Lesson 7: The placebo effect
Evaluate and improve a treatment trial
SkillsDesigning treatment trialsThe placebo effect
Worked example
A fictional clinic wants to show that its new relaxation app reduces anxiety. Forty clients with GAD choose whether to use the app for six weeks; 20 do and 20 don't. The clinic's psychologist, who designed the app, rates every client's anxiety before and after. App users improve by a mean of 9 points and non-users by 3 points, and the clinic claims the app "reduces anxiety by 6 points".
Evaluate the clinic's claim and propose an improved design. (6 marks)
Step 1: Identify the allocation problem
Clients chose their own group. People who choose an app may be more motivated or less severely anxious, a participant variable that could cause the difference instead of the app. Random allocation would spread these differences evenly.
Step 2: Identify the placebo problem
App users knew they were being treated and non-users knew they were not, so part of the 6-point difference could be the placebo effect. A placebo condition (an app that looks similar but has no relaxation training) would separate it.
Step 3: Identify the experimenter problem
The rater designed the app and knew who used it, so experimenter expectancy could bias the ratings. A rater who does not know the groups (a double-blind procedure with a placebo app) would control it.
Step 4: Judge the claim
Because of these confounds, the 6 points cannot be attributed to the app: the claim is not supported.
Step 5: Propose the improved design
Randomly allocate the 40 clients to three groups: the real app, a placebo app and a waiting list. Keep participants and the rater blind to the app condition, use a standardised rating scale, and compare real app with placebo app using an unpaired t-test.
Answer
The claim is not supported: self-selection (a participant confound), no placebo control (expectations) and a non-blind rater who designed the app (experimenter expectancy) could each explain the difference. An improved trial randomly allocates clients to a real app, a placebo app and a waiting list, double-blind for the two app groups, and tests real against placebo.
Try a similar one
A fresh question of the same kind, checked as you answer, with hints and a full solution.
Method
Look for problems with allocation, expectations and who measures the outcome.
Say how each could produce the result without the treatment working.
Make a judgement about the claim.
Propose a design that fixes each problem.
Watch out for
Suggesting "a larger sample" as the only improvement: it improves reliability but not the confounds.