0

I estimate a local linear model for the data with one continuous dependent variable and multiple explanatory variables (continuous and dichotomous).

Is it possible to estimate the derivatives of this function for each bundle of explanatory variables and save them for the further usage? How can one do it in SAS?

In principle, I would like to get some kind of analogue to the parameter estimates in a simple parametric regression. But now I would have not a one-point estimator but rather a distribution for each of the variable.

Thanks for any suggestions, comments, clarifications.

UPDATE:

For example, if I use the SAS datasample "ExperimentA" and run the local linear model, how can I get after the estimation the derivative for each indep.variable at each row of the data?

data ExperimentA;
    format Temperature f4.0 Catalyst f6.3 Yield f8.3;
    input Temperature Catalyst Yield @@;
datalines;
80  0.005 6.039  80 0.010 4.719  80 0.015 6.301
80  0.020 4.558  80 0.025 5.917  80 0.030 4.365
80  0.035 6.540  80 0.040 5.063  80 0.045 4.668
80  0.050 7.641  80 0.055 6.736  80 0.060 7.255
80  0.065 5.515  80 0.070 5.260  80 0.075 4.813
80  0.080 4.465  90 0.005 4.540  90 0.010 3.553
90  0.015 5.611  90 0.020 4.586  90 0.025 6.503
90  0.030 4.671  90 0.035 4.919  90 0.040 6.536
90  0.045 4.799  90 0.050 6.002  90 0.055 6.988
90  0.060 6.206  90 0.065 5.193  90 0.070 5.783
90  0.075 6.482  90 0.080 5.222 100 0.005 5.042
100 0.010 5.551 100 0.015 4.804 100 0.020 5.313
100 0.025 4.957 100 0.030 6.177 100 0.035 5.433
100 0.040 6.139 100 0.045 6.217 100 0.050 6.498
100 0.055 7.037 100 0.060 5.589 100 0.065 5.593
100 0.070 7.438 100 0.075 4.794 100 0.080 3.692
110 0.005 6.005 110 0.010 5.493 110 0.015 5.107
110 0.020 5.511 110 0.025 5.692 110 0.030 5.969
110 0.035 6.244 110 0.040 7.364 110 0.045 6.412
110 0.050 6.928 110 0.055 6.814 110 0.060 8.071
110 0.065 6.038 110 0.070 6.295 110 0.075 4.308
110 0.080 7.020 120 0.005 5.409 120 0.010 7.009
120 0.015 6.160 120 0.020 7.408 120 0.025 7.123
120 0.030 7.009 120 0.035 7.708 120 0.040 5.278
120 0.045 8.111 120 0.050 8.547 120 0.055 8.279
120 0.060 8.736 120 0.065 6.988 120 0.070 6.283
120 0.075 7.367 120 0.080 6.579 130 0.005 7.629
130 0.010 7.171 130 0.015 5.997 130 0.020 6.587
130 0.025 7.335 130 0.030 7.209 130 0.035 8.259
130 0.040 6.530 130 0.045 8.400 130 0.050 7.218
130 0.055 9.167 130 0.060 9.082 130 0.065 7.680
130 0.070 7.139 130 0.075 7.275 130 0.080 7.544
140 0.005 4.860 140 0.010 5.932 140 0.015 3.685
140 0.020 5.581 140 0.025 4.935 140 0.030 5.197
140 0.035 5.559 140 0.040 4.836 140 0.045 5.795
140 0.050 5.524 140 0.055 7.736 140 0.060 5.628
140 0.065 6.644 140 0.070 3.785 140 0.075 4.853
140 0.080 6.006
;
run;

ods graphics on;
ods output ScoreResults=PredLOESS;
proc loess data=ExperimentA;
    model Yield = Temperature Catalyst
                  / scale=sd select=gcv degree=2;
    score;
run;
Vlada Pleshcheva
  • 81
  • 6
  • @Close voter: this is an appropriate Stack Overflow question. It perhaps wouldn't be off topic on [stats.se], but it is certainly on topic here. It is asking how to do something programmatic in SAS. (Typically, [stats.se] prefers not to host questions asking in a _particular_ language, though they generally allow them when they are asked over there if they're not too specific. If this were asking if it is possible _in general_ this would be off topic here. But asking about SAS specifically makes it appropriate here.) – Joe Feb 05 '15 at 17:02
  • Vlada - You will probably get a better answer if you provide some sample data and code to show what you're doing right now. This is out of my domain, but answers in this general field often depend on what procedures you're using, as there are often multiple ways to do the same thing. – Joe Feb 05 '15 at 17:05
  • @Joe That was me with the close vote. I was suggesting it be migrated to Cross Validated (I think close voting in this way is the appropriate mechanism to do this right?). I felt like he's more likely to get a helpful response there. – Robert Penridge Feb 05 '15 at 18:42
  • @RobertPenridge That's the correct mechanism, but that's only appropriate when the question is explicitly off topic here. You can search [meta] for some discussions of this (and it comes up a lot with [tag:r]), but in general the agreed-on rule is that if it asks in a specific language, it's SO, and if it asks about some generic statistical thing, it's [stats.se]. – Joe Feb 05 '15 at 19:10
  • @Joe Thanks for the links and clarification. Close vote retracted. – Robert Penridge Feb 05 '15 at 19:53
  • Thank you all for the comments and links. I will try to specify my question later on. I know that R provides a lot of possibilies for this task, e.g. package "NP" of Hayfield and Racine (2008), but I would like to solve it with SAS. – Vlada Pleshcheva Feb 06 '15 at 09:35

0 Answers0