I’m using mixed model because its repeated measures data, so I allow for each individual to have her own trajectory (and it’s not perfectly balanced in that some individuals have more data points than others). I’m interested in the fixed effects estimates really, but I wanted to take into account the individual variability and correlated error. I want to report the value of B at which the slope of A is zero, which I calculate from the fixed effect parameter estimates. Then I want to output model predicted values so that readers can visualize the interaction. So here’s where it seems conventional to output the conditional predicted values, or is it? And I'd like to know why.
Dear Cora,
This article on mixed models might be helpful:
http://www2.sas.com/proceedings/sugi29/189-29.pdf
Best regards,
Ann
One possibility is to fit models with and without interaction terms, and judge or test whether the interaction terms improve the fit. Another possibility is to fit the model without interaction and plot the residuals against the independent variables. For a graphical display I would plot the ordinary predicted values to demonstrate the predictions produced by the model. If the model is too complex (e.g. more than 2 independent variables), it may be necessary to plot the predicted values at selected combinations (low, medium, high) of the independent variables.
Dear Cora, you can find some free software of Genotype x Environment interaction model developed CIMMYT on internet. I am sure that sofware can help you in this regards,
Best wishes,
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude