If the results of the previous research are available in sufficient detail that you can produce a covariance matrix (or at least, a correlation matrix) that includes all the variables you want in your model, then yes, you can do SEM on that matrix, without (new, raw) data. Otherwise, I believe the answer is no, you can't.

Hello Dr. Kenubih

As noted by Dr. Grandjean, you can conduct SEM analysis by utilizing variance-covariance matrix from previous research. If you were familiar with R, an open-source software in Statistics and Data Analysis, its lavaan library can do the SEM with the matrix, easily.

Enjoy your works.

yes, you can use it.

for further clarification I suggest you to read this article

Article Beyond logistic regression: Structural equations modelling f...