2
$\begingroup$

I have four sets of data grouped by stages (Stage 1-4) with each having their own correlation coefficient. The 4 sets of data has the same dependent and independent variables with 1st n=106, 2nd n=23, 3rd n=89, 4th n=132. My hypothesis is that at stage 4 the correlation would get weaker.

From the numbers alone it supports the idea since the correlation is smaller on the 4th stage than on the 1st, 2nd and 3rd but is there a way to prove it statistically? Or is it just okay to say that there is a weakening in the correlation without having to resort to statistical significance testing? I have read about the Fisher z transform but can it be used to test 4 coefficients? And since I am only comparing the 4th stage to the rest can I just test 1-4 2-4 and 3-4 using Fisher z? Oh and is there a way to do the test using SPSS?

$\endgroup$
  • 1
    $\begingroup$ I would suggest using (strafied on Stage) bootstrapping since it tends to be more robust towards the distribution of your variables. I don't whether it can be done in SPSS, but it is straightforward in R. $\endgroup$ – Erik Aug 5 '13 at 14:30
  • 1
    $\begingroup$ Do you have AMOS? You could test this in AMOS (or another SEM program). $\endgroup$ – Jeremy Miles Oct 4 '13 at 21:35
1
$\begingroup$

I would suggest using the Chow Test for constancy of parameters. Construct an appropriate (common) XARMAX model for each stage and 1 globally (collectively). Generate the F value which tests the equivalence of the model parameters. If you don't have time series your model would be a simple OLS. if however you have time series data and I guess you do then identifying and constructing the XARMAX model night require software like SAS/SPSS or AUTOBOX (which I am involved with).

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ I have read the wiki about time series but I don't think my data are time series, my data are two laboratory readings of different subjects that are grouped according to the disease stage. Can the chow test be used for that kind of data ? $\endgroup$ – Ardo Aug 5 '13 at 23:28
  • 1
    $\begingroup$ Very definitely yes . $\endgroup$ – IrishStat Nov 3 '13 at 21:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.