Hi
Im currently doing a working with multiple regression in STATA and have a question about linearity and
Linearity
I'm currently going through the regression assumptions and have issues testing for linearity.
My dependent variable has been made into an index, but my independent variable and many of my control variables are ordinal scaled.
So how do I check for linearity? When running the regression analysis I simply put an i. in front of the ordinal variables to get my results, but when checking for linearity that doesn't seem to work. How do I proceed? Is it only possible to check for linearity with interval scaled variables? Should I simply not check for linearity with ordinal/categorigal variables?
Multicollinearity
When doing a VIF test I get multicollirinarity inside 1 of my categorical variables. The variable is about level of education broken into dummy variables (upper secondary school, short-cycle higher education, medium-cycle higher education, long cycle higher education). The reference category is secondary upper school. When doing VIF test long cycle higher education and medium-cycle higher education have a VIF on 7.08 and 6.60.
Should there be done anything about this as they're a part of the same variable or not?
Thank you on beforehand
Kind regards,
Malte
Dear Malte,
I think the following links will help you at least to have an idea
https://www.google.com.tr/url?sa=i&source=web&cd=&ved=2ahUKEwj2ksm_y5LfAhXJJlAKHdEzAT8QzPwBegQIARAC&url=https%3A%2F%2Fwww.researchgate.net%2Fpost%2FWhat_is_the_right_regression_model_to_use_when_my_dependent_variable_is_an_index_rated_from_1_to_16_point&psig=AOvVaw1Lg3EkpHMH9ZkkmODZPkrU&ust=1544439466677621
https://stats.stackexchange.com/questions/95608/which-regression-to-use-for-ordinal-variables
Hello Malte, what really is your dependent variable whose values would assume some relationship with the categories of education you mentioned? By index do you mean measurement constructed by summing up other simpler measurements? I just want to know how you classify your dependent variable (nominal, ordinal, interval or ratio).
Maurice Ekpenyong my dependent variable is a reflective index. We did factor analysis with 10 different items (all 7-point-likert scales), found 1 factor and made a reflective index ranging from 0-10 (of course checking for reliability and validity).
My OLS regression looks fine and I guess I don't have to check for linearity when my independent variable and controlvariables are ordinal? I guess we just assume linearity here?
About the multicollinearity I guess it's important to mention, that education is just a control variable, so would it be acceptable to not do anything about the 2 VIF values above 5? At the same time it's two dummies inside one categorical variable that get high VIF, so it's not taking away explanatory power from any other of the categorical variables.
Kind regards,
Malte
Variance inflation factors (VIF) measure how much the variance of the estimated regression coefficients are inflated as compared to when the predictor variables are not linearly related. It is used to explain how much multicollinearity (correlation between predictors) exists in a regression analysis. There is no formal VIF value for determining presence of multicollinearity. Values of VIF that exceed 10 are often regarded as indicating multicollinearity, but in weaker models values above 2.5 may be a cause for concern. So it is really a matter of model strength. What parameter determines model strength? R-squared, or more correctly adjusted R-squared. The factors especially residuals that influence R-squared values will influence the strength of your model and the strength of the model subsequently determines the VIF value that suggests the presence of multicolinearity of your predictor variables. For a strong model, your 2 VIF values above 5 are OK (no muliticolinearity) but if your model was weak, then you have good reason to worry about multicolinearity. Please read the attached. I hope it helps you
Hi Malte,
You might like to read the following:
Good luck!
Shah
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics