Is there a formal test for linearity?

[MarieK]

Using SPSS, in preparation for correlation analysis,

apart from scatterplots, is there a formal statistical test for linearity for two continuous variables?

Someone suggested using Analzye -> Compare means -> Means -> under options, test for linearity?

Is that the best test for this purpose?

Many thanks!

 

[hlsmith]

Look at linear regression and residuals. You can also fit polynomials and see if fit is better or worse.

 

[obh]

Hi,

The white test is testing the homogeneity of variance.
I assume if the variance is homogeneous, and the linear regression model is built with average(residuals)=0 so the data is also linear.

So can we say the white test is also a test for linearity?

 

[Dason]

Definitely not

 

[obh]

Hi Dason,
In one word: Definitely
In two words: Definitely not

I'm sure that you are correct. Can you please help me understand the reason? thanks

 

[obh]

Okay got it. that was a silly question ...
But is there any other non-graphical way? maybe the moving average to reduce the standard deviation?

 

[noetsi]

for linear regression two words Box Tidwell.

this is the only formal test I know of (the only one that has a score associated with it). It has been pointed out that this test does not commonly tell you very much in practice which is why graphical tests are preferred.

Partial residuals are often seen as a better way to do this test but this is not a formal test.

 

[hlsmith]

Reset test

 

[obh]

Thanks, @hlsmith that looks like a very nice and easy test
The methodologic reminds me of the white test