Introduction to Econometric Data

Every day, governments, businesses, and individuals face economic choices. Should the central bank raise interest rates? Does a higher minimum wage reduce employment? How much will a new tax affect consumer spending? To answer such questions, we need more than theory — we need data. This chapter introduces the kinds of data economists work with and the two main goals of any data analysis: understanding cause and effect, and making accurate predictions.

The Big Picture#

Econometrics is the toolkit that lets us turn raw data into answers about the economy. This chapter lays the foundation: we’ll see why data matters, what it means to ask a cause‑and‑effect question versus a prediction question, and how the structure of our data shapes what we can learn. By the end, you’ll be able to look at a real‑world dataset and recognize whether it’s cross‑sectional, a time series, or a panel — and you’ll understand the crucial difference between experiments and observational data.

Economic Questions and Quantitative Answers#

Economists are rarely satisfied with a simple “yes” or “no.” Suppose a policymaker asks, “Does education increase earnings?” The answer “yes” is qualitative — it tells us the direction, but not how much. They need a quantitative answer: by how many dollars does an extra year of schooling raise annual income? This is a quantitative question, and answering it requires data.

Econometrics is the branch of economics that uses statistical methods to measure economic relationships from data. To measure the return to education, we collect information on individuals’ years of schooling and their earnings. But we can’t observe every person in the country — that would be the entire population. Instead, we work with a sample, a manageable subset of the population.

Population: The entire group of individuals, firms, or countries we want to study. Sample: A subset of the population that we actually observe.

We use the sample to compute statistics (like the average earnings difference between high‑school and college graduates) and then draw conclusions about the population. The leap from sample to population is where statistical reasoning comes in, and it’s a central theme of econometrics.

Think of a chef who wants to know whether a new ingredient improves a recipe. She doesn’t cook for the entire world; she prepares a few dishes, tastes them, and figures out the effect. Econometrics does the same with data — but instead of taste buds, we use statistics and probability.

📝 Section Recap: Econometrics provides quantitative answers to economic questions by using data from a sample to learn about a larger population.

Two Goals: Causal Inference and Prediction#

When we analyze data, we usually have one of two very different goals in mind.

Causal inference asks: “If I change X, what happens to Y?” For example, “If we raise the minimum wage by $1, how many jobs are lost in the restaurant industry?” Answering this requires isolating the effect of the wage change while holding everything else constant — a tough job, because in the real world many things change at once.

Prediction asks: “Given what I know about X, what is my best guess for Y?” For example, “Given today’s unemployment rate, inflation, and consumer confidence, what will next quarter’s GDP growth be?” Here we don’t need to know why the variables are related; we just need a pattern that forecasts well.

A weather forecast is a perfect analogy. Meteorologists predict rain using air pressure, humidity, and wind patterns. They don’t claim that high humidity causes rain — they just know that when humidity is high, rain often follows. That’s prediction. A causal question would be: “Does cloud seeding cause more rainfall?” To answer that, you’d need an experiment.

Both goals are important. A central bank needs accurate forecasts to set interest rates, but it also needs to know the causal effect of a rate change on inflation. Econometrics provides tools for both, but the methods differ. Mixing them up is one of the most common mistakes in data analysis.

📝 Section Recap: Causal inference asks what happens when we deliberately change something; prediction asks what we can expect to see based on observed patterns. Each requires its own set of tools.

The Ideal Experiment: Randomization#

If we want to nail down a causal effect, the gold standard is a randomized controlled trial (RCT). Here’s how it works.

Suppose we want to know whether a new teaching method improves test scores. We take a group of students and randomly split them into two groups. The treatment group gets the new method; the control group gets the standard method. Random assignment ensures that, on average, the two groups are identical in every other way — motivation, prior knowledge, family background, and so on. Any difference in test scores at the end can then be confidently attributed to the teaching method.

Randomized controlled trial (RCT): An experiment where subjects are randomly assigned to a treatment group or a control group, so that any difference in outcomes can be credited to the treatment.

Randomization acts like a fair coin toss that balances out all the hidden factors we can’t measure. This is why RCTs are so powerful: they eliminate confounding — the nasty situation where a third variable influences both the cause and the effect, creating a misleading relationship.

Economists have used RCTs to study everything from the effect of class size on learning to the impact of health insurance on medical spending. The famous RAND Health Insurance Experiment randomly assigned families to different insurance plans and found that higher cost‑sharing reduced healthcare use without harming health — a result that shaped policy for decades.

But here’s the catch: many economic questions can’t be answered with an experiment. You can’t randomly assign countries to different trade policies, or force some people to get less education. When experiments are impossible, we must rely on observational data and clever statistical designs that try to mimic randomization. That’s where much of econometrics lives.

📝 Section Recap: Randomized experiments are the cleanest way to learn cause and effect because they eliminate confounding. In economics, experiments are valuable but often not possible, so we need other methods.

Experimental Data and Observational Data#

The distinction between experimental data and observational data is fundamental.

Experimental data: Data generated by a controlled, randomized experiment. Observational data: Data collected by simply observing real‑world behavior, without any intervention by the researcher.

Experimental data gives us a clear, apples‑to‑apples comparison. If we randomly assign fertilizer to some plots of land and not others, any difference in crop yield can be safely attributed to the fertilizer.

Observational data is far more common in economics, but it comes with a big problem: selection bias. Suppose we want to measure the effect of college on earnings. We compare the average wages of college graduates and high‑school graduates using a survey. College graduates earn more — but is that because of college? Or are people who go to college already more able, more motivated, or from wealthier families? If we just compare averages, we mix up the effect of college with the effect of those pre‑existing differences. That’s selection bias.

Selection bias: The distortion that arises when the treatment and control groups differ in ways other than the treatment itself, because individuals (or firms, or countries) chose their own group.

Think of it this way: if you only observe that people who wear expensive suits tend to have higher incomes, you wouldn’t conclude that buying an expensive suit causes a raise. The kind of person who buys an expensive suit is also the kind of person who tends to earn more. Observational data forces us to confront these kinds of self‑selection problems.

Econometric techniques — like adding control variables, using instrumental variables, and difference‑in‑differences — are designed to tease out causal effects from observational data by making it behave more like experimental data. We’ll spend much of the course learning how to do that.

📝 Section Recap: Experimental data comes from randomized trials and gives clean causal comparisons. Observational data is abundant but plagued by selection bias, so we need special methods to extract credible causal conclusions.

Data Structures: Cross‑Sectional, Time Series, and Panel#

Economic data comes in three main shapes. Recognizing which shape you’re holding is the first step in choosing the right analysis.

Cross‑sectional data is a snapshot of many units at a single point in time. For example, a survey of 1,000 households in 2023 that records income, consumption, and family size. Each row is a different household, and we often denote the outcome for unit as . Cross‑sections are great for comparing different groups at one moment — say, income differences between urban and rural households.

Time series data tracks a single unit over many time periods. The quarterly GDP of the United States from 2000 to 2023 is a classic time series: one country, many quarters. We use notation where indexes time. Time series data lets us study trends, cycles, and make forecasts. But it’s tricky because observations are ordered and often depend on their own past.

Panel data (also called longitudinal data) combines the two: we observe multiple units, each followed over time. For instance, annual income for the same 500 individuals from 2010 to 2020. We write , where is the individual and is the year. Panel data is powerful because it lets us control for unobserved characteristics that don’t change over time. If we want to study the effect of job training on wages, we can compare a person’s wage before and after training, removing the effect of their innate ability — something a single cross‑section cannot do.

Cross‑sectional data: Observations on many individuals, firms, or countries at one point in time. Time series data: Observations on a single entity over many time periods. Panel data: Observations on multiple entities, each observed over multiple time periods.

A panel can be balanced (every unit is observed in every period) or unbalanced (some units drop out or enter late). In practice, unbalanced panels are common, and econometric software handles them.

The data structure determines what questions we can answer. A cross‑section can tell us whether richer countries have lower infant mortality, but it can’t tell us how a country’s own growth affects its own mortality over time — that requires panel or time series data. Choosing the right data for the question is half the battle.

📝 Section Recap: Economic data comes in three forms: cross‑sectional (many units, one time), time series (one unit, many times), and panel (many units, many times). The structure dictates which causal and predictive questions we can tackle.

Summary#

We’ve taken our first step into econometrics by understanding the raw material — data. You now know that economic questions demand quantitative answers, and that the goal of our analysis can be either causal or predictive. The ideal way to learn cause and effect is through a randomized experiment, but most economic data is observational, which requires careful handling. Finally, you can identify the three main data structures — cross‑sectional, time series, and panel — and you appreciate that the structure determines what questions we can answer and which methods we’ll use. With this foundation, you’re ready to start building the statistical tools that turn data into insight.