I am running a survey-weighted logistic regression (svy: logit) to determine the significant predictors of a particular medical outcome. I have included a number of important categorical and continuous variables that are potential predictors. As I am also looking to see the impact of a particular diagnosis on this outcome, I have also included the presence of this diagnosis as an independent variable, as well as interactions with some of the predictor variables. The interactions are actually what I am most interested in - that is, are the variables significant predictors for people with the diagnosis.
Included in the interactions is an interaction with age (i.e., c.age#diagnosis), but due to potential nonlinearity in the age-outcome relationship I have also included age*age as a predictor (as a new variable agesq), as well as agesq#diagnosis.
The coefficients for this regression were significant for age, agesq, age#diagnosis, and agesq#diagnosis. Now I am stuck on how to explain the interaction. With other predictors I can use margins to demonstrate predicted probabilities but I don't know how to do that here. What I ultimately want to be able to display or depict is the additional impact that the diagnosis has on the effect of age on the outcome.
Any help would be greatly appreciated.
S.
Included in the interactions is an interaction with age (i.e., c.age#diagnosis), but due to potential nonlinearity in the age-outcome relationship I have also included age*age as a predictor (as a new variable agesq), as well as agesq#diagnosis.
The coefficients for this regression were significant for age, agesq, age#diagnosis, and agesq#diagnosis. Now I am stuck on how to explain the interaction. With other predictors I can use margins to demonstrate predicted probabilities but I don't know how to do that here. What I ultimately want to be able to display or depict is the additional impact that the diagnosis has on the effect of age on the outcome.
Any help would be greatly appreciated.
S.
Comment