It should be reasonable to use the anxiety sum score as a continuous DV. However, if this variable (or other variables in the model) is not normally distributed (which may be the case for anxiety), and you plan to use maximum likelihood (ML) estimation for your SEM, then I would recommend using a robust/scaled ML estimation method such as the Satorra-Bentler correction or the Bollen-Stine bootstrap method. This will help improve the accuracy of your fit statistics, standard errors, and tests of statistical significance.

Thank you Christian Geiser for your answer. I have checked, all my variables are normally distributed.

Using Total Score as DV in AMOS

Yes, you can use the total score of a variable measured on a binary scale as the dependent variable (DV) in AMOS (Analysis of Moment Structures), but consider the following:

Assumptions and Limitations

- Binary scale: Binary scales are typically analyzed using logistic regression or other techniques suitable for categorical data.

- Total score distribution: The distribution of the total score may not meet the assumptions of normality required for some SEM (Structural Equation Modeling) techniques.

Alternatives and Considerations

- a different analysis techniqueConsider using logistic regression or generalized linear mixed models (GLMMs) for binary data.

- Use a different software: Software like R, SAS, or SPSS may be more suitable for analyzing binary data.

AMOS Specifics

- AMOS can handle categorical data: AMOS can handle categorical data, but it's essential to ensure that the analysis technique and model specification are suitable for binary data.

If you're unsure about the best approach, consider consulting with a statistician or exploring alternative analysis techniques.

Ayesha Akram While treating your 12-item summed binary score for existential anxiety as a continuous dependent variable in SEM is a common and often necessary pragmatic approach, and is arguably justifiable as an observed approximation of a latent continuum (especially with 13 potential values and a sample size of 220), it is crucial that you first rigorously test the summed score's distribution for severe non-normality and then employ a robust estimator (e.g., maximum likelihood with bootstrapping) to mitigate the potential biases introduced by this practice namely, the attenuation of path coefficients due to unmodeled measurement error, which must be explicitly acknowledged as a limitation in your research.