Hello,
It could be great if anyone could share with me how to calculate the confidence interval for partial correlation using SPSS or JASP.
Thanks!
Muhammad Badarnee There is at least one (somewhat complicated) way in which this can be done in SPSS that I can think of. (Maybe somebody else knows an easier way.)
The partial correlation is equal to the zero-order correlation of regression residuals R1 and R2 obtained from two linear regressions (partialling out the third variable Z from the two variables X1 and X2 for which you wish to compute the partial correlation). You can save these residuals in SPSS by using the REGRESSION option twice (once to partial Z from X1 and once to partial Z from X2). You can then correlate the residuals R1 and R2 from those two regressions. This (also) gives you the partial correlation. To my knowledge, SPSS does not give confidence intervals for correlations directly. However, you can obtain them indirectly via the REGRESSION option (see below).
To obtain the confidence interval for the residual (=partial) correlation, you can run the correlation between the residuals "indirectly", that is, by again using the linear regression function, this time regressing one residual on the other, e.g., R1 on R2. The reason is that for a bivariate regression, the standardized regression coefficient BETA is equal to the Pearson product-moment correlation. In the case of regression residuals, BETA is equal to the partial correlation.
The REGRESSION option in SPSS allows you to print confidence intervals (CIs) for B (the unstandardized regression slope coefficient). If you z-standardize your residuals (or save standardized residuals to begin with) before running the regression, B will be equal to BETA (the partial correlation; see https://www.youtube.com/watch?v=-dSoWqDyT4E) so that you can obtain the CIs for the partial correlation that way.
Slightly complicated, but it should work...
I almost always agree with Christian Geiser's advice, and I frequently recommend his responses. But in this case, I disagree with the method described in the YouTube video he mentioned. Why? Because it yields CIs for rho that are symmetrical. The usual method for computing the CI for rho, on the other hand, yields a symmetrical CI only when the observed correlation = 0. Otherwise, the interval is asymmetrical--see the 3rd slide in the attached PDF. While it is true that the CORRELATIONS command in SPSS still has no option to show confidence intervals, there are other options, including the (Python) extension command STATS CORRELATIONS. One of the attached PDFs has instructions on how to find and install it. And sorry for disagreeing with you, Christian. ;-) HTH.
Bruce Weaver Thanks for pointing this out. I hadn't thought about it.
Christian Geiser & Bruce Weaver thank you for your responses. I'll try to get the CI using your suggestions. Thanks again.
Bruce Weaver Hey again and thank you for your help.
Unfortunately, the links in the pdf file titled "IntroBiostats..." are not available. Kindly let me know if there is a way to reach them.
Thanks.
Hi Muhammad. Sorry about that. Not sure why the hyperlinks did not come through in the PDF. Here they are.
The STATS CORRELATION extension command
- http://spssx-discussion.165.s1.nabble.com/New-Extension-Command-STATS-CORRELATION-td5732755.html
The !rhoCI macro
- https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/spss/my-spss-page/rhoci
The syntax in this IBM Tech Note
- https://www.ibm.com/support/pages/confidence-intervals-correlations
See this page for help on how to find, install and update user-written programs for Stata
- https://www.ssc.wisc.edu/sscc/pubs/4-16.htm
Stata users can type ssc describe corrci for info about the corrci package (and to install it).
HTH.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS