When it's quantitative You calculate eigenvalues to detect as much 2 variables. On qualitative You need to use regresions and logistic models. But the treatment to the model is the same. And anova is the same. And acf is the same.

Hello,

I am a bit puzzled as you can only apply factor analysis with quantitative variables (and not qualitative). The main question with regard to your set of variables is: does the factor analysis makes sense. A (common) factor model implies that there is an underlying singular hidden causal force that generated/affects the indicators that you choose and that is responsible for the intercorrelation of these variables. What should this be in your case?

Best,

Holger

Hi Holger Steinmetz , let me explain further. I am working with a panel data size of about 700 million rows. Having a number of proxies to assess some dimensions, such as the number of bank branches and kindergartens to assess the presence of services in municipalities, I hope to capture with factor analysis these kinds of dimensions with a limited number of factors.

Hi Antonello,

it seems that something like PCA might be more appropriate as mere data reduction seems to be the goal. If you try to predict anything, then either a PCR (combination of PCA and regression or any other predictive model) or a machine learning approach that either applies some kind of penalty of unnecessary predictors (e.g. LASSO) or tree model (e.g., random forest) would work. These approaches select most powerful predictors and eliminate unnecessary ones. As you have panel data, incorporating some time series forecasting might be fancy too.

As I said, by applying a factor model you commit yourself to a statement about the data generating process which is not necessary (and rather unrealistic) in my view.

Best,

Holger