topic ERROR: The TEST is not consistent or has redundant column. in Statistical Procedures

ERROR: The TEST is not consistent or has redundant column.

Demographer — Mon, 20 Nov 2017 10:57:32 GMT

Hello,

I run linear regressions for different groups and I want to test if the parameter is statistically significant from 1 rather than 0, so I add a "test" statement, such as:

proc reg data=work.mincer plots=none;
model CEp=CEIp;
test CEIp=1;
by EU28 sex;
run;

The variables EU28 and sex have 3 categories each. For some case, there is no problem, but in some others, I don't get the output for the test. I get this error message:

ERROR: The TEST is not consistent or has redundant column.

I find nothing in the documentation about this issue. What does that mean?

Re: ERROR: The TEST is not consistent or has redundant column.

SAS_Rob — Mon, 20 Nov 2017 13:13:18 GMT

This is usually indicative of a scaling problem with your data. When this happens the standard errors for some of your parameter estimates become very small and the chi-square test produced by the TEST statement is unreliable.

I would try standardizing the data first.

proc stdize data=mincer out=new;
run;

proc reg data=work.new plots=none;
model CEp=CEIp;
test CEIp=1;
by EU28 sex;
run;

Re: ERROR: The TEST is not consistent or has redundant column.

Demographer — Mon, 20 Nov 2017 15:00:33 GMT

It fixes the error message, but parameters are different and as so, I'm not really sure if the interpretation is the same.

These are Mincer-Zarnowitz regressions, that try to seek if there is a systematic and significant over/under-estimation between two sets of data (CEp and CEIp).

When the parameter is, for instance, 1.10 and statistically different from 1, it should mean that there is a systematic overestimate by 10%. When I use proc stdize on the dataset, I'm not sure that I can still interpret parameters this way.

Re: ERROR: The TEST is not consistent or has redundant column.

SAS_Rob — Mon, 20 Nov 2017 15:35:02 GMT

Standardizing simply removes the units associted with the variables...not sure though how that would change the interpretation for your model.

Either way the scale of the variables is causing the issue, so you may need to find a way to rescale the variable in a way that makes sense.