Limit distribution fitting function to only non-missing values

[Minh2710]

Hello. I have been using proc nlin to fit a pre-determined model to a set of value with relative success using the following line of code:

data example;
	input Notch Rate;
	datalines;
1 0.101750388
2 0.02797293
3 0.007380199
4 0.003547882
5 0.002219785
6 0.000631796
7 0.000507642
8 0.000361024
9 0.000105668
10 0.000315689
11 .
12 .
13 .
14 .
15 .
16 .
;
run;

proc nlin data = example;
  parameters  alpha = 0
              beta = 0;
  model Rate = alpha * exp (beta * Notch);
  output out = exponential predicted = pred_exp sse = sse_exp;
run;

data exponential;
	set exponential;
	format pred_exp Rate percent10.4;
run;

However, as I get further, I need the fitting procedure to use only inputs to the last non-missing value (e.g. in this particular dataset, minimize the sum of squared errors of only up to first 10 observations (Notch = 10), but I also want to generalize the code to use on other datasets).

I really appreciate your help.

[FreelanceReinh · Accepted solution]

Hello @Minh2710,

You can exclude observations with missing Rate values by means of a WHERE statement:

proc nlin data = example;
where n(rate);
...

But PROC NLIN doesn't use those observations anyway in the process of estimating the parameters, see these lines in the output table "Estimation Summary" (obtained without the WHERE statement):

Observations Read                 16
Observations Used                 10
Observations Missing               6

So the WHERE statement doesn't change the model.