I have a problem regarding modelling strategy of a non-linear effect, as follows. In the plot, the blue line is a smoothed function, the red line would be the linear model.
As you can see in the plot, ratio is dependant on age but in a non-linear fashion. I have tried to model it using age^2, but the non-linear effect is still there.
The best model I have come up so far, is to binarize age between young (<50) and old (>50) and fit a linear model through both of them using an interaction term between the age group and the age.
Somewhat this is not very satisfying as the binarization is not based on any biological foundation.
I was therefore wondering if anyone had a good idea for fitting this kind of data?
Thank you in advance!