nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Multiple responses and multiple predictors

FR60

I would like to know which is the best tecnique to model a set of data with multiple responses Yi and multiple predictors X.

I tried the fit model option but the results are not so good.

BY using the PCA I was able to reduce the Y responses from ~50 at ~10. Still a high number.

Rgds. Felice . .

Accepted answers

Mark_Bailey

If you want to model all the responses simultaneously, not individually, then consider using PLS or neural network. If you want to model individual responses, then you might first use Analyze > Screening > Response Screening and Predictor Screening to explore possible relationships.

All comments

Mark_Bailey

You seem to be asking two questions: modeling techniques and dimension reduction. The PCA result does not seem like a lot of predictors to me. Why is ~10 still too large?

There are many modeling techniques for such a case. Are you building an explanatory model or a predictive model? The former is about estimating and testing parameters, such as a design experiment. The latter is about estimating the response, as in machine learning and AI.

What do you mean when you say, "I tried the fit model option but the results are not so good." What were you expecting? Why were you disappointed? What was lacking?

Are you trying to model multiple responses independently? That is, each response is modeled with its own model. If not, how are the responses related?

FR60

Ciao Mark thanks for your input.

For the PCA I was referring to Y responses and not to predictors X.

Generally I prefer to work with few responses, for this reason I said that 10 for me was still large.

I'm building an explanatory model by using production data. In other words I want to check if exist any relationship betweens the X predictors and the multiple responses Y.

About the fit model I was saiyng that results were not good due to low Rsquare (<5%) and high pval (0.18).

I'm triyng to model all the responses togheter. No collinarity for them.

Are you suggesting to run as many models as there are responses Y?

Felice

Mark_Bailey