How do you use a scanning electron microscope to measure mental health?

Hi Oluyinka,

actuallly, you should avoid using subscales, parcels or composites but rather try to model the relationships between observed and latent variables. Sometimes, the variables are of no special interest (e.g., as controls) - but if they are focal variables, mashing together observed variables often means mashing together different empirical enties (ie. latent variables) which a) clouds the ontological aspect of the composite (ie. is it one thing or several?) and b) the interpretation of effects stemming from this composite.

Best,

Holger

Hi Oluyinka,

  In SEM, the subscales would be considered indicators. When placed on a single factor, they create a latent variable. That latent variable (and hence an individuals place on the latent variable... or composite as afforded by your model) can then be used to predict other things you care about or be predicted by other variables. Common software for doing this is: AMOS, EQS, LISREL, Mplus, SAS (has some capabilities). R also has SEM capabilities and is free.