I want to run a nonparametric or semiparametric regression on data which I suspect to be non-linear. At first I thought of using LOWESS -lowess- and Yatchew's semiparametric regression -plreg-(https://journals.sagepub.com/doi/pdf...867X0600600306). However, 41,329 out of 75,582 observations for my independent variable are 0. Theoretically, this is due to a corner solution. (But, as explained by Woolridge, at least for the Tobit case, corner solutions use the same estimation procedures as censored at 0.)

I'm worried that the high amount of zeroes may be throwing off my estimation, and I would rather not simply truncate the data. I haven't been able to find an implementation that allows me to do this sort of local regression in Stata. I found an R implementation (http://ugrad.stat.ubc.ca/R/library/l...it.censor.html), but I'm hesitant to use it because according to "Local Regression and Likelihood" by Clive Loader, this method relies heavily on the normality assumption, but this is difficult to check because in the presence of censoring the estimated residuals are not guaranteed to be normally distributed even in large samples.