the sake of honest statistical procedure I want to treat as
>> nominal.
>>
>> Measures of association for nominal variables, such as
>> Goodman-Kruskal's Lambda and Cramer's V, are found under Crosstabs,
>> rather than under Correlate. So far, I have not found a way to
>> generate matrix output from Crosstabs, and I'm afraid that
>> constructing a matrix from the piecemeal output of Crosstabs will make
>> it more likely for mistakes to creep in.
>>
>> Does anyone know if it is possible to create a matrix of Lambda and/or
>> Cramer's V correlation values?
>>
>> All the best,
>>
>> Nicola Knight
>>
>
>