AMOS: "the model is probably unidentified"

What the error means

Identification asks whether there is enough information in your covariance matrix to estimate every free parameter uniquely. When AMOS says the model is probably unidentified, it has detected that one or more parameters cannot be pinned down — many different values fit equally well. AMOS hedges with "probably" because it diagnoses this numerically; the practical message is the same: fix the identification before trusting any estimates.

Common causes

• A latent variable with no scale — AMOS needs either one loading fixed to 1 or the factor variance fixed to 1 for each construct.
• Too few indicators on a factor (a standalone factor needs at least three; two can work only if it correlates with another factor).
• Negative degrees of freedom — more free parameters than the data's known elements, so the model cannot be identified.
• An error term left without a scale (residual paths not fixed to 1).
• Empirical under-identification — a structurally fine model with a near-zero loading or correlation.
• A feedback loop or correlated structure specified without the constraints needed to identify it.

How to fix it

1. For every latent variable, set a scale: fix one indicator loading to 1, or fix the latent variance to 1. AMOS adds the 1's on residual terms by default — make sure you did not delete them.
2. Check that each error/residual term has its regression weight to its indicator fixed to 1.
3. Give each factor at least three indicators; rework two-indicator factors so they correlate with another latent or merge them.
4. Look at the degrees of freedom in the output — they must be zero or positive. Negative df means the model is not identified and you must add constraints or remove parameters.
5. If the structure is fine but estimation still fails, check for near-zero loadings (empirical under-identification) and consider a larger sample.
6. Re-run and confirm the unidentified message is gone and standard errors are produced.

How to avoid it next time

Sketch the model and count parameters versus known covariances before drawing it in AMOS, and give every latent a scale from the start. If you also analyse the model in Explonia's browser-based CB-SEM, it reports degrees of freedom and flags weak loadings, which makes a fragile or under-identified structure easy to catch early. See CB-SEM and the concept page on identification.

Frequently asked questions