In nonlinear regression like kernel regression, you do not have a single coefficient as in linear regression but a smooth over the domain of a variable (or when in the case of an interaction over the domain of two variables).

You can explore the effect of a single regressor by plotting the effect of the regressor over all values of the corresponding variable by the outcome. This is often called a termplot.

See the graphs of the pdf document you linked on page 11 on the left hand side for an example.

Note that this type of nonlinear regression can also help in finding optimal transformations of the original variables, so you can go back to a linear model with transformed variables.

Nikolaj Tollenaar Many thanks for your response! Makes sense.

Unfortunately, plot(model.np) for npconmode object did not yield the graphs as shown for plot(model.np) for the npreg object in the manual. But will try to figure a workaround.

Thank you once again.

That's because the 2 methods are different. Readup on both and see what data is actually plotted. the first response actually used a different method from these two. But the ideas are similar. Now the comparison.since the RHSs are different the individual parts are not compatible. It is the overall estimates that you should compare. Be sure that both methods apply to your data situation otherwise they won't be compatible.generally 2 different method s are for different situations. If both did the same thing would you need 2? Be safe be cool and remember that the big kahuna is on the way. Best wishes, David Booth

David Eugene Booth Thank you, Sir!