AtlasSession 3L8: Study designp.75–85
Quasi-experimental designs
- Study designs primer
- What is systematic error
- Confounding
- Quasi-experimental designs
Fig. 1
Surgical-site infections per 100 implant operations around a suture policy; days to wound closure in two burns centres
Fig. 1: surgical-site infections per 100 operations, three quarters before and three after triclosan-coated sutures: 6.0, 5.6, 5.2, 3.8, 3.4, 3.0. A dashed line carries the old trend on past the change; the after points sit 1.0 below it. Of the whole drop of 2.2, 1.2 would have happened anyway. A dashed comparison series with no change keeps going the way it was.
- Before the change, each quarter
- falling by 0.4
- Whole drop, before → after (per 100 operations)
- 2.2
- Would have happened anyway
- 1.2
- Extra drop after the change
- 1.0
- Comparison with no change
- keeps going the way it was
Study Timeline · First look
Level 1 · First look · Lecture 8 · p.75–85
Judge a change introduced without randomisation by looking at trends and a comparison group.
Read an interrupted time series and a difference-in-differences design, compute the DiD (ext), and name the assumption each one depends on.
New road, fewer crashes? (the step you are on)
Check whether crashes were already falling before it opened.
A new road opens and crashes fall, but were they already falling? Here, before the new sutures, infections per 100 operations were already falling by 0.4 each quarter.