near singular matrix

[sid85]

Hello,
I'm trying to estimate an equation: freq c sex ag1 ag2 ag3 ag4 ag5 ag6 ag7 ag8 ap1 ap2 ap3 ap4 av1 av2 av3 av4 us1 us2 us3 us4 m1 m2 m3 m4 m5 by count method, each of these variables can take 1 or 0, the problem is that I have near singular matrix and I don't know how to resolve it :( , any solutions ? :D

[EViews Gareth]

Your regressors are probably co-linear.


You solve it by removing the co-linear regressors.


There's been numerous similar posts.

[sid85]

I've looked to the topics but I haven't found my answer: I know that there is collinarity, the problem is : I haven't found the collinear variables, the correlation coefficients are <0.7
Please help, I upload the data if someone can help me

[EViews Gareth]

The correlation between the coefficients doesn't matter, they can still be perfectly co-linear. Try running the following program, for example:

Code: Select all

create u 100 rndseed 1 series y=nrnd>0 series x1=nrnd>0 series x2 = nrnd>0 series x3 = 1-x1-x2 group g x1 x2 x3 show g.cor count y c x1 x2 x3

X1, X2 and X3 are all dummy variables. By construction they are co-linear. Yet the highest correlation between any two of them is only 0.7 ish

[startz]

I've looked to the topics but I haven't found my answer: I know that there is collinarity, the problem is : I haven't found the collinear variables, the correlation coefficients are <0.7
Please help, I upload the data if someone can help me

Remember that no one knows your variable definitions. Have you checked for the dummy variable trap?

(By the way, why upload an excel file instead of an EViews file?)

[sid85]

I tried your code QMS Gareth and I have the same problem: near singular matrix :s
Sorry for the upload, here is the eviews file

[startz]

Your equation is misspecified. You have indeed fallen into the dummy variable trap.

[EViews Gareth]

I tried your code QMS Gareth and I have the same problem: near singular matrix :s

That was the point.