A model that is based on a causal theory and partial fit with past data is better than
A model that has no causal theory and has a perfect fit with the past data.
If we have data about a lot of variables then we can create a model with a lot of variables and then use factor analysis.
If we are planning an expensive data collection then we want to be parsimonious.
Theories, background knowledge and values of the researcher can influence what is observed
but we can pursue objectivity by recognizing the possible effects of biases. Human knowledge
is based on challengeable conjectures. we can only know about reality probabilistically and
imperfectly. it is impossible to verify that a belief is true, though it is possible to reject false
beliefs if they are phrased in a way amenable to falsification. It is not simply individual
theories but whole worldviews that must occasionally shift in response to evidence.