I'm not sure I fully understand what you are asking. Do you mean "is it OK to run the analyses at the item level?"? If so, then the answer is yes, there are factor analytic techniques for binary and ordinal (item-level) indicators. For example, Mplus uses tetrachoric/polychoric correlations and WLSMV estimation in this case.

Thanks Christian Geiser for your answer.

I am asking, Does having a definite measured scale for latent variable from previous researches is necessary to conduct structural equation modelling?

Structural Equation Modeling consists of two components: a measurement model that includes the "latent" concepts (scales) and a path model that links the causal paths between the variables (both measured and latent). Examining just the measurement model is equivalent to doing to Confirmatory Factor Analysis, while path models can be constructed without any lanes variables using ordinary regression.

The minimum number of observed variables for a latent variables is three.

@Nida Yes for covariance based SEM, one should have at least three observed items under each latent variable (as already mentioned by @Morgan), however PLS-based SEM allows to work with less than three observed items. Using Smart-PLS, you can do CFA and SEM with two observed.

I do agree with David L Morgan , thanks for your answer.

The minimum number of observed variables for a latent variable is three. My question is that how we can determine the number of observed variables for an unobserved/latent variable? Is it necessary to be determined by measured scale identified by any other researcher?

For example, team-building was measured with a six-item scale developed by Potnuru et al., 2019. Does it mean only six observed variables come under team building(latent variable)?

Imran Anwar With two variables, there is only one correlation, so the loadings are by definition the square root of that correlation.

As far as reproducing another scale goes, if you want to rely on the reliability and validity evidence from the original study, then you need to specify the same version of the that scale.

There is no definitive rule that says you have to use the exact same measures as somebody else, but it is certainly helpful if you want to ensure direct comparability of your results to previous work. In any case, you should be able to justify your choice of measures on theoretical/substantive grounds, whether it be the full set of indicators or a reduced subset.

You can use conventional CFA and SEM with just two measures per factor (and the loadings do not have to be set equal within the same factor) provided each factor is substantially correlated with at least one other factor or an observed variable in the model. However, models with just two indicators per factor seem to be more prone to improper solutions (e.g., negative residual variances) than models with three or more indicators.

Thanks Christian Geiser for your answer.

But how one can compare one's observed measures of latent (unobserved) variable with previous researches if he/she used different definition of latent variable on which observed measures are based?