encouragement design is a type of randomized controlled trial. However, instead of randomizing the intervention, the researcher randomizes the encouragement to receive the treatment: some individuals or groups are encouraged to participate in a treatment, and some are not.
Usage
- when researchers have less control over the subjects’ compliance.
- when treatment is known to be beneficial to at least some individuals and unlikely to be deleterious to anyone (because flatly assigning people to control can be unethical)
Characteristics
In order for the “encouragement design” to work, the encouragement must significantly increase the likelihood that units will follow through with what they are being encouraged to do.
Key features of an encouragement design may include:
- Randomization: Participants are randomly assigned to either a treatment group, where they receive encouragement or incentives, or a control group without such encouragement. Randomization helps control for potential confounding variables, allowing researchers to make causal inferences about the impact of encouragement.
- Encouragement Mechanism: The encouragement can take various forms, such as providing information, offering incentives, or creating a supportive environment. The goal is to motivate participants to engage in the desired behavior or treatment.
- Outcome Measurement: Researchers measure the outcomes of interest to assess the effectiveness of the encouragement. This could include changes in behavior, adoption of a new practice, or other relevant outcomes.
- Causal Inference: The random assignment helps researchers attribute any observed differences in outcomes between the treatment and control groups to the encouragement rather than other confounding factors. This is crucial for establishing a causal relationship.
Type
- Compliers: Those who follow the rule - Those in the encouraged group who are treated, and those in the non-encouraged group who are not receiving the treatment
- Defiers (non-complier): Those who defies the rule - those in the encouraged group who don’t receive treatment, and those in the non-encouraged group who are treated
- Always takers: Always take the treatment, regardless of the encouragement.
- Never takers: Never take the treatment, regardless of the encouragement. Usually assumed as none in the encouragement design
Analysis
- as-treated analysis compares those who received treatment with those who received control, ignoring treatment assignment.
- Per protocol analysis compares people who were assigned to and received treatment with those who were assigned to and received control, i.e., compares those who appeared to comply with the protocol, ignoring observe true non-compliers
- intention-to-treat effect: represent the effect of treatment assignment on the outcome (regardless of whether the units follow the treatment or not)
We can estimate three treatment effects
- ITT: measures the average effect of encouragement on treatment outcomes, for the whole sample
- LATE: measures the impact of treatment on outcomes, for the compliers,
- Compliance rate:
- The effect of encouragement (vs. non-encouragement) on whether or not the treatment happens, for all units. To calculate this, we would take the take-up rate (% got treatment) in the encouraged group and SUBTRACT the take-up rate (% got treatment) in the non-encouraged group
Local average treatment effect (Complier average causal effect)
Since treatment is not randomized, it is difficult to calculate an unbiased average treatment effect. However, we can still estimate the effect of the treatment by exploiting a randomized encouragement design, calculate the average treatment effect among compliers (LATE)
where Compliance Rate = Fraction of Subjects that were treated in the encouraged group - Fraction of Subjects that were treated in the non-encouraged group.
If the compliance rate is 100%, LATE = ITT, we have perfect compliance. Everyone who gets encouraged receives the treatment, and everyone who isn’t encouraged doesn’t receive the treatment. Then encouragement is the marker for treatment, and we can calculate the average treatment effect as LATE = ATE
There are a few important assumptions that need to hold for the LATE to give an unbiased estimate.
- Encouragement cannot make subjects less likely to receive the treatment. This is often a reasonable assumption, but needs to be considered carefully on a case-by-case basis.
- Encouragement is in fact randomly assigned (or is “as good as random”)
- Encouragement has no direct effect on results, except via increasing the probability of receiving treatment. This is why encouragement should be kept as simple as possible. For example, if encouragement takes the form of a training session that could have its own effect on results in addition to encouraging take-up of treatment, then this assumption would be violated.
Difference between