Suggestion for a statistical test

Question

I hope this is the right forum for chemometrics related questions well. Imagine that we have a list of tabulated molecular properties for 40 different molecules such as A) dipole moments B) water solubility C) molecular surface area and so on. Let us just call them A, B, C, D,..., Z.

Now if we have a set of an experimental observable for those 40 compounds called "chromatographic efficiency", N. The value N does not directly depend on A,B,C, D but may be influenced by them. We don't know. By that I mean, there is no chemical theory which relates N with A,B,C,D,... directly.

All the molecular properties A, B, C, D ..., may have a combined effect on N. Is there a test which can show that N is "somewhat" influenced by property B, but not by A and D i.e. something like a correlation test but for multi-variables.

Is there a statistical test which can collectively see the influence of the effect of molecular properties (A, B, C, D,...) on the experimental observable?

Answer 1

This sounds like an appropriate situation for regression analysis. In this form of analysis you would let the chromatographic efficiency be your response variable and you would let the other properties (dipole moments, water solubility, molecular surface area, etc.) be your explanatory variables. From your description it sounds like each data point would be for one molecule.

In this method you would posit some kind of appropriate regression model, fit your model and check its assumptions using diagnostic testing. Once you have an appropriate regression model that fits the data well, you will be able to make inferences about the statistical relationship between the explanatory variables and the response variable. Care must be taken in interpreting these things causally (e.g., as "influence" on the response variable), but you will at least be able to get evidence of how these variables are statistically related.