Prev: Import from Access
Next: testing for random effects in proc mixed
From: BruceBrad on 21 Nov 2006 21:05
I have the following code. I want to test for the significance of the
different transforms of income as a group.


data test;
set input.desc1;
income2 = income*income;
incomelog = log(income+1);
run;
proc reg;
model alrndom = income income2 incomelog /corrb collin;
test income, income2, incomelog;
quit;run;



I get an error message:
ERROR: The TEST is not consistent or has redundant column.

What's going on here? There doesn't seem to be undue collinearity as
far as I can see. The incomelog variable is not significant, and is
thus correlated with the intercept, but not extremely so.

The regression output is below.


Number of Observations Read 4890
Number of Observations Used 4601
Number of Observations with Missing Values 289


Sum of
Mean
Source DF Squares
Square F Value Pr > F

Model 3 17258
5752.62073 61.86 <.0001
Error 4597 427508
92.99710
Corrected Total 4600 444766


Root MSE 9.64350
R-Square 0.0388
Dependent Mean 100.36552
Adj R-Sq 0.0382
Coeff Var 9.60838


Variable Label DF Estimate Error
t Value Pr > |t|
Intercept Intercept 1 97.14704 2.67290
36.35 <.0001
Income 1 0.00367 0.00067637
5.43 <.0001
Income2 1 -3.16237E-7 8.755566E-8
-3.61 0.0003
IncomeLog 1 -0.12193 0.46761
-0.26 0.7943


Correlation of Estimates
Variable Label Intercept Income
Income2 IncomeLog
Intercept Intercept 1.0000 0.7767
-0.5564 -0.9908
Income 0.7767 1.0000
-0.9025 -0.8484
Income2 -0.5564 -0.9025
1.0000 0.6350
IncomeLog -0.9908 -0.8484
0.6350 1.0000

From: Peter Flom on 22 Nov 2006 12:04
>>> BruceBrad <BruceBrad(a)INAME.COM> 11/21/2006 9:05:46 pm >>> wrote
<<<<
I have the following code. I want to test for the significance of the
different transforms of income as a group.


data test;
set input.desc1;
income2 = income*income;
incomelog = log(income+1);
run;
proc reg;
model alrndom = income income2 incomelog /corrb collin;
test income, income2, incomelog;
quit;run;



I get an error message:
ERROR: The TEST is not consistent or has redundant column.

What's going on here? There doesn't seem to be undue collinearity as
far as I can see. The incomelog variable is not significant, and is
thus correlated with the intercept, but not extremely so.

The regression output is below.
>>>

snip

I am not sure exactly what is going on, as I don't often use the TEST
statement. However, have you tried centering or possibly standardizing
the income variable before transforming it?

I have often found this helpful with similar problems

HTH and Happy Thanksgiving


Peter
From: David L Cassell on 23 Nov 2006 01:27
BruceBrad(a)INAME.COM wrote:
>
>I have the following code. I want to test for the significance of the
>different transforms of income as a group.
>
>
>data test;
>set input.desc1;
>income2 = income*income;
>incomelog = log(income+1);
>run;
>proc reg;
> model alrndom = income income2 incomelog /corrb collin;
> test income, income2, incomelog;
> quit;run;
>
>
>
>I get an error message:
>ERROR: The TEST is not consistent or has redundant column.
>
>What's going on here? There doesn't seem to be undue collinearity as
>far as I can see. The incomelog variable is not significant, and is
>thus correlated with the intercept, but not extremely so.
>
>The regression output is below.
>
>
>Number of Observations Read 4890
>Number of Observations Used 4601
>Number of Observations with Missing Values 289
>
>
> Sum of
> Mean
> Source DF Squares
> Square F Value Pr > F
>
> Model 3 17258
>5752.62073 61.86 <.0001
> Error 4597 427508
> 92.99710
> Corrected Total 4600 444766
>
>
> Root MSE 9.64350
>R-Square 0.0388
> Dependent Mean 100.36552
>Adj R-Sq 0.0382
> Coeff Var 9.60838
>
>
>Variable Label DF Estimate Error
> t Value Pr > |t|
>Intercept Intercept 1 97.14704 2.67290
> 36.35 <.0001
>Income 1 0.00367 0.00067637
> 5.43 <.0001
>Income2 1 -3.16237E-7 8.755566E-8
> -3.61 0.0003
>IncomeLog 1 -0.12193 0.46761
> -0.26 0.7943
>
>
> Correlation of Estimates
>Variable Label Intercept Income
>Income2 IncomeLog
>Intercept Intercept 1.0000 0.7767
>-0.5564 -0.9908
>Income 0.7767 1.0000
>-0.9025 -0.8484
>Income2 -0.5564 -0.9025
>1.0000 0.6350
>IncomeLog -0.9908 -0.8484
>0.6350 1.0000

If all you want is a test of the three variables as a group, then
the omnibus F test does that without a TEST statement. Since
your F is highly significant, you have statistical significance.

But do you really want to have these three variables all in your
regression at once, or is this just experimenting with data?

As for the problem that is giving you an error, try using the /PRINT
option to get the intermediate steps in the computations so you
can see what is going on.

HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Fixing up the home? Live Search can help
http://imagine-windowslive.com/search/kits/default.aspx?kit=improve&locale=en-US&source=hmemailtaglinenov06&FORM=WLMTAG
From: BruceBrad on 7 Dec 2006 22:35
> I am not sure exactly what is going on, as I don't often use the TEST
> statement. However, have you tried centering or possibly standardizing
> the income variable before transforming it?

Yes, this solved it. I was testing the effect of income (mean approx
1300) and income squared. Dividing income by 100 solved the problem.
Obvious the test algorithm doesn't handle large numbers well.

 | 
Pages: 1
Prev: Import from Access
Next: testing for random effects in proc mixed