What econometrics viva questions are testing
Econometrics viva questions ask you to explain what a model can establish, what its assumptions require, and how its results should be interpreted. They test more than whether you can state a formula. You may need to translate an estimate into plain language, defend an identification strategy, diagnose a problem, or explain what evidence would change your conclusion.
There is no single international format for an econometrics viva. The format and syllabus depend on your course, institution or examination. Check your course handbook or the relevant institution’s official website for details such as permitted materials, duration and assessment criteria; do not assume that a practice station reflects your own examination’s format.
The practice below uses three common discussion areas: interpreting a regression, omitted-variable bias, and instrumental variables. For each, answer the examiner’s opening question aloud. Then compare your answer with the feedback and model response. The aim is not to memorise a script, but to practise a reliable sequence:
- State your conclusion directly.
- Explain the relevant model or assumption.
- Apply it to the facts in the question.
- Name a limitation or check that could affect the conclusion.
A practice examiner strip for a general econometrics viva can keep the main areas of attention visible. In MySummaries, it can look like this:
These are useful practice emphases, not the official criteria of any particular examination. If your course provides its own rubric, use that instead.
Build stations from the material you need to explain
A good oral station begins with a decision or interpretation, not a request to recite a chapter. For example, you might be given a coefficient and asked what it means, or asked whether a proposed instrument is credible. Keep each station focused on one main skill, while allowing follow-up questions to probe assumptions and limitations.
Make your answer audible and organised. A candidate who lists every econometric caveat without answering the question can sound less precise than one who gives a clear conclusion, supports it, and then qualifies it. If you are unsure, say what follows from the information given and what cannot be concluded without more information.
Here is a station set that covers three central topics and leaves one station available for further practice. The attempted stations show a plausible mid-preparation record, not a target score.
Station 1: interpret a regression coefficient
This station tests whether you can interpret a semi-log coefficient accurately, use the standard error without overclaiming, and distinguish association from causation. The percentage interpretation is approximate; the exact percentage change for a one-unit increase is calculated by exponentiating the coefficient.
Examiner
A regression of log wages on years of schooling gives a coefficient of 0.08, with a standard error of 0.025. How would you interpret this estimate, and what would you need to know before making a causal claim?
You interpreted 0.08 as an approximate 8% difference in wages for one additional year of schooling.
ImproveState that the exact semi-log change is 100 × (exp(0.08) − 1), about 8.3%.
You used the standard error to recognise that the estimate is distinguishable from zero under conventional large-sample reasoning.
ImproveShow the calculation: 0.08 / 0.025 = 3.2, and qualify inference by the assumptions behind the standard error and test.
You noted that the regression alone does not establish causation.
ImproveExplain that confounding, including omitted ability or family background, can make schooling correlated with the error term.
You mentioned controls but did not connect them to a specific threat or design.
ImproveName the identifying assumptions or a credible design you would examine, rather than implying that adding controls automatically solves bias.
A strong answerHolding the included regressors fixed, one additional year of schooling is associated with an approximately 8% higher wage; the exact semi-log change is about 8.3%, calculated as 100 × (exp(0.08) − 1). The coefficient’s t-ratio is 0.08 divided by 0.025, or 3.2, but valid inference depends on the model and standard-error assumptions. This regression alone does not show that an extra year of schooling causes wages to rise. For that claim, I would need a credible identification strategy that addresses confounding, such as a defensible research design or instrument, and explain its assumptions.
That is a MySummaries station, filled with econometrics material. Yours is written from your own notes. Start free
A useful refinement is to avoid saying that a statistically significant coefficient is necessarily large, important or causal. Statistical significance concerns compatibility with a null hypothesis under specified assumptions. Economic importance requires interpreting the size in context, and causal interpretation requires an argument about identification.
If the dependent variable were wages in levels rather than log wages, the same coefficient would have a different interpretation: it would represent a change in wage units per additional year of schooling, conditional on the regressors. Always identify the scale of the dependent variable before translating a coefficient.
Station 2: explain omitted-variable bias
This station tests whether you can give the conditions for the direction of omitted-variable bias, rather than just saying that an omitted variable “causes bias”. The direction depends on two relationships: how the omitted factor affects the outcome and how it is associated with the included regressor.
Examiner
Suppose ability is omitted from a regression of wages on years of schooling. Under what conditions would the estimated schooling coefficient be biased upwards, and what could a researcher do about it?
You identified that ability must affect wages and be related to schooling for its omission to bias the schooling coefficient.
ImproveState the sign condition explicitly: if ability raises wages and is positively associated with schooling, the omitted-variable bias is upward.
You explained that the schooling coefficient can partly pick up the effect of ability.
ImproveConnect the omitted effect to the error term and explain why correlation between schooling and that error term violates exogeneity.
You suggested adding a measure of ability but treated it as a complete solution.
ImproveExplain that a control helps only if it is a suitable measure and the remaining assumptions are credible; alternatives may include a defensible design.
You avoided claiming that the observed coefficient is the causal return to schooling, but did not say what evidence would help.
ImproveSay what additional data, design or assumptions would be needed to support a causal estimate.
A strong answerIf ability raises wages and is positively correlated with years of schooling, omitting ability creates an upward bias in the schooling coefficient. Ability is then part of the regression error, and schooling is correlated with that error, so the OLS exogeneity condition fails. A suitable ability measure may help, but a noisy or incomplete proxy may leave residual confounding, and controls do not guarantee identification. I would consider the data and research design before claiming a causal return to schooling. The sign could differ if either relationship had the opposite sign.
For a compact verbal explanation, use the omitted-variable bias formula in words: the bias depends on the effect of the omitted variable on the outcome and its relationship with the included regressor. For this example, a positive effect of ability on wages combined with a positive relationship between ability and schooling implies an upward bias. If either sign changes, the direction may change too.
When asked how to address the problem, do not present “add controls” as a universal fix. A control must be measured and included appropriately; controlling for a variable affected by the treatment can create other problems. A natural experiment, random assignment or another research design may offer a route to identification, but each requires its own assumptions. State the proposed remedy and the risk it is meant to address.
Station 3: assess an instrumental-variable strategy
This station tests whether you can separate the three ideas often bundled into an instrument argument: relevance, independence or exogeneity, and exclusion. The question also asks what effect an instrumental-variable estimate may describe, rather than assuming it applies to everyone.
Examiner
A researcher uses distance to the nearest college as an instrument for years of schooling in a wage equation. What conditions would support this strategy, and what population effect might the estimate describe?
You said distance must predict schooling, which is the necessary first-stage idea.
ImproveDescribe how you would examine the first stage and acknowledge that a weak relationship undermines the design and can make inference unreliable.
You named both conditions but did not fully apply them to distance to college.
ImproveDiscuss whether distance is independent of unobserved wage determinants and whether it affects wages only through schooling.
You raised geography as a possible concern.
ImproveGive a concrete pathway, such as distance capturing local labour-market or family-background differences that also affect wages.
You noted that the estimate may not be a population-wide effect.
ImproveUnder the relevant assumptions, explain that with heterogeneous effects and monotonicity, the standard IV interpretation is a local average treatment effect for compliers.
A strong answerThe instrument must predict schooling, so I would examine the first stage and how precisely that relationship is estimated. It must also be independent of unobserved determinants of wages, conditional on the design’s stated controls, and affect wages only through schooling. Those assumptions are contestable here: distance to college could reflect local labour markets or family circumstances that directly affect wages. If treatment effects differ and the additional assumptions, including monotonicity, hold, the usual IV interpretation is a local average treatment effect for people whose schooling responds to the instrument. I would present the causal claim as conditional on these assumptions, not as established by the first stage alone.
Do not treat a strong first stage as proof that an instrument is valid. Relevance is empirically assessable, but the exclusion restriction is a substantive claim about the causal pathways from instrument to outcome. In this example, ask whether college distance could be related to local economic conditions or other factors that affect wages through routes other than schooling.
If the examiner asks what the IV estimate means, qualify the answer. Under the relevant assumptions, a standard IV estimate with heterogeneous effects can be interpreted as a local average treatment effect for compliers: people whose treatment status changes in response to the instrument. It need not equal the average effect for every person in the population.
Make the feedback useful
An oral answer can lose marks through a missing condition, an unsupported causal claim, or an interpretation that uses the wrong scale. Review the transcript of a station and identify the exact phrase where the reasoning became vague. Replace “the instrument is valid because it predicts schooling” with separate statements about relevance and the assumptions that cannot be demonstrated by the first stage alone.
Here is a short transcript review from the instrumental-variable station. The marked phrase is the point where the answer needs more precision.
Distance to college predicts schooling, so that supports relevance. The instrument is valid because of that first stage. I would also ask whether distance affects wages through local conditions. The estimate may apply to compliers if the additional assumptions hold.
The instrument is valid because of that first stage.
A first stage supports relevance but does not establish independence or the exclusion restriction.
Say: Say: The first stage supports relevance. I would separately defend the claim that distance is independent of unobserved wage determinants and affects wages only through schooling.A concise debrief can reinforce the distinction you need to remember. In practice, listen for whether you stated what the evidence supports and what remains an assumption.
You identified relevance, but the causal claim still depends on defending exclusion and independence.
A repeatable method for your own viva
Build practice stations from your syllabus, lecture material, problem sets and past feedback. For each topic, write an opening question that requires a decision or interpretation, then prepare follow-ups: What assumption is needed? What would violate it? What would you check? What can the result not establish?
After each attempt, record one specific improvement rather than writing out a full script. For instance: “state exact semi-log interpretation”, “give both signs for omitted-variable bias”, or “separate instrument relevance from exclusion”. Repeat the station later and check whether that point appears naturally in your answer.
If you are preparing for a course assessment, confirm the expected topics and format with your course materials or institution. Use this practice to sharpen your explanations, not as a substitute for the syllabus or official assessment guidance.
How MySummaries helps
You can build a revision board from your own econometrics notes, slides or problem sets, then use it to create oral stations and record answers against the material you are studying. The feedback can help you turn repeated gaps into focused revision. Open MySummaries.