Marginal Structural Model (MSM)
- Estimates the average causal effect (ACE) of an intervention on a population.
- Designed to control for confounders whose effects vary over time.
- Applicable to data from randomized controlled trials and observational studies.
Definition
Section titled “Definition”A marginal structural model (MSM) is a statistical method used in causal inference that allows researchers to estimate the effects of an intervention on a population. This is often done by using data from randomized controlled trials, but can also be applied to observational data.
MSMs are used to estimate the average causal effect (ACE) of an intervention on a population by taking into account the time-varying effects of confounders on the outcome of interest.
Explanation
Section titled “Explanation”Confounders are variables that influence both the intervention and the outcome; if they are not properly controlled for, they can bias estimates of the intervention’s effect. MSMs address this by explicitly accounting for confounders whose relationships with the outcome change over time.
The typical approach described in the source material is:
- Specify variables likely to be confounders (for example, diet, exercise, stress levels).
- Use statistical techniques to estimate the average causal effect of the intervention while taking into account the time-varying effects of those confounders on the outcome.
MSMs can therefore reduce bias in estimated intervention effects that arises from time-varying confounding.
Examples
Section titled “Examples”Medication effect on blood pressure
Section titled “Medication effect on blood pressure”A researcher randomly assigns participants to receive either a new medication or a placebo. After several weeks of treatment, the researcher measures participants’ blood pressure and compares the results between groups. Other factors such as diet, exercise, and stress levels may influence blood pressure; to control for these confounders the researcher specifies them and uses an MSM to estimate the average causal effect of the medication while accounting for their time-varying effects.
Educational program effect on student achievement
Section titled “Educational program effect on student achievement”A researcher randomly assigns schools to implement a new educational program or continue their current curriculum. After one year, the researcher measures student achievement scores and compares results between the two groups. Factors such as parental education level, socioeconomic status, and prior achievement may influence outcomes; the researcher specifies these confounders and uses an MSM to estimate the average causal effect of the program while accounting for their time-varying effects.
Use cases
Section titled “Use cases”- Estimating average causal effects in randomized controlled trials.
- Estimating average causal effects in observational studies where confounders vary over time.
Notes or pitfalls
Section titled “Notes or pitfalls”- Confounders that are not properly specified and controlled for can bias the estimated effect of an intervention.
- Time-varying confounding is a central concern that MSMs are designed to address.
Related terms
Section titled “Related terms”- Average causal effect (ACE)
- Confounders
- Randomized controlled trials
- Observational data