IDK¶

Pre-Survey/Exploratory Designs¶

Sequential approach: simple screening pre-survey experiment first (such as 2-level factorial design), followed by through investigation

Use \(\le 25\%\) of total available data for collection into screening

We want to know: what vars are the most important and over what ranges

Do low-cost pre-survey DOE that focuses on high-priority

Based on this outcome, perform full experiment

Experiment Pointers¶

• Always perform experiment with both trial & control samples

• Always get the raw data; processing should be done by analyst, not data providers

• Every data point should have central tendency & uncertainty associated

• Incorporate all potential uncertainty associated with collecting the data & use Uncertainty Propagation
• Observation: Use robust summary statistics: Median
• Spread/Uncertainty of estimate
  • Standard error, not standard deviation
  • Use non-robust summary statistics
• Every data point fed to model should be iid observation

Data Template¶

For Collection¶

For Modelling¶

We cannot use the collection data directly for modelling as each row is not iid observation. Hence aggregation is required to obtain the central tendency & uncertainty for each iid observation.

Comments