Optimal scaling

[Sofia Cruz]

Hello there!!,

I am trying to create an index to measure the quality of employment using Principal Component Principal, but I have some categorical variables on my data, so I need to do an optimal scaling, I am not finding a command to do it, can I help you.

Thanks for your attention.

[Marcos Almeida]

You may think about - gsem - models.

[wbuchanan]

Sofia Cruz
Why not fit an IRT model so you can also check for other types of measurement invariance and then predict the latent scores? This is similar to what Marcos Almeida suggested, but would give you a bit more information about which items provide the most information for specific parts of the distribution of theta.

[Sofia Cruz]

Thank you Guys!!,

I will checking these models, but how do you think if I use homals method?

[Joseph Coveney]

how do you think if I use homals method?

What is homals method?

[Carlo Lazzaro]

I assume that Sofia refers to https://www.jstatsoft.org/article/view/v031i04

[Sofia Cruz]

Hello there,

Thanks Carlo, Yes I am talking about method on your link, I read there are exit methods to do an optimal scaling

• Categorical regression analysis: Undertakes multiple regression when some of the variables are ordinal or nominal.
• Nonlinear Principal Components Analysis: Undertakes conventional principal components analysis, when some of the variables are ordinal.
• Nonlinear Canonical Correlation: Undertakes canonical correlation when some of the variables are ordinal and nominal.
• Correspondence Analysis: Undertakes something like a factor analysis on two categorical variables.
• Homogeneity Analysis (also called Multiple Correspondence Analysis): Undertakes something like a factor analysis on more than two categorical variables.

My problem I want to built a quality index using PCA, for that my variables need to be continue, I have categorical variables, like contract: Yes (1) No (2), Income= 1 (less of minimum wage)
= 2 ( between 1 and 3 minimum wage)
= 3 (more 3 minimum wage)
Subsidy = 1 (Yes)
= 2 (No)

Between others; So, Stata has a command named mca to do Multiple Correspondence Analysis, then I can use these results to built the index using PCA?

Please, somebody help me.

[wbuchanan]

I’m still not clear why it isn’t possible for you to use the irt commands to do this? Your example of a nominal scale variable above can be considered a special case of an ordinal scale with two categories. So you really have a series of ordinal scale variables with varying numbers of categories on which the responses are keyed. Depending on the assumptions you are willing to make you could probably use either the graded response or partial credit model to accomplish creating a scale based on the underlying latent construct you are trying to measure; this of course rests heavily on the assumption that it is a unidimensional construct that you are attempting to measure.

[Sofia Cruz]

Thanks wbuchanan!!, I had doubt about command mca because I have ordinal variables with different categories but I am working on.