Questions: Time Series Cross-Section (TSCS) Models

Question 1

A researcher analyzes a panel of 30 countries over 50 years using standard OLS with no correction for temporal dependence. A global financial crisis hits all countries simultaneously in year 30. What is the primary statistical problem?

• A. The sample size is too small to produce reliable coefficient estimates
• B. Serial autocorrelation within each country inflates the number of effective observations
• C. Contemporaneous cross-unit correlation means standard errors underestimate true uncertainty, since all units are hit by the same shock
• D. OLS cannot be applied when the number of time periods exceeds the number of units

Answer: (C)

A common global shock creates contemporaneous cross-unit correlation: observations for different countries in the same year are not independent. Standard OLS assumes independence across observations, so its standard errors are too small — results appear more statistically significant than they are. Panel-corrected standard errors (PCSEs) are designed precisely to account for this structure alongside serial autocorrelation and panel heteroskedasticity.

Question 2

Why can TSCS data with 25 countries observed over 40 years not be analyzed adequately with standard cross-sectional regression techniques?

• A. The effective sample size of 1,000 is too large for reliable standard error estimation
• B. Cross-sectional regression cannot accommodate more than one observation per unit
• C. Ignoring serial autocorrelation within countries produces artificially small standard errors and overconfident inference
• D. Standard regression requires that the number of units exceed the number of time periods

Answer: (C)

In TSCS data, observations for the same country across years are not independent — past outcomes predict future ones (serial autocorrelation). Standard cross-sectional regression assumes independent observations; applying it to panel data understates standard errors and creates false precision. This is only one of three dependence structures in TSCS; contemporaneous cross-unit correlation and panel heteroskedasticity compound the problem.

Question 3

Clustering standard errors at the unit level is almost always appropriate in TSCS analysis because all observations within a unit are correlated across time.

• T. True
• F. False

Answer: True

Within a country (or firm, region, etc.), observations across years are rarely independent — economic, political, and institutional dynamics create strong temporal correlation. Clustering at the unit level allows for arbitrary within-cluster correlation rather than assuming independence, producing standard errors that reflect actual uncertainty. Not clustering when temporal dependence exists typically leads to confidence intervals that are too narrow.

Question 4

Fixed effects models preserve most variation in TSCS data by controlling for time-invariant unit characteristics without discarding any information.

• T. True
• F. False

Answer: False

Fixed effects models discard between-unit variation entirely — they demean each unit, so only within-unit variation over time is used for estimation. Variables that do not vary over time within a unit (e.g., a country's colonial history) cannot be estimated at all. This is a deliberate tradeoff: fixed effects powerfully control for unobserved unit-level confounders, but at the cost of all between-unit comparative information.

Question 5

What three distinct dependence structures characterize TSCS data, and why does each require a methodological response beyond what standard cross-sectional or time-series methods provide?

Model answer: 1) Serial autocorrelation: within-unit observations are correlated across time (past GDP predicts current GDP), so errors within units are not independent — ignoring this inflates apparent precision. 2) Contemporaneous cross-unit correlation: units observed in the same period co-move due to common shocks (global recessions, regional contagion), violating cross-unit independence — ignoring this understates true uncertainty. 3) Panel heteroskedasticity: error variance differs across units (large economies have larger absolute shocks than small ones) — ignoring this produces inefficient estimates. PCSEs address all three simultaneously.

Each problem exists in isolation in other data types, but TSCS data has all three at once, making standard approaches from either cross-sectional or time-series analysis insufficient. This is what makes TSCS a distinct methodological domain.