Fairly novice in statistics, so this will surely have a simple explanation. I ran bivariate analyses to check for normality of a series of variables and also to test the relationship of each with two outcome variables (one binary the other continuous). In the bivariate analyses (using either linear or logistic regression) none of the variables appear to be related to the outcome variables. However, when I ran the analyses including multiple variables in a GLM and GLMZ model, I did get significant results for some of the variables and good model fit statistics. How should I interpret this?
Different p-values with different procedures
- Thread starter RNG
- Start date
- Tags general linear model generalized linear model linear regression logistic regression p-values
Univariate models and multiple regression (regardless of whether you use GLM, logistic, linear regression etc) will commonly generate different results. Variables that are signficant in one won't be in another. It is because univariate analysis just looks at the explained variance in the DV by the single variable. In multivariate models only the variance explained by that IV which is unique (that is not shared with any other independent variable) is used to calculate effect size and thus signficance. When various IV predict parts of the same variance in the DV, and commonly they will, the results will be very different than with just a single variable predicting the DV.