Thank you this was very usefull! i know that you adviced for the usage of confidence interval to test for significant slopes. However, if i decide against that and use an anova on the segmented model it becomes somewhat tricky to interpret it... And i cant find resources regarding anovas on segmented models. Do you have any idea where i can get some insights?
I have a doubt. The segmented package returns only one coefficient of determination (Multiple R-squared). Does this Multiple R-squared consider the R-squared of the two models if I have one breakpoint or tree models if I have two breakpoints? Is there a way to know the R-squared of each model fitted? Thank you
While I can't speak to all possibilities from the package or the ability to derive separate R-squared values for each segment, the output from the package should represent the overall variability in our outcome explained by the given model. So, if we have 2 breakpoints, the R-squared included in the output is the overall summary for that model having 3 segments. One option for calculating an R-squared for each segment could be to estimate the cutpoints, then fit individual linear regression models restricted to the data within each segment. However, I am not sure that this information is as useful as the overall R-squared from the segmented package, since it is not clear if one segment had lower R-squared if that is important relative to the overall performance of the given model.
Thank you this was very usefull! i know that you adviced for the usage of confidence interval to test for significant slopes. However, if i decide against that and use an anova on the segmented model it becomes somewhat tricky to interpret it... And i cant find resources regarding anovas on segmented models. Do you have any idea where i can get some insights?
This is really excellent, really useful
I have a doubt. The segmented package returns only one coefficient of determination (Multiple R-squared). Does this Multiple R-squared consider the R-squared of the two models if I have one breakpoint or tree models if I have two breakpoints? Is there a way to know the R-squared of each model fitted? Thank you
While I can't speak to all possibilities from the package or the ability to derive separate R-squared values for each segment, the output from the package should represent the overall variability in our outcome explained by the given model. So, if we have 2 breakpoints, the R-squared included in the output is the overall summary for that model having 3 segments. One option for calculating an R-squared for each segment could be to estimate the cutpoints, then fit individual linear regression models restricted to the data within each segment. However, I am not sure that this information is as useful as the overall R-squared from the segmented package, since it is not clear if one segment had lower R-squared if that is important relative to the overall performance of the given model.