Regressors are linearly dependent

[ygor]

With the following code, i get an error "matrixengine.rb:56:in `initialize': Regressors are linearly dependent (Statsample::Regression::LinearDependency)":

ds = Statsample::Dataset.new({
'a' => [5,5,4,5,5,2,2,4].to_scale,
'b' => [5,2,5,2,5,1,1,4].to_scale,
'c' => [4,5,2,4,5,2,1,4].to_scale,
'd' => [4,5,5,5,5,5,1,5].to_scale,
'e' => [5,4,1,5,1,5,4,2].to_scale,
'f' => [4,5,4,5,5,1,1,5].to_scale,
'g' => [4,2,4,4,5,5,4,1].to_scale,
'h' => [4,1,5,5,5,1,4,5].to_scale,
'i' => [4,2,2,1,4,5,5,5].to_scale
})

lr = Statsample::Regression.multiple(ds, 'a')
puts lr.summary

Any idea what is going on? (Excel does provide a linear regression)

[clbustos]

Ygor:

On matrixengine I test if determinant of correlation matrix is 0 or near to
it. The easy way to override this behaviour is comment out the line 56 on
matrixengine.rb. I will test the engine with your data, anyway.

On Mon, Mar 28, 2011 at 9:40 AM, Ygor <
[email protected]>wrote:

With the following code, i get an error "matrixengine.rb:56:in
`initialize': Regressors are linearly dependent
(Statsample::Regression::LinearDependency)":

ds = Statsample::Dataset.new({
'a' => [5,5,4,5,5,2,2,4].to_scale,
'b' => [5,2,5,2,5,1,1,4].to_scale,
'c' => [4,5,2,4,5,2,1,4].to_scale,
'd' => [4,5,5,5,5,5,1,5].to_scale,
'e' => [5,4,1,5,1,5,4,2].to_scale,
'f' => [4,5,4,5,5,1,1,5].to_scale,
'g' => [4,2,4,4,5,5,4,1].to_scale,
'h' => [4,1,5,5,5,1,4,5].to_scale,
'i' => [4,2,2,1,4,5,5,5].to_scale
})

lr = Statsample::Regression.multiple(ds, 'a')
puts lr.summary

Any idea what is going on? (Excel does provide a linear regression)

Reply to this email directly or view it on GitHub:
#2

Claudio Bustos
Psiclogo
[email protected]

[ygor]

If i comment out the line on 56, it goes wrong a bit further. It appears it cannot perform a linear regression. Any idea why excel can do it?

[clbustos]

On R appears the same problem. The predictors are linearly dependent between them, so R^2 is 1.
So, is an Excel problem not warning about problem in your data...

[ygor]

Ok, thnx a lot for looking into this!

[clbustos]

No problem at all.