Final answer:
A simpler model is chosen over a more complex model when overfitting is a problem, computational resources are limited, or there is a need for easier interpretation. It can also be the right choice in fields where data is hard to obtain, when making less critical decisions, or when analyzing effects of individual variables on complex problems.
Step-by-step explanation:
You would choose a simpler model over a more complex one when overfitting is a concern. Overfitting occurs when a model learns the detail and noise in the training data to the extent that it negatively impacts the performance of the model on new data. This means the model is too complex and captures random fluctuations which are not present in the general data set.
A simpler model might also be preferred when dealing with a large dataset if computational resources or time are limited. A complex model requires more computational power and time to run, which might not be feasible in certain scenarios. A simpler model might also be chosen when there's a need to understand the model fully or explain its predictions to stakeholders who are not experts in the field.
Formal incorporation of previous data is also a way to reduce uncertainty without necessarily increasing model complexity. This approach helps in building upon past scientific efforts, especially in fields where data are challenging to obtain. Furthermore, when the decision to be made is not overly critical, or when the aim is to analyze complex problems ceteris paribus - considering one variable at a time - a simpler model is advisable. Lastly, some information criteria such as BIC tend to favor simpler models, especially when sample sizes are large.