What is econometrics? A plain-language introduction
Econometrics uses data and statistical methods to measure economic relationships, test theories and evaluate policies. Here is what that means in practice.
Econometrics combines economic reasoning, data and statistical methods to study economic relationships. It helps answer questions such as how earnings vary with education, whether a policy changed employment and how inflation might develop. A useful result reports the size of a relationship, the uncertainty around it and the assumptions needed to interpret it.
The IMF's introduction to econometrics describes the connection between economic theory and measurement. Theory supplies possible mechanisms; observations help assess how well those explanations describe the world.
An example: did a training programme raise wages?
Suppose, in a hypothetical study, workers who completed training earn an average of $24 per hour, while other workers earn $20. The observed difference is $4. That is a descriptive comparison, and a reasonable starting point.
It is not yet the effect of training. Participants may have had more experience, different jobs or higher wages before the programme. The causal question asks what those same participants would have earned without training. That unobserved alternative is the counterfactual.
An econometric analysis therefore asks how workers entered the programme, what data are available before and after it, and which comparison could approximate that counterfactual. Random assignment, a defensible policy comparison or another research design may help. Merely adding many variables to a regression does not automatically resolve selection.
This is why econometrics involves both calculation and an argument about how the data were generated. The correlation versus causation article develops that distinction.
What is econometrics used for?
| Purpose | Example question | Evidence to report |
|---|---|---|
| Measurement | How does spending vary with income? | Units, sample and estimated relationship |
| Theory testing | Are the observations consistent with a proposed demand model? | Restrictions tested and uncertainty |
| Forecasting | What might inflation be next quarter? | Forecast horizon, benchmark and prediction interval |
| Policy evaluation | What did a training programme change? | Comparison group, identification assumptions and effect estimate |
These tasks overlap, but they need different evidence. A model can predict wages well while failing to identify the effect of schooling. A credible policy estimate may explain only a modest share of individual wage variation. Define the task before choosing the technique.
The main kinds of econometric data
Cross-sectional data observe many units in one period: for example, a survey of workers in a given year. Time-series data observe a quantity over time, such as monthly unemployment. Panel data follow the same units repeatedly, such as the same firms each year.
The structure changes the comparisons you can make. A time series needs attention to persistence, trends and the information available at each date. A panel lets you study changes within units, while still requiring assumptions about time-varying confounding. Read the introductions to time series analysis and panel data for examples.
How an econometric analysis works
A practical workflow is:
- State the question. Specify the outcome, population, period and quantity you want to learn about.
- Understand the data. Check what a row represents, measurement units, missing observations and how the sample was collected.
- Choose a research design and model. Explain what comparison the estimate uses and why its assumptions are reasonable.
- Estimate and investigate. Examine uncertainty, diagnostics and sensitivity to defensible alternative choices.
- Report the result. Keep the data decisions, code and outputs together so someone else can understand and rerun the work.
The econometric analysis guide expands that workflow. If you are earlier in the process, begin with data analysis and exploratory data analysis.
Which methods should you learn first?
Start with regression analysis, which describes how an outcome varies with predictors. Learn to read coefficients, confidence intervals and sample definitions before treating the output as an answer. Then study the distinction between association and causation.
Further methods depend on the question: binary-outcome models for yes-or-no responses, panel methods for repeated observations, time-series models for temporal dependence and causal designs for intervention questions. The methods reference explains individual families, implementation boundaries and worked examples. A longer estimator list cannot replace a suitable research design.
Econometrics shares much of its mathematics with statistics. Its economic questions and data settings shape how those tools are used; econometrics versus statistics explains the overlap.
A brief history
Ragnar Frisch and Jan Tinbergen received the first economics Nobel Memorial Prize in 1969 for their work on dynamic economic models. Later developments placed increasing emphasis on credible research designs. Angrist and Pischke's 2010 account of the credibility revolution describes that shift in empirical microeconomics.
Start with a question and a small project
Choose one economic question, write down what evidence would answer it and work through a complete example. The teaching labs pair questions with data and Python analyses. The reproducible workflow guide explains how to keep the work inspectable, while the glossary provides short definitions as you go.
References
- Ouliaris, S. Econometrics: Making Theory Count. International Monetary Fund, Finance & Development, Back to Basics.
- The Nobel Prize. The economics prize, 1969–2007: new methods of economic analysis.
- Angrist, J. D. and Pischke, J.-S. (2010). The credibility revolution in empirical economics: How better research design is taking the con out of econometrics. Journal of Economic Perspectives, 24(2), 3–30. Publisher page and open paper.