PLCP 1010 Basic terms

causal claim

X made Y happen, instead of some other outcome.

counterfactual

What would have happened without the cause. Never observable.

association

Two variables change together. Not proof of causation.

independent variable (x) and dependent variable y

X is the cause or treatment. Y is the effect or outcome.

bias

Systematic error that stays even with unlimited data.

random error

Error with no pattern. It averages out and shrinks with more data.

selection bias

Bias from looking at a non-random subset of cases. On the causal graphs it is the collider.

potential outcome

The outcome a unit would show under treatment, or under control.

fundamental problem of causal inference

We see each unit under only one condition, so the effect on any unit is never observed.

average treatment effect (ATE)

The average difference treatment makes: Ȳtt − Ȳtc. Never directly observable.

Y bar

An average outcome for a group.

Y bar tt, Y bar tc, Y bar ct, Y bar cc

First letter = group assigned; second = outcome under. Observable only when the letters match (Ȳtt, Ȳcc).

front door path

X → Y, or X → M → Y: the causal path.

back door path

A non-causal route from X to Y through a common cause: X ← Z → Y.

spuriousness

An association produced entirely by a common cause. it disappears when you adjust for it

endogeneity

Reverse causation: Y causes X (X ← Y).

collider

A variable caused by both X and Y (X → C ← Y). Selecting on it creates a fake association.

mediator

A variable between X and Y (X → M → Y). It is the causal mechanism.

exchangeability

Assignment is independent of potential outcomes. Swap the groups and results wouldn't change.

comparability

The two groups look alike on background traits.

null hypothesis

Assumes no effect (ATE = 0, or coefficient = 0).

p value

The chance of seeing evidence this strong if the null were true.

statistically significant

p < 0.05: reject the null.

regression coefficient

The slope: the change in Y linked to a one-unit change in X

control variable

A variable added to a regression to adjust for a possible confounder.

process tracing

Using within-case evidence to test causal steps and eliminate rival explanations.

hoop test

The hypothesis must pass to stay alive. Failing eliminates it; passing doesn't confirm it.

smoking gun test

Passing strongly confirms the hypothesis; failing doesn't eliminate it.

framework theory

Very general and abstract. Can't be tested directly.

political culture

Outcomes are mainly shaped by shared values.

political economy

Outcomes are mainly shaped by economic interests.

preference

A ranking of options by how much you want them

rational choice

Rank your options and choose the top one. It says nothing about what people should want.

backward induction

Solve a game from the last move back to the first.

power (overt)

A gets B to do something B wouldn't otherwise do.

structural power

A holds an advantageous position in a social field. No overt act is needed.

situated power (Wartenberg)

Power sustained by surrounding "peripheral social others," like employers who give grades meaning.

bottom-up causation

Ordinary people's shared beliefs create political systems, not elites.

civic culture

A cluster of interpersonal trust, life satisfaction, and low support for revolution.

female subordination

Fish's proposed link between Islam and authoritarianism: women's low status, measured by literacy gap, sex ratio, and women in government.

factor endowments

A country's land, labor, and capital.

isomorphism

The idea that hierarchical family relations reproduce themselves in politics.