Denim regression

[COMPAREMISSING]

DO I have Denim regression in SAS? I saw POC CALIS but I don't know how to use it.

Below some data for which I would like to performs deming regression.

data example;

do id=1 to50;

  x=10 + NORMAL(0);

  y=20 + NORMAL(1);

  output;

end;

run;

Thanks in advantage,

Barbara Pomili - Biostatistician specialist at Novartis vaccines

[PaigeMiller]

Do you mean "Deming Regression"?

[COMPAREMISSING]

Yes sorry I made I big mistake.

[COMPAREMISSING]

I meant DEMING regression

[PaigeMiller]

Google comes up with the answer quickly

http://analytics.ncsu.edu/sesug/2009/CC014.Deal.pdf

[COMPAREMISSING]

I know this macro. I need to know if there is a PROC that perform direcly Deming regression and if yes how to use it.

Thanks,

BArbara

[PaigeMiller]

Well, again, searching the SAS documentation doesn't turn up anything for Deming Regression, and searching that Internet thing doesn't turn up anything other than that macro, so ...

I point out that a very special case of Deming Regression, called Orthogonal Regression, is available in SAS.

Why is that macro not sufficient?

[Ksharp]

Did you see the sixth row of this Paper ?

Though PROC NLP and CALIS in SAS® can accommodate Deming regression models,

[gergely_batho]

I guess you already looked at proc calis here:

Example 29.8 Measurement Error Models: SAS/STAT(R) 13.2 User's Guide

You have to learn one of the languages of proc calis.

I am not able to analyze your generated data, because I don’t know what model (equations) to use. It depends on your model of the measurement system.

Obviously in your generated data there is no correlation between x and y. You can “prove” it using proc calis, but in this simple case proc corr is suitable.

Gergely

[gergely_batho]

I have looked into A SAS® Macro for Deming Regression. One way to analyze the example dataset generated there:

data example;

do myid=1 to 200;

value=UNIFORM(0)*10;

x=value + NORMAL(0);

y=value + NORMAL(1);

output;

end;

run;

proc calis data=example meanstr;

   lineqs

      x   = 0 * Intercept + 1. * Fx  +  e1,

      y   = mu * Intercept +  beta * Fx  +  e2;

  variance e1=1, e2;

  mean Fx;

run;

The result show, that variance of the new measurements ~equal to the old.

Also the slopes of two measurements (old and new) are ~equal (=1).

[COMPAREMISSING]

Thanks a lot this was really helpfoul.

Barbara