ECMT1020 Chap.1 Econometric Questions and Data Structures
Econometric Questions and Data Structures
Econometrics begins by translating an economic question into observable variables and a defensible comparison. This chapter distinguishes cross-sectional, time-series, panel and repeated cross-sectional data, then identifies the observational unit and population target. It also separates sample description, statistical inference and causal interpretation.
That discipline prevents a regression command from outrunning the structure and limitations of the data. A data label is useful only when it changes the analysis. Repeated observations on one household can share persistent behaviour, neighbouring regions can experience a common shock, and a fresh survey sample each year does not follow the same people. These details affect what counts as independent information.
The chapter therefore asks five linked questions: what is one row, which dimensions vary, how was the sample obtained, what population is targeted, and which claim is descriptive rather than causal. Answering them produces a compact design statement that can guide later choices about models, standard errors and interpretation.
What this chapter covers
- 01
Economic questions and measurable variables
- 02
Observational units and population targets
- 03
Cross-sectional data
- 04
Time-series dependence
- 05
Panel and repeated cross-sectional data
- 06
Description, inference and causality
Worked example · free
Classify a household energy dataset
- +1The repeating households and monthly dates make the dataset a panel.
- +1One observation is a household-month, located by household and date together.
- +1The sample can describe average use before and after the change.
- +1Weather, prices or household changes can move use at the same boundary.
- +1A causal interpretation needs a design or controls that address those rival changes.
Key terms
- Cross-sectional data
- Observations on many units at one date or short period.
- Time-series data
- Ordered observations on one unit through multiple dates.
- Panel data
- Repeated observations on the same entities across time.
- Observational unit
- The entity and occasion represented by one row.
- Causal inference
- Reasoning about what an outcome would have been without a proposed cause.
- Repeated cross-section
- A sequence of surveys taken at different dates using newly sampled units each time. It provides population snapshots but does not track individual change, so changing sample composition must be separated from genuine movement in the population distribution.
- Population target
- The complete group, period and quantity about which an empirical conclusion is intended. Stating it prevents a descriptive feature of the observed rows from being silently generalised to units, places or dates that the sample does not represent.
- Serial dependence
- A relationship between observations on the same unit at different dates. Persistence means nearby time observations may carry overlapping information rather than behaving like unrelated sample draws.
Econometric Questions and Data Structures FAQ
How do I identify panel data?
Look for the same named entities appearing at several dates. Many entities plus many dates is not enough if each date samples different entities; that structure is repeated cross-sectional.
Why does the row definition matter?
It determines the unit at which variables are measured and where dependence may occur. A household-month row, for example, can be related to other months from the same household.
Does regression establish causality?
No. A regression summarises a conditional relationship under a specification. Causal language additionally needs a design and assumptions that rule out competing explanations in the target population.
Exam move
For every dataset description, write four labels before calculating: observational unit, time index, population target and data structure. Then add one sentence distinguishing description from inference and one threat to a causal interpretation. Practise with household, firm, state and national examples so the classification depends on the dimensions rather than familiar subject matter.
Build a classification deck from short scenarios. On the front, record only the units, dates and sampling language. On the back, write the structure and a one-line reason. Mix cases where many dates are present but entities do not repeat, because those are the situations most likely to expose a memorised label.
Add a dependency note to each answer: serial dependence for one entity through time, shared shocks across entities at one date, or repeated-unit dependence in a panel. Next, rewrite broad claims into targets. Replace “income increased” with a statement that names the measured population, period and whether the same people were followed.
Replace “the policy caused the change” with the observed comparison and a list of competing changes. This editing exercise trains restraint before any formula appears. For worked questions, spend the first minute drawing a small entity-by-time grid. Shade the cells actually observed and identify the row represented in the file.
A full rectangle with repeated entity labels suggests a balanced panel; gaps signal an unbalanced panel; separate columns of new units suggest repeated cross-sections. Finish by explaining which conclusion the observed design can support without extra assumptions. Review mistakes by type rather than by scenario.
Keep separate error-log headings for the wrong observational unit, wrong time structure, unsupported generalisation and unsupported causal language. Reattempt one new example under each heading until you can produce the four-label audit quickly. When regression arrives later, return to the same audit and ask whether the software output respects the sampling and dependence structure you identified at the start.
Working through Econometric Questions and Data Structures in ECMT1020? Sia is AskSia’s AI Economics tutor — ask any ECMT1020 Econometric Questions and Data Structures question and get a clear, step-by-step explanation grounded in how ECMT1020 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
