AtlasSession 3L7: Systematic errorsp.34–37, 80–81
Classification of interventions
- Study designs primer
- What is systematic error
- Classification of interventions
Fig. 1
Implant patients by exposure as recorded and outcome: true and misrecorded tables (patients)
Fig. 1. Two groups of patients, one dot each. Exposed: 20 with the outcome out of 100. Unexposed: 10 with the outcome out of 100. Lit: everyone in each group. Dividing gives 2.00 times; the names for these are risk and risk ratio.
- Exposed: with the outcome
- 20 out of 100
- Unexposed: with the outcome
- 10 out of 100
- Dividing (risk ratio)
- 2.00 times
- Subtracting (risk difference)
- about 10 more in every 100
- Odds ratio, as recorded
- 2.25
- Risk ratio, as recorded
- 2.00
- Misclassification
- none
2×2 Workbench · First look
Level 1 · First look · Lecture 7 · p.34–37, 80–81
See how putting people in the wrong exposure group can blur or fake a link.
Explain how misclassifying exposure biases results: non-differential usually towards the null, differential (e.g. recall bias) in either direction.
Swapped labels (the step you are on)
If some cola cups are tagged 'lemonade', the two drinks' scores look more alike than they are.
Tag some cola cups ‘lemonade’ and the two drinks seem more alike. Record 1 in 5 smokers as non-smokers (the table as recorded) and gum disease around their implants looks 2.00 times as common, not 2.00 times.