from description to causation

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

From Descriptionto Causation

Department of GovernmentLondon School of Economics and Political Science

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

1 Correlation vs. Causation

2 Over-Time Changes

3 Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

What makessomething a cause?

Write for 1 minute.

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

1 Correlation vs. Causation

2 Over-Time Changes

3 Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation

Correlation is the non-independence of twovariables for a set of observations

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation

Synonyms: correlation, covariation,relationship, association

Any correlation is potentially causalX might cause YY might cause XX and Y might be caused by ZX and Y might cause ZThere may be no causal relationship

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Flashback!

Two Categories of Inference:

1 Descriptive InferenceWhat are the facts?

2 Causal InferenceWhy does something occur?

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation is Causation?

The mind tends to interpretcorrelations and patterns as evidence

of causal relationships!

But this is rarely correct!

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: Wikimedia Commons

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: Wikimedia Commons

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: The Economist, 8 July 2016

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: The Economist, 8 July 2016

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: Randal Olson

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: Ministry of Justice, “Statistics on Race and theCriminal Justice System 2010”

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: StackExchange

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Source: TylerVigen.com

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Naive Causal InferenceCorrelations are not necessarily causal

Our mind thinks they are becausehumans are not very good at the kind ofcausal inference problems that socialscientists care about

Instead, we’re good at understandingphysical causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Physical causalityAction and reaction

Example:Picture a ball resting on top of a hillWhat happens if I push the ball?

Features:ObservableSingle-caseDeterministicMonocausal

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Physical causalityAction and reaction

Example:Picture a ball resting on top of a hillWhat happens if I push the ball?

Features:ObservableSingle-caseDeterministicMonocausal

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Physical causalityAction and reaction

Example:Picture a ball resting on top of a hillWhat happens if I push the ball?

Features:ObservableSingle-caseDeterministicMonocausal

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

1 Correlation vs. Causation

2 Over-Time Changes

3 Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Pre-Post Change HeuristicOur intuition about causation relies tooheavily on simple comparisons ofpre-post change in outcomes before andafter something happens

No change: no causationIncrease in outcome: positive effectDecrease in outcome: negative effect

Why is this flawed?

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Pre-Post Change HeuristicOur intuition about causation relies tooheavily on simple comparisons ofpre-post change in outcomes before andafter something happens

No change: no causationIncrease in outcome: positive effectDecrease in outcome: negative effect

Why is this flawed?

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Threats to Validity

Campbell and Ross talk about six “threats tovalidity” (i.e., threats to causal inference)

related to time-series analysis

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Flaws in causal inferencefrom pre-post comparisons

1 Maturation or trends2 Regression to the mean3 Selection4 Simultaneous historical changes5 Instrumentation changes6 Monitoring changes behaviour

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Maturation or trends

Is a shift in an outcome before and aftera policy change the impact of the policyor a small part of a longer time trend?

Case Study: Connecticut crackdown onspeeding (1955)

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Regression to the mean

Is a shift in an outcome before and aftera policy change the impact of the policyor a function of statistical variation?

Case Study: Connecticut crackdown onspeeding (1955)

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Selection

Is a shift in an outcome before and aftera policy the impact of the policy or theresult of the policy being implementedwhen outcomes are extreme?

Case Study: Connecticut crackdown onspeeding (1955)

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Simultaneous changes

Is the shift in an outcome before andafter a policy the impact of the policy orthe result of a simultaneous historicalshift?

Case Study: US Great Depression Policy

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Instrumentation changes

Is the shift in an outcome before andafter a policy the impact of the policy ora change in how the outcome ismeasured?

Case Study: Age-adjusted mortalityrates

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Monitoring changesbehaviour

Is the shift in an outcome before andafter a policy the impact of the policy ora change in response to measuring theoutcome per se?

Case Study: Educational testing

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

The more any quantitative socialindicator is used for socialdecision-making, the more subjectit will be to corruption pressuresand the more apt it will be todistort and corrupt the socialprocesses it is intended to monitor.

– Donald T. Campbell (1979)

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Flaws in causal inferencefrom pre-post comparisons

1 Maturation or trends2 Regression to the mean3 Selection4 Simultaneous historical changes5 Instrumentation changes6 Monitoring changes behaviour

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

1 Correlation vs. Causation

2 Over-Time Changes

3 Getting Systematic about Causality

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Directed Acyclic Graphs

Causal graphs (DAGs) provide a visualrepresentation of (possible) causal relationships

Causality flows between variables, which arerepresented as “nodes”

Variables are causally linked by arrowsCausality only flows forward in time

Nodes opening a “backdoor path” fromX → Y are confounds

“Selection bias” or “Confounding”

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Directed Acyclic Graphs

Causal graphs (DAGs) provide a visualrepresentation of (possible) causal relationships

Causality flows between variables, which arerepresented as “nodes”

Variables are causally linked by arrowsCausality only flows forward in time

Nodes opening a “backdoor path” fromX → Y are confounds

“Selection bias” or “Confounding”

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Directed Acyclic Graphs

Causal graphs (DAGs) provide a visualrepresentation of (possible) causal relationships

Causality flows between variables, which arerepresented as “nodes”

Variables are causally linked by arrowsCausality only flows forward in time

Nodes opening a “backdoor path” fromX → Y are confounds

“Selection bias” or “Confounding”

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Smoking Cancer

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Smoking Cancer

Age

Environment

GeneticPredisposition

ParentalSmoking

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Smoking Cancer

Age

Environment

GeneticPredisposition

ParentalSmoking

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Smoking Cancer

Age

Environment

GeneticPredisposition

ParentalSmoking

Correlation vs. Causation Over-Time Changes Getting Systematic about Causality

Causal InferenceCausal inference (typically) involves

gathering data in a systematic fashion inorder to assess the size and form of

correlation between nodes X and Y in such away that there are no backdoor paths

between X and Y by controlling for (i.e.,conditioning on, holding constant) any

confounding variables, Z.