0

I have an interesting problem (interesting at least for beginners like myself :) ). I have a dependent variable Y (class-variable, 2 possible values 0 and 1). I have multiple variables that all together can predict the value of Y with p-value smaller than 0.05. Logistic regression was used to see which variable will stay, and which should be removed. So, I got final model, after eliminating statistically unsignificant variables from the model, with 3 variables a, b, and c. All those 3 variables have p-values < 0.05 in the final model. Correlation (Pearson) analysis done on the ancestor of the Y (that is correct, Y was created from continuous variable X) shows that variable a is not correlated at all with X (p=0.7), b and c correlate with p < 0.05. In my analysis, I would like to highlight the meaning of variable c and somehow to eliminate or diminish the meaning of the variable b. c variable's correlation is 30% with X, and b variable's correlation is 15% with X

The question is, can be partial correlation analysis used to highlight the meaning of the c variable? Does it even make sense to do something like this?

user1415536
  • 101
  • 2
  • It seems, at least according to the SPSS Statistics documentation that with partial correlation we can show which relationship will depend on other variable. I think, it can be enough to highlight the effect of the variable c from the newbie point of view – user1415536 Oct 31 '19 at 14:23
  • Partial correlation in linear regression is closely related to regression coefficient: https://stats.stackexchange.com/a/76819/3277. This is where you could find its meaning and hence probably something about the predictor's meaning. – ttnphns Oct 31 '19 at 14:39

0 Answers0