AtlasSession 3L7: Systematic errorsp.13–21, 74–75
Confounding
Fig. 1
MSC cultures reaching target yield at passage 2, adipose- vs marrow-derived (cultures, n = 120)
Fig. 1: 120 cultures, 60 fat and 60 marrow. 43 of 60 fat (72%) and 23 of 60 marrow (38%) reached target yield at passage 2.
- Fat cultures reaching the target
- 43 of 60, about 72%
- Marrow cultures reaching the target
- 23 of 60, about 38%
- From younger donors
- not revealed yet (shown once donor age is revealed)
- Younger donors
- not split yet (shown once the cultures are split by age)
- Older donors
- not split yet (shown once the cultures are split by age)
- Within each age group
- not split yet (shown once the cultures are split by age)
Causal Diagram Lab · First look
Level 1 · First look · Lecture 7 · p.13–21, 74–75
See how a hidden third factor can make one option look better when it isn't.
Spot a prognostic variable that also predicts the intervention received, and show how stratifying on it changes the crude estimate.
Ice cream and sunburn (the step you are on)
Both rise in summer; sunshine drives both, ice cream doesn't cause burns.
Ice cream sales and sunburn rise together because sunshine drives both; ice cream doesn’t cause burns. Here fat-derived cultures reach the target more often (43 of 60, about 72%) than marrow-derived ones (23 of 60, about 38%). Could something hidden drive that too?