I would appreciate if someone could give me a idea on what I could possibly infer from a residual vs fitted plot for a non linear regression. If i have understood residuals correctly it would indicate whether or not my model predicts the data well. In the figure I have plotted the sinc fit to the data and the actual data vs the right ascension(or distance)and below that is the residual vs fit plot. The inferences I could make from this is:

  1. the sinc function does not predict my data well and I would probably need a more complex function like sinc² .

2.If I am assuming the system noise(0.027) to indicate standard deviation of the fit, would it make sense that I consider fit to be good if the residual variance of the amplitude lies within 2std.dev of the zero line.

enter image description here

Any other comments, corrections and explanations would really help. Thanks

enter image description here


The "figure 8" appearance to the right of 25 is due to the asymmetry of the data (the fit is symmetric, the curve joining the data is slightly "skew"):

connections between regions of misfit of curve and patterns in residual plot

The top right of the residual plot (indicated by the red ellipse) corresponds to the part just before the peak, while the bottom right (indicated by the green ellipse) corresponds to the part just after the peak. We can see the data lie above the curve before the peak and below it just after, then the curves cross over near the points where the variable is about 30, and the figure-8 in the residual plot crosses over in that region.

There's similar kinds of lack of fit at the far left of the original plot (where it looks "tangled" in your plot) but it's hard to disentangle in detail because there are several such "loops" overlaid in a narrow space.

In this case a plot of residual against Right Ascension would probably be more directly informative.

  • $\begingroup$ Thanks for the explaination.I have also uploaded the is the ra vs residual plot now. $\endgroup$ – carlton xavier Dec 5 '16 at 10:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.