www.cmm.bristol.ac.uk/forum

I wonder if you can help. My colleagues and I have run multi-level mediational analysis in MlWin. Assuming that the values from the output we have are the beta coefficients (we are unsure), is there any way of using the beta and its standard error to generate the unstandardised b so that we can conduct a Sobel test?

We are assuming that a Sobel test is the most appropriate test to run to explore the significance of the medation.

We have centered the data by group (level 2), but from our reading we understand that this does not standardise the data. The manual suggests that the output we have has provided us with the beta, however, we assume that the software package has standardised the raw data as we have not done this manually.

Thanks very much for any help provided.
Nicky

Hello Nicky,

I'm not 100% what you mean by beta and b here. Have you been using SPSS previously? In SPSS the beta column in the output refers to the standardised coefficients. However in MLwiN, the beta values in the output are the equivalent on the b's in SPSS, they are unstandardised coefficients. You could try specifying a single level model in MLwiN and SPSS (or other package) and compare the coefficients to see how they correspond, they should be reasonably close (the standard errors will be different though).

Let me know if I've understood this correctly or if you need anything else. I'm not familiar with the sobel test, what is it you are trying to find?


Thanks,
Mark