Causal Inference in Economics

• Causal Inference focuses on leveraging theory and institutional details to estimate the impact of events and choices on a specific outcome.
• Humans have been interested in causality for a very long time

Causal Questions in Economics

• Does a rooster's crow make the sun rise?
• Does rain cause people to catch a cold?
• Can higher drug use rates result in increased crime rates?
• Does drinking salabat make colds or sore throats better?
• Does offering eggs to Sta. Clara make the weather nicer?
• Does massaging certain parts of the foot make internal organs healthier?
• Do Covid-19 vaccines cause death when injected?
• Do face shields reduce Covid-19 transmission?
• Does increasing the minimum wage lead to less employment for low-skilled workers?
• Can microfinance improve household income and reduce poverty?
• Does raising taxes on cigarettes lower smoking rates?
• Does using social media reduce mental health?
• Are happier people more likely to volunteer or does volunteering improve happiness?
• Does expanding health insurance coverage lower out-of-pocket expenditures?
• Do greater investments in education increase wages?
• Does using renewable energy subsidies lower carbon emissions?
• Does corruption in public procurement raise infrastructure costs?
• How do programs for fiscal stimulus affect rates of unemployment?
• Is economic growth impacted by foreign direct investment (FDI)?
• Did economic benefits come from martial law in Mindanao?
• Does democracy promote economic development?

The Start of Causal Inference

• John Snow's cholera outbreak work in the 1850's is considered the first instance of difference-in-difference
• Sir Ronald A. Fisher's The Design of Experiments was released in 1935
• Trygve Haavelmo's "The statistical implications of a system of simultaneous equations" was released in Econometrica 11(1): 1-12
• Donald Rubin's "Estimating causal effects of treatments in randomized and nonrandomized studies," was released in the Journal of Educational Psychology 66(5): 688-701
• Causal inference gained momentum and entered economics mainstream, gaining the name "Credibility Revolution
• Correlation does not mean Causation; correlation between 2 variables does not imply causal relationship

Global Rice Consumption

• Lack of correlation may not mean no causal relationship at play
• Though not obvious, a causal relationship could exist
• Those navigating choppy waters must endogenously turn rudder to offset wind and maintain a straight line
• No clear correlation between a straight-line path and a sailor's carefully planned rudder movements
• Since humans engage in optimal behavior, correlations do not reveal causal relationships, because human beings do not often act randomly
• Endogeneity is everywhere and is very difficult to rule out
• To remove endogeneity, experiments mimic the laboratory setting, however they're often infeasible, too costly, and involve ethical/moral dilemmas
• Researcher is a passive actor in data generation when using non-experimental or observational data
• From observational data, correlations do not reflect causal relationships because people choose the variable and are making optimal decisions
• For correlation to measure a causal effect, it must measure a choice independent of potential outcomes

Economic Correlation

• Economic theory states choices are endogenous, and therefore correlations between them and outcomes rarely represent a causal effect
• True scientists do not collect evidence to prove their beliefs and this is propaganda not science
• Scientific methodologies are methods for forming a particular kind of belief
• Methodologies allow acceptance of unexpected answers, are process oriented, not outcome oriented, and without it, causal methodologies are unbelievable
• Empirical analysis in the sciences depends on data to test theories and estimate relationships between variables

Models for Relationships

• Models are developed to describe the relationship between variables and form testable hypotheses
• These models can be translated into econometric models, which are then directly estimated using data
• Comparative statistics are theoretical descriptions of causal effects in models
• If one changes factor X ceteris paribus, and Y changes, one can infer that X causes Y
• Confounds can come from other factors moving simultaneously, hindering estimates of X's causal impact on Y

Causality in Income

• There's a correlation between people's choice of transportation and income but income has endogenous effects
• Reverse causality shows people relying on slower transport may face higher commute times, limiting their job opportunities
• The urban infrastructure quality is affecte by both transport choice and income (omitted variable bias)
• Lower-income may reside in areas with high access to transport, high income may want more private use

Endogeneity Problems

• Social media use may reduce mental health, or people prone to mental health may use it more (self-selection bias)
• Volunteering may improve happiness, or happier people may volunteer more (self-selection bias)
• Reverse causality may relate foreign direct investment (FDI) in economics and growth and economic development may cause democracy with other variables

Price Elasticity of Demand

• Empirically estimating the price elasticity of demand faces complexities
• A model can describe variable relationships and form testable hypotheses
• Econometric models can be made through translation for using data and testing theories
• By changing factor X ceteris paribus, and Y changes, one can reasonably infer that X causes Y in comparative statistics
• Demand is theoretical/unobservable. So, the price-quantity values at various equilibrium points are instead observed.
• Price-quantity pairs won't reflect the curves but connected points if there are simultaneous shifts, making them useless
• From Economics 102, price elasticity of demand is ∈ =dlogQ/dlogP
• In exogeneity, P is exogenous, and holds other factors fixed.

Measuring and Estimating Relationships

• Completely independent changes in P from other factors of supply and demand are needed
• A correlation between P and Q won't measure elasticity of demand
• The relationship empirically can be done with logQd = α + δlogP + yX + u formula, α is the intercept, δ is the elasticity, and X is the vector of factors like other prices or consumer income, y is the coefficient (factors and Qd.)
• Variation in prices must be independent of u (exogeneity). So, being correlated with some stuff captured/darkness
• The study of Education has literature of school inputs on student achievement (Class and spending)
• Student outputs use regression to explain amount of variables

Regression in Functions

• Coleman et al. (1966) used sensitivity analysis by looking at results arising from many specifications (Robus checks.)
• These studies failed and were reverse causations, omits variables bias
• Perverse: Test Score bad if it is in small classes(control for demographics) The struggling children and small classes go together
• School spending and outcomes is in areas of rich districts and large urban districts (where struggling minority students).
• Class and pupil are linked; You cannot add spend or have small classes

Interpretation Issues

• The weak foundation/causal of specification in text (J. D. Angrist & Pischke, 2010).
• Leamed (1983) criticized previous research by Ehrlich (1975b, 1975a) for it punishment had a significant deterrent.
• Ehrlich's findings sensitivity changes in functional or adds control (More data in 1960s.)

Reverse Causality

• Ehrlich studies credible, but failed in source of rate variation
• Murder rates have a two-way relationship w/ executions
• Used 2 stage lease estimator

Understanding Correlations

• Native regressions common(1960s-1980s).
• The issues and no overreaching causal and there was a growing frustration of the state or empiric economics, Spurious with confounding
• Given the key assumption held, results are not robust, data analysis
• Leamer(1983) and urged empirical researches to take con out.
• Different in non-experiment based on degree

Credibility Revolution

• Definition: Economics that shifts from transform to simple relation(Good designs by simple).
• Emphasis research mimics conditions( J. Angrist & Pischke, 2010).
• Random controlled and experiments methods
• Better data and less emphasis is econometric (More big data now and reduce causal.)
• Olden days is obsessive over analysiss, more focus with less specification
• Designs are credibility prima facie (J. D. Angirist & Pischke, 2010).
• Good designs typically lead methods that are simpel with easy results.
• Revolution was fueled by experimental designs that are experimental or quasi.
• Many were costly, clever desings by people for assignemnts etc.

Seminal Papers

• Card (1990) study 1980 Mariel Boatlift increased Miami's labor by %7 with skills
• D. Agnist from Vietnam with a lottery as experiment in natural with veeran earnings
• D. Angrist & Krueger 1991 did compulsory impacts

Nobel Memorial Prizes

• Banerjee, Duflo, and Kremer won prizes(2019) and Card, Angrist, and Imbens(2021) and Acemoglu, Johnson, and Robinson (2024)
• To be credible is to study and spell and research
• Not like Haavelmo Econometrica, which physicist experiments is an essential appendix or economic theorist

Economic Research

• The changing nature of studies, Hamemer mesh(2011).

External Validity

• Experimental from quasi offers little beyond the question
• Look evidence until a moral picture emerges
• Economic theory in framework

Second Critique

• It's focus to much on methodologicals
• Without what can not know or depend, estimative

Third Critique

• Economist Raj Chetty: They thin questionless with method
• James Heckman the problem is to much is to the New Yorker
• Use to trivial studies has causal or big

Basic Econometrics

• A random variable uses to mapping space (for all outcomes that set) to use analysis in processes

Expected Values

• Expected or expectations or random value with an average for a population

Expectation Properties

• For all, expectation has variables which use
• Population concept apply

Conditional Expections

• For variable Y under variable x, use if Y has variables under X.
• If indepdenthen E(YIX) = E(Y) because X has nothing.

Variance

• Deviation of the value.

(x−x)n2

Variable Estimations

• For population: y= Beta0+ Betal x + M.
• models explicitly show form

Coefficent Beta

• population impossible