I realize one does not need to explain this upfront as there is no statistical significance... however, in preparing for questions or arguments, and simply understanding the data- what are steps that you would go through in order to "explain" something that does not make sense on the surface.
I have a negative coefficient for a zscore'd variable and a non signigicant pvalue (.494). I would have anticipated the effect would be positive, yet maybe non significant.
Unrelated factors here but I also have the opposite happening, a positive relationship with a significant pvalue... and yet I would not have expected the positive relationship. In the next iteration of the model, adding one more variable, the relationship changes and becomes negative and non-significant.
How would we explain these things? What would be your next steps for testing your model is solid? (regression)
Sounds like there may be an issue with multicolinearity (highly correlated covariates).
for more information take a look at the following website:
https://onlinecourses.science.psu.edu/stat501/node/343
You seem to use different models. The coefficients in these models always refer to the whole model and they are interpreted in the context of the model. If you have different models, the coefficients have different meanings and require different interpretations. So if you are puzzled that the value or the statistical significance of a coefficient is different between different models, then your key problem is that you do not realize that you are comparing different models.
I don't know why you create different models (by adding one predictor after the other). You should have an idea about what predictors are relevant for your purpose and use them, then you have one model and you can start to interpret the meanings of the coefficients in that model.
Richard pointed out a possible difficulty: if there is a linear dependency of predictors (that is: the value of one predictor can be calculated from a linear combination of the values of the other predictors) what is called "(multi-)colinearity", then the model can not decide what predictor actually determines the value of the response what eventually causes large standard errors of the coefficients. These are things, however, that should be considered before starting to think about a model anyway, at least when the model is used to shed light on the functional relationship between the predictirs and the response. If the model is used for prediction, colinearity is not an issue anyway.
Lisa, if your library has Mosteller & Tukey's (1977) classic book Data Analysis and Regression, Chapter 13 (Woes of Regression Coefficients) is recommended reading. HTH.
https://www.pearson.com/us/higher-education/program/Mosteller-Data-Analysis-and-Regression-A-Second-Course-in-Statistics/PGM154255.html
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression