[go: up one dir, main page]
Scribd
Upload
44 views2 pages

Introduction Causal Inference

1. This document introduces causal inference and the goal of estimating causal effects from observational data. It discusses examples of causal questions and the fundamental challenge that we only observe one potential outcome for each unit. 2. Key concepts are introduced, including treatment and control conditions, potential outcomes, sample average treatment effect (SATE), population average treatment effect (PATE), and the stable unit treatment value assumption (SUTVA). 3. Methods for estimating causal effects are outlined, including using a difference-in-means estimator to calculate the SATE by taking the difference between average outcomes under the treatment and control conditions. Conditions for when this estimator provides an unbiased estimate of the true causal effect are discussed further in L

Uploaded by

denniswalleck
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
44 views2 pages

Introduction Causal Inference

1. This document introduces causal inference and the goal of estimating causal effects from observational data. It discusses examples of causal questions and the fundamental challenge that we only observe one potential outcome for each unit. 2. Key concepts are introduced, including treatment and control conditions, potential outcomes, sample average treatment effect (SATE), population average treatment effect (PATE), and the stable unit treatment value assumption (SUTVA). 3. Methods for estimating causal effects are outlined, including using a difference-in-means estimator to calculate the SATE by taking the difference between average outcomes under the treatment and control conditions. Conditions for when this estimator provides an unbiased estimate of the true causal effect are discussed further in L

Uploaded by

denniswalleck
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Introduction to Causal Inference

March 28, 2018

1 Introduction
• In social sciences, we care about causal claims. For example,
1. Does Chinese censorship change behavior?
2. Do GOTV efforts impact turnout?
3. Does speaking a dominant language increase lifetime earnings?
• In order to answer these questions, we need to know a theory of causality.
• Causality is an action (treatment, manipulation, intervention) applied to a unit

• For Simplicity: we consider two conditions, (1) Treatment and (2) Control
• Unit-level Causal Effect:
Difference in outcomes for a given unit between treatment and control conditions

1. Example 1: My headache went away because I took Tylenol.


2. Example 2: My earning is higher because I went to college. (Multiple Versions Problem).
3. Example 3: She has long hair because she is a girl. (Manipulation Problem).

1.1 Mathematical Notations


• Sample: i ∈ {1, 2, . . . , N }.
• Treatment: Ti ∈ {0, 1}.

• Propensity Score πi : Pr(Ti = 1) or more in general, Pr(Ti = 1 | Xi ).


• Potential Outcomes: Yi (t): A deterministic function that maps each treatment level t to an observable
outcome. Importantly, this function exists prior to and independent of the treatment administration.
• Ti = 0: No Tylenol. Yi (0): the headache status if you receieve no Tylenol.

• Ti = 1: with Tylenol. Yi (1): the headache status if you receive Tylenol.

1.2 Estimands (what we want to estimate)


• A Unit Causal Effect: Yi (1) − Yi (0)
1
PN
• Sample Average Treatment Effect (SATE): N i=1 {Yi (1) − Yi (0)}
• Population Average Treatment Effect (PATE): E[Yi (1) − Yi (0)]

• The Fundamental Problem of Causal Inference


– We only observe one potential outcome for each unit. We never observe both.
– “Counterfactual Outcome”: The unobserved potential outcome for a given unit.

1
1.3 SUTVA
• In this class, we make the Stable Unit Treatment Value Assumption (SUTVA; Rubin 1980).
a) No interference between units (note: including the absence of the general equilibrium effects).

b) Single value (i.e., version) of each treatment condition.

1.4 Table of the Complete Data

Table 1: Table of the Complete Data


Units Yi (1) Yi (0) Causal Effect Ti Yi
1 7 1 6 1 7
2 5 6 -1 1 5
3 5 1 4 0 1
4 7 8 -1 0 8
Average 6 4 2

4
1X
SATE ≡ τ = {Yi (1) − Yi (0)} = 2
4 i=1

• Consistency of the Potential Outcomes: Yi = Ti Yi (1) + (1 − Ti )Yi (0).

This is the way in which the treatment assignment mechanism connects the potential outcomes and observed
outcomes.

2 Next Topic: Estimation of the SATE


1
PN PN PN PN
• Difference in Means; τ̂ = N1 i=1 Ti Yi − N10 i=1 (1 − Ti )Yi where N 1 = i=1 Ti and N 0 = i=1 (1 − Ti ).

• When does this difference-in-means estimator work? Why? → Lecture 2

You might also like