There are four general ways that patterns emerge in bi/multivariate data.
- A causes B
- B causes A
- X (known or unknown) causes A & B
- Random chance
In a statistical analysis, there is a typical sequence of steps.
- Interview the data – where did it come from? Is it high quality? Does it omit any details? Are omitted details important?
- Check for significant correlation – Are variables related to one another? What happens if you randomly re-distribute one or more variables? Is there a trend? Does it change with re-distribution?
- Check for causality – Does one variable affect the value of another?
- Generalize causality – Can the causality apply to broader contexts? Is causality limited to a certain area, scope, demographic, timeframe, etc?
Each step tends to refine data, questions, etc.. Some analysis attempts may not make it through the series of steps.