Hi Jennifer

I believe you mean 12 items measuring a single (latent) construct.

Generally, when AMOS reports in such way, there's likely you have encountered Heywood case in your dataset. It's very likely that this is caused by outliers or extreme violation of normal distribution or sometimes, multicollinearity.

A guessing game is that, you may want to run factor analysis using with:

1. Extraction method: Maximum Likelihood, rotation: any oblique (e.g. direct oblimin).

2.  Extraction method: Principal axis factoring, rotation: any oblique (e.g. direct oblimin).

Look at the factor loading, scan for any 'impossible' value, i.e. loading >1.0

If there's any, almost certainly your dataset suffers Heywood cases. 

You can later check and treat for outliers which is very time consuming. Alternatively, you can opt for PLS-SEM instead of covariance-based SEM (e.g. AMOS).

The link below gives you an idea how PLS-SEM works. 

Best wishes

Saiyidi

Chapter Partial least square in a nutshell | Saiyidi MAT RONI 2 0 1 4

Hi Jennifer

I believe you mean 12 items measuring a single (latent) construct.

Generally, when AMOS reports in such way, there's likely you have encountered Heywood case in your dataset. It's very likely that this is caused by outliers or extreme violation of normal distribution or sometimes, multicollinearity.

A guessing game is that, you may want to run factor analysis using with:

1. Extraction method: Maximum Likelihood, rotation: any oblique (e.g. direct oblimin).

2.  Extraction method: Principal axis factoring, rotation: any oblique (e.g. direct oblimin).

Look at the factor loading, scan for any 'impossible' value, i.e. loading >1.0

If there's any, almost certainly your dataset suffers Heywood cases. 

You can later check and treat for outliers which is very time consuming. Alternatively, you can opt for PLS-SEM instead of covariance-based SEM (e.g. AMOS).

The link below gives you an idea how PLS-SEM works. 

Best wishes

Saiyidi

Chapter Partial least square in a nutshell | Saiyidi MAT RONI 2 0 1 4

Thank-you Siayidi - I'll go looking for my Heywood cases again

Jennifer, 

take a look on this website which contains the best explanations.

http://www2.gsu.edu/~mkteer/npdmatri.html

Best,

Holger

Thank-you Holger

If you have outliers, you could also consider deleting anyone 3sds > M.  Also, stata has a nifty program called g ladder which checks the distribution and tells you the best way to normalize it.  Good discussion!

Thanks Carolyn - I have already deleted outliers greater than 3sds, but I will have a look at the g ladder option. Thanks

Dear Jennifer,

In additional to the excellent contributions by Saiyidi, Holger and Carolyn, please find attached the following reference on AMOS and Structural Equation Modelling (SEM) (Please see Page 31) and it further provides detailed explanations and interpretations to the question that you posed on ResearchGate. In fact among the responses included are to the same query “The following covariance matrix is not positive?”, and a number of strategies or suggestions to rectify the problem are outlined. It would be worthwhile to read through. In summation, as pointed out by Saiyidi, it’s mostly issues related to ‘multi-collinearity’ among the variables. Among the suggested remedies includes the following:

1. Removing highly correlated items (check for values of r > = 0.85)

2. Also as previously suggested, using ULS rather than MLE estimation might be the best way forward.

Hope you will find the attachment useful

Nicholas

Thanks Nicholas - I'll look into the ULS option

These are all great suggestions.  I hear that using EQS or Mplus are better programs for SEM.  Any thoughts on these?

Hi,

you should give R (package lavaan) a try (www.lavaan.org). It is very similar to Mpuls and its free :)

Best,

Holger

R is a bit hard to learn to use as it is old dos format. Many of my colleagues have switched to R rather than having to pay the annual SPSS fee for AMOS.  Also, some people claim there are problems in how AMOS conducts its calibration of effects.

Interesting thoughts thanks Mary