Can a commitment contract promote weight loss? A field experiment in London

Participants were randomised into two groups: comparison and treatment.The treatment was the offer of a commitment contract that they were invited to personalise by writing in their names, signature, and date. The contract was designed to have a certain degree of visual gravitas and formality, on card of A5 size that could be carried around in a handbag or satchel, or stuck on a fridge or wall. They were advised to keep it somewhere they would see it on most days, and to discuss it with friends and family if they wished but not with the other group members. Comparison group members completed a baseline survey and returned to the weight loss group, and asked to continue with the programme as normal.

2016-01-20

2016-01-31

The primary outcome variable is change in weight at the end of the Shape Up programme. Other outcome variables of interest include:
• Attendance and completion rates over the programme (%).
• Whether there was any change in food and exercise behaviours.

The primary outcome variable weight loss is calculated as (start weight) – (end weight), with a positive number indicating weight loss. The earliest possible weight reading is taken, which is sometimes the second or third group meeting if participants were late to register for the programme. The final weight reading is taken from week 7 onwards. Participants with missing outcome data after week 7 were contacted to gather follow-up data. Some were not reachable, and interim data suggests an attrition rate of 20%, which is not out of line with published weight loss studies.

I designed an experiment to run in collaboration with the ‘Shape Up’ weight loss programme, run by Camden Council’s Active Health team across various sites in North London. Shape Up is a 12-week, group-based programme that sets a 5% weight loss target for all clients. The group meets every week to discuss a different aspect of weight loss and lifestyle. Class tutors measure and record weight, facilitate group discussion, and coach individuals on their food journal and self-reflection.

The target population is the Active Health team’s client base. These individuals have been screened for eligibility based on their current BMI and home postcode, and might have been put forward for the programme either through a GP or self-referral. These same eligibility criteria apply to participation in the field experiment, and did not change during the trial. Under a normal Shape Up programme, the population have made no financial commitment, but arguably may be conscious of a mild reputational commitment to the group tutor.

The registration process involved a baseline survey with questions on lifestyle, wellbeing and health motivation using the 19-item scale from the Healthy Foundations Segmentation tool - a social marketing model intended to identify which of 5 health motivational groups an individual belongs to.

Randomisation done in advance by a computer using Stata 12.0. In some circumstances client lists were not shared in advance, and in this case a simple rule was applied based on the sum of digits in their date of birth to assign them to either treatment or comparison group.

Individual

No

N/A

200

Aiming for 190-200 participants based on ex ante sample size calculations.

Sample size calculations (for individual randomisation designs) are reported below in 3 scenarios using weight loss in lbs as a main outcome variable. The Baseline, Moderate and High weight loss scenarios relate to there being a 1lb, 1.5lb, and 2lb per week difference between control and treatment groups. Each specifies 0.9 power and a two-sided test. I also test the effects of low or high standard deviation assumptions, using 2lb and 3lb respectively. Baseline (1 lb weight loss difference) 44 (low SD 1 lb) 190 (med SD 2 lbs) 380 (high SD 3 lbs) Moderate (1.5 lbs weight loss difference) 20 76 190 High weight loss (2 lbs weight loss difference) 12 44 96 Calculations suggest a sample size of 190 is needed in an individual randomised design in the baseline weight loss scenario with moderate standard deviation across the participants. These sample size calculations carry a degree of uncertainty, as they require estimates of the mean and standard deviation of weight loss for which data is not readily available; and they assume a full sample and no problematic attrition.