Variance calculations in Onyx

3 replies [Last post]
mdr's picture
mdr
Offline
Joined: 02/12/2014

I've been trying to match the variances reported by Onyx with those reported by AMOS. Following the example in the AMOS User Guide, I've created a model with four observed variables (recall1, recall2, place1, place2 in the file Attg_yng.xls file) connected to each other with double-headed arrows to determine the covariances of the variables, as is done in AMOS.

The values displayed by Onyx are all different from those displayed by AMOS. For example, the variance of recall1 in AMOS is 5.79 and 110.85 in Onyx. (Excel's VARP function also gives a value of 5.79.) Similarly, the covariance between recall1 and recall2 is 2.56 in AMOS and 121.2 in Onyx (Excel's COVAR gives 2.56).

I'd appreciate an explanation for these differences. Thank you

MDR

brandmaier's picture
Offline
Joined: 02/04/2010
Missing means?

The variance estimates suggest that you haven't modeled the mean structure.

Onyx 0.9 assumes zero means if you have no explicit mean structure. This is different in Onyx 1.0 (about to be released soon), in which we assume a fully saturated means model as default if no mean triangles are present as this seems to be more intuitive.

brandmaier's picture
Offline
Joined: 02/04/2010
Would you mind sharing your

Would you mind sharing your data file so that we can reproduce this?

Thanks,
Andreas

mhunter's picture
Offline
Joined: 07/31/2009
Alternatives

You could also generate simulated data based on your data by following instructions here (http://openmx.psyc.virginia.edu/wiki/generating-simulated-data) and using the R file at the bottom of that page.

Another possibility is sharing your data privately with brandmaier so it's not posted online but he can replicate your error.