# Lavaan in R : some estimated lv variances are negative (higher order model)

I'm trying to fit a higher order model with five correlated secondary order factors and 14 primary order factors, but I receive following warning in R using cfa() in lavaan package in R:

lavaan WARNING: some estimated lv variances are negative.

What are (overall) possible reasons for these?

(That's a BIG model - what is your sample size?)

This means that your model is mis-specified. Your model is trying to get the factor loadings right, and the variances right, and the covariances right.

Here's my attempt at a simple example. You have four variables, A, B, C and D. A and B load on F1, C and D load on F2.

A and C are highly correlated. B and D correlate zero.

The model cannot find estimates for the parameters which satisfy everything. Any model that allows A and C to correlate also allows B and D to correlate. In an attempt to fix this, it makes values nonsensical (like negative variance and standardized loadings > 1).

I suspect your model fit is also poor.

• Thank you! Sample size is 150. By looking at one of the second order factors and corresponding primary factors I realized it is indeed a problem with the model. Also I shouldn't fix a negative variance of -0.312 (unstand.) to zero as it isn't small enough, right? p-value is .116 – InsertMon Feb 5 at 21:42
• That's a huge model for a small sample size. It's almost certain that you will run into problems. – Jeremy Miles Feb 5 at 22:46