29. A fitted model has a high overall fit statistic. What would you examine before trusting its inference?
HardStatistics & inference
Technical
Show model answer
Check whether the model matches the design and outcome, including dependence, residual patterns, influential observations and variance assumptions. Strong fit to observed data does not establish that the inferential assumptions or scientific interpretation are valid.
Use appropriate diagnostics, alternative models and sensitivity analysis with domain and statistical input. Keep prediction performance separate from causal interpretation, and report important limitations rather than selecting a model solely for an attractive fit statistic.
What interviewers listen for
- Match model and study design
- Inspect diagnostics and influence
- Separate fit from inference