Explore one approach to keeping model metadata during the modeling process for documentation of rationale for modeling decisions.
- [Instuctor] Hi there.…Welcome to the next section, where I will introduce you…to documenting model metadata.…In the previous course,…we went through what curation files are,…and we revisit the concept here.…When it comes to remembering what you did with modeling,…just saving the code is not good enough.…The modeling process already has you making model…after model, removing covariates, then adding covariates,…and then removing them again.…And suddenly, you are at your final model.…But then you ask yourself, why that model?…Did I really try all the models I was supposed to try?…What if someone on my research team asks me why…I removed a certain covariate, or why I left one in?…Panic sets in as you ask, "What was I thinking?"…which is why you need model metadata,…to keep track of what you were thinking.…
When making metadata for your models,…you have to imagine yourself running a model,…which tends to go pretty quick,…then pausing to document that model in metadata.…And you don't want that to be too much work,…
- Differentiate between modular code and spaghetti code and explain when to use each.
- Explain the data-set transformation approach.
- Assess the right time to remove identifiers from the data set.
- Cite the considerations for categorical outcomes.
- Recognize that with large data, even small differences are statistically significant.
- Determine when using a stepwise model is appropriate.