Estimand

CategoriesCausal Inference , Study Design , Method

An estimand is a key concept in statistical inference: the quantity of interest that a researcher aims to estimate.

An estimand is a concept in statistics that defines to the quantity of interest that a researcher aims to estimate using statistical methods. In other words, it is the parameter or population value that the researcher wants to learn about from a particular study.

The concept of an estimand is particularly important in causal inference, where the goal is to estimate the causal effect of an intervention or treatment on an outcome variable. In this context, the estimand specifies what the causal effect is, in terms of the population parameter that represents the effect.

The choice of estimand is critical in causal inference, as different estimands can lead to different conclusions about the causal effect. For example, the intention-to-treat (ITT) estimand estimates the causal effect of receiving the intervention, regardless of whether the intervention was actually received or not. On the other hand, the per-protocol (PP) estimand estimates the causal effect of actually receiving the intervention, as opposed to merely being assigned to receive it.

This video by Brady Neal does a great job of explaining the process of defining an estimand and then using it to produce causal models which can be use to estimate effects:

https://www.youtube.com/watch?v=aeOp8LJuzDw

Example Estimand: Average Treatment Effect (ATE)

An Average Treatment Effect (ATE) estimand is a commonly used concept in causal inference that aims to estimate the average causal effect of a treatment or intervention on an outcome variable in a population.

The ATE estimand is defined as the difference between the average outcome of the treated group and the average outcome of the control group, where the treatment status is randomly assigned to the study participants. Many estimands, such as this one, rely on a key assumption, known as SUTVA (stable unit treatment value assumption).

Stable Unit Treatment Value Assumption - SUTVA

SUTVA is a key assumption in causal inference that requires that the potential outcomes of each unit are unaffected by the treatment assignment of any other unit. In other words, the assumption requires that the treatment received by one unit has no direct or indirect effect on the outcome of any other unit in the study population. This assumption is important because it allows for the estimation of the causal effect of the treatment, assuming that the assignment to treatment does not affect other units' outcomes.

In practice, SUTVA may be difficult to satisfy in some contexts, such as when the treatment is a public policy that affects a community or a geographic region, or when there are spillover effects from the treatment that affect other units in the study. In these cases, alternative causal estimands that relax the SUTVA assumption may be more appropriate.

To estimate the ATE, researchers typically use statistical methods such as randomized controlled trials (RCTs) or propensity score matching (PSM) to create a valid comparison group of untreated individuals.

The ATE estimand is often used in policy evaluations, clinical trials, and other studies where the goal is to understand the overall effectiveness of a treatment or intervention in a population. It provides a useful summary measure of the average causal effect of the treatment, which can inform decision-making and resource allocation.

However, it is important to note that the ATE estimand may not be applicable in all situations, and alternative estimands such as the controlled direct effect (CDE) or the natural direct effect (NDE) may be more appropriate in certain contexts.

Related articles
In categories