Why Treatment Assignment and Potential Outcomes Are Not Independent in Observational Studies ?

 

 Step-by-Step Explanation:

1. What Are Potential Outcomes?

  • In causal inference, for each individual we define two potential outcomes:

    • Y(0)Y(0): the outcome the person would have if they did not receive the treatment.

    • Y(1)Y(1): the outcome the person would have if they did receive the treatment.

📌 These are hypothetical — for each person, we only get to observe one of them in real life (depending on their treatment assignment).

2. What Is
T?

  • TT is the treatment assignment:

    • T=1T = 1: treated

    • T=0T = 0: control (not treated)

3. What Does "Not Independent" Mean?

  • Saying that the potential outcomes (Y(0),Y(1)) and TT are not independent means:

    The likelihood of receiving treatment depends on something that also affects the outcome.

In math:

(Y(0),Y(1))⊥T

That is: the treatment assignment is related to the outcome you’d have had with or without treatment.

4. 🔄 Why Is This a Problem?

In randomized experiments, treatment is assigned randomly, so:

(Y(0),Y(1))⊥ ⁣ ⁣ ⁣
T(Y(0), Y(1)) \perp\!\!\!\perp T

That’s good — it means treatment is not confounded with outcome.

But in an observational study, people choose treatment (or are selected for it) based on characteristics like:

  • Age

  • Disease severity

  • Socioeconomic status

  • etc.

Those same characteristics may also affect outcomes — so there's confounding.

🧠 Example:

  • Sicker patients are more likely to get an aggressive treatment.

  • But sicker patients also tend to have worse outcomes, even if the treatment is effective.

  • So if we just compare outcomes in treated vs. untreated, we confuse the effect of treatment with the effect of baseline health.

🔑 Why It Matters

If potential outcomes and treatment assignment are not independent, then naïvely comparing treated vs. untreated will give a biased estimate of the treatment effect.

This is the fundamental challenge of causal inference in observational data.

🧪 What Can We Do?

To deal with this lack of independence, methods like:

  • Propensity score matching / weighting / stratification

  • Regression adjustment

  • Inverse probability of treatment weighting (IPTW)

  • Targeted maximum likelihood (TMLE)

  • Instrumental variables

are used to simulate a situation as close as possible to randomization by adjusting for confounders.

✅ Summary in Plain Language

In a randomized trial, treatment is random — it has nothing to do with what the outcome would have been.

But in an observational study, people who get treated are systematically different from those who don't.

As a result, treatment assignment and potential outcomes are not independent, and this creates bias in estimating the true causal effect.

Comments

Post a Comment

Popular posts from this blog

Analysis of Repeated Measures Data using SAS

Image
 1. Basic Concepts of Repeated Measures: In this basic setup of a completely randomized design with repeated measures, there are two factors, treatments and time. Treatment is called the between-subjects factor because levels of treatment can change only between subjects; all measurements on the same subject will represent the same treatment. Time is called a within-subjects factor because different measurements on the same subject are taken at different times.  In repeated measures experiments, interest centers on (1) how treatment means differ, (2) how treatment means change over time, and (3) how differences between treatment means change over time. 2. Four-step procedure for mixed model analysis: Step 1: Model the mean structure, usually by specification of the fixed effects. Step 2: Specify the covariance structure, between subjects as well as within subjects. Step 3: Fit the mean model accounting for the covariance structure. Step 4: Make statistical inference bas...
Read more

Medical information for Melanoma, Merkel cell carcinoma and tumor mutation burden

  Background: Melanoma Also called: malignant melanoma The most serious type of skin cancer. Melanoma occurs when the pigment-producing cells that give color to the skin become cancerous. Symptoms might include a new, unusual growth or a change in an existing mole. Melanomas can occur anywhere on the body. Treatment may involve surgery, radiation, medications, or in some cases chemotherapy. Merkel cell carcinoma  (MCC) it is a very rare ,  highly aggressive, neuroendocrine carcinoma of the skin ;  it has a worse survival probability compared with more common cutaneous malignancies, including metastatic melanoma, and has been termed the “most lethal” skin cancer  as   p atients with metastatic MCC have a historical 5-year overall survival (OS) rate of 18% What Is Tumor Mutational Burden? TMB is defined as the frequency of certain mutations within a tumor’s genes.  To be counted towards TMB, mutations have to alter the protein that is made from a gene. ...
Read more

Four essential statistical functions for simulation in SAS

  PDF function : This function is the probability density function. It returns the probability density at a given point for a variety of distributions. (For discrete distribution, the PDF function evaluates the probability mass function.) CDF function : This function is the cumulative distribution function. The CDF returns the probability that an observation from the specified distribution is less than or equal to a particular value. For continuous distributions, this is the area under the PDF up to a certain point. QUANTILE function : This function is closely related to the CDF function, but solves an inverse problem. Given a probability, P, it returns the smallest value,  q , for which CDF( q ) is greater than or equal to P. RAND function : This function generates a random sample from a distribution. In SAS/IML software, use the  RANDGEN subroutine , which fills up an entire matrix at once.
Read more