- Decide on change
- Chose a Metric
- review statistics (observational)
- Design the test
- Analyze results
Things to keep in mind:
- Risk to participants
- Benefit - what will the experiment yield
- Choice - for instance: can provide the users the choice to select privacy level
- Privacy - how is the user privcy protected in the data (make data not personally identifiable)
- Define problem and suitable metrics
- Build intuition - measure the change between control and expt & the variability of the metric
- Characterize
- Choose the subject
- Choose the "population"
- Calculate Size of both classes needed
- Calculate the time the experiment needs to run
- Sanity Check
- Choose single metric or multiple metrics
- Calculate change
- Check for statistical significance
- Sign test - Positive/Negative change
- Statistical significance of sign test