When is a model good?

statmodeling.stat.columbia.edu/2009/05/07/bayes_jeffreys/

There is a body of work behind why model parsimony is good, but the reasoning doesn't apply everywhere. When does it not?

From simplystatistics.org/2018/06/04/trustworthy-data-analysis/

• How was the data gathered?
• How were the data processed?
• Sampling frame?
• Is there a reason why the variables might be causally related?

Donald Rubin has two questions he likes to ask any researcher:

• What would you do if you had all the data?
• What were you doing before you had any data?