A/B Testing

Predictor Variable

自变量

 

Experimental Variable

因变量，实验变量

Control Variable

1. list all potential var
2. if have no data, remove var
3. combine similar var
4. test the correlation between predictor and outcome

Lurking Var? Logic

we thought A -> B, but in fact C->A, C->B. So A-/->B, but C->B.

 

How Long?

Full Cycle

Design Unit

1. Randomized

random treatment unit, control

2. Matched Pair

a treatment unit <-> a similar control unit