warning: 8956 observations were not used to compute the mean function because they
         violated the model identification assumptions. These observations are
         marked as 1 in the system variable _unident_sample. You may use the
         unidentsample() option to use a different variable name.

all observations violate identification condition
    Most likely, you have included a discrete covariate without specifying factor
    variable notation. npregress interprets the variable to be continuous.  Please
    use i.varname for discrete covariates.
r(198);
请大神指教，为什么所有观测值都违反识别条件。