Hello Youness,

"Manifest" variables are observed/observable scores, directly measured. These can be individual responses (e.g., "How much pain do you feel right now, on a 0-10 scale?"), or composite scores, formed by summing, averaging, or otherwise combining individual responses on related items into some overall value.

In general, composite scores are more likely to have better reliability than individual item/question responses, and, as a result, more likely to have better validity for representing the variable of interest. Since most instruments used with human responses have measurement error, the composite scores are likely to be a better choice than individual item scores. The affiliation of individual item responses should be checked explicitly, of course, prior to settling on any method to combine item responses into a single score.

Why do you so frequently see individual responses (even Likert-type response scales) used as manifest variables? Because it's so easy to do (and the researcher then doesn't have to engage in factoring or some other step in order to worry about how to form a composite.

Good luck with your work.

Thank you Dr. David Morse and Dr. Erling Solheim, your answers are very helpful. I still wonder if the researcher has the choice to analyze items individually or construct composite scores? Dr. David Morse said that composite scores have better reliability and validity, but how to select items for composite scores?

Are there any criteria to avoid subjectivity?

Thank you