2

I would like to compare a control group against a test group using something like ANCOVA. However, the covariates are not simple, as some items may have already been increasing with time, others decreasing, all in a non-linear fashion. How would I do a similar analysis with a nonlinear model?

I found a relevant answer here (Repeated measure t test with covariates in R) but it doesn't include cases where the covariates have a nonlinear relationship. I'd like to think I could just do a nonlinear predictive model and then do ANOVA on the residuals. I feel like as long as it follows the assumptions of ANOVA are satisfied I should be okay (ANOVA assumption normality/normal distribution of residuals). Any suggestions?

Thanks!

Community
  • 1
ecksc
  • 93
  • 5
  • What do you mean by non-linear? Do you mean in the sense that you can't solve with least squares? Or in the sense that your covariates don't steadily increase/decrease in a linear, straight-line fashion? – bill_e Feb 13 '15 at 22:39
  • My primary covariate is time. The test and control groups vary in a nonlinear fashion with time. If the test had not happened, I should be able to predict the value fairly well, but with some error. I could just predict the next days' results and compare the residuals against that prediction. My question is whether that is statistically rigorous, as long as the residuals follow those assumptions. – ecksc Feb 13 '15 at 23:04

0 Answers0