Skills

Econometric Analysis with Real Data

10 weeks · 0 milestones

Build and run an econometric model (regression or equivalent) on a real, named publicly available dataset (World Bank, FRED, ONS, or equivalent). Document your dataset source, model specification, results, and a limitations section naming the main threats to validity. Proof is your analysis output plus a methodology challenge session with a statistics or economics academic who questions at least one model assumption or specification choice — your written response addressing their challenge in full is a required part of the proof. The model code or methodology must be reproducible.

Part of: Economics · Proves for: UG

Milestone map

3 milestones

Identify a real research question and select a suitable econometric model

4 weeks

Select a real research question that can be addressed with econometric methods using a publicly available dataset — questions about the effect of a policy on an outcome, the relationship between economic variables, or the determinants of a social or economic outcome. Choose a dataset (World Bank Open Data, FRED, OECD.Stat, or similar), formulate a testable hypothesis, and specify the econometric model: OLS, IV, difference-in-differences, panel data, or logit/probit. Document why this model is appropriate for the question and what identification assumptions it relies on.

Proof required

Research design document (500+ words) covering the research question, dataset source and key variables, econometric model specification, justification for the model choice, and identification assumptions with a discussion of whether they are likely to hold.

What gets checked

Research question is specific enough to produce a falsifiable empirical result
Model choice is justified by the structure of the question and the data — not chosen because it is familiar
Identification assumptions are stated specifically — not just 'OLS assumptions hold'

Common mistakes

Choosing a model and then finding a question to fit it — the question drives the model, not the other way round
Claiming OLS is appropriate without discussing endogeneity — most social-science research questions have endogeneity concerns that require IV, DiD, or panel methods
Not discussing the data source's limitations — every publicly available dataset has measurement and coverage limitations relevant to the research question

Resources

Foundationstart here

Depthgo deeper

What a verifier looks for

Confirm the dataset is publicly available — request the source link
Review the model justification — confirm it addresses why this model is appropriate for the question structure
Check identification assumptions — confirm they are stated specifically for this model and data

You'll sign in first, then come straight back here.

Estimate the model and interpret results with diagnostics

5 weeks

Estimate the specified econometric model using Python (statsmodels, linearmodels), R, or STATA on the dataset from M1. Run appropriate diagnostic tests (heteroscedasticity, autocorrelation for time series, multicollinearity for OLS, instrument strength for IV). Interpret the results: report coefficient estimates with standard errors and confidence intervals, discuss practical significance alongside statistical significance, and state what the results imply for the research question.

Proof required

Results document (500+ words) covering coefficient estimates with standard errors and confidence intervals, diagnostic test results with interpretation, practical significance discussion, and the data and code available in a public repository (GitHub, OSF, or similar).

What gets checked

Coefficient estimates are reported with standard errors and confidence intervals — not just p-values
Diagnostic tests are run and interpreted — not just mentioned as passed
Practical significance is discussed: is the effect size economically meaningful, not just statistically significant?

Common mistakes

Reporting p-values without effect sizes — econometric results are often statistically significant but economically trivial; both must be addressed
Running diagnostics without interpreting failures — finding heteroscedasticity and doing nothing about it undermines the inference
Not making code and data publicly available — reproducibility is a basic standard for quantitative social science

Powstik Guide

A 500-word results write-up with estimates and diagnostics.

Steps

Report coefficients with standard errors.
Add confidence intervals.
Run and report diagnostic tests.
Interpret results in plain terms.

Template

| Variable | Coefficient | SE | 95% CI |
|----------|-------------|----|--------|
Diagnostics run and results:
Interpretation:

What gets sent back

Under 500 words.
Sources uncited.
Hypothetical instead of real.

Resources

Foundationstart here

Depthgo deeper

Masteryfor the dedicated

What a verifier looks for

Access the code repository — confirm the estimation code is present and runs on the dataset
Review diagnostic test results — confirm they are interpreted, not just listed
Check the practical significance discussion — confirm it addresses economic meaningfulness alongside statistical significance

You'll sign in first, then come straight back here.

Defend methodology and results under economist challenge

1 week

Present the econometric analysis to an economist or applied researcher for a methodology challenge. The reviewer must probe identification — 'your instrument — why is it truly exogenous?', 'you're claiming a causal effect but your model doesn't rule out reverse causality here', 'your sample period — do these results hold in different periods?'. Document the challenges and your responses. This Q&A satisfies the ADVERSARIAL VERIFICATION RULE for social-* outcomes.

Proof required

Q&A notes (250+ words) documenting the reviewer's credentials (economist or applied researcher), at least three specific identification or methodology challenges, your responses, and any analysis updates committed to based on the session.

What gets checked

Reviewer is an economist or applied researcher with econometrics experience
At least three specific identification or methodology challenges are documented
At least one analysis update is committed to based on the session

Common mistakes

Choosing a reviewer who approves the approach without probing identification assumptions — identification is where most applied econometrics breaks down
Not being able to defend the instrument choice or causal claim under challenge — if the identification assumption cannot be defended in real time, the causal claim was unjustified
Not updating any analysis based on the challenge — a legitimate identification critique requires a response

Powstik Guide

Your identification strategy, challenged by an economist.

Steps

Write your causal claim in one sentence.
State the assumption it rests on.
List the best argument against that assumption.
Book an economist or applied researcher.
Ask for three identification challenges.

Template

Challenge session log
Date:            Duration:
Reviewer name:
Reviewer role and credentials:

Challenge 1 (identification or exogeneity):
  What they asked:
  My answer:
  Did it hold?
Challenge 2:
  What they asked:
  My answer:
  Did it hold?
Challenge 3:
  What they asked:
  My answer:
  Did it hold?

Update I am committing to:

What gets sent back

A reviewer who approves without testing identification.
A causal claim you can't defend live.
No update after the session.

Resources

Foundationstart here

Depthgo deeper

Masteryfor the dedicated

What a verifier looks for

Confirm the reviewer is an economist or applied researcher with econometrics experience
Review Q&A notes — confirm at least three identification or methodology challenges are documented
Check that at least one analysis update is documented based on the session

You'll sign in first, then come straight back here.

Part of

Prove Your Economics Chops →Economics 4: Statistics and econometrics →

Related outcomes

Milestone map

3 milestones

Identify a real research question and select a suitable econometric model

4 weeks

Proof required

What gets checked

Research question is specific enough to produce a falsifiable empirical result
Model choice is justified by the structure of the question and the data — not chosen because it is familiar
Identification assumptions are stated specifically — not just 'OLS assumptions hold'

Common mistakes

Choosing a model and then finding a question to fit it — the question drives the model, not the other way round
Claiming OLS is appropriate without discussing endogeneity — most social-science research questions have endogeneity concerns that require IV, DiD, or panel methods
Not discussing the data source's limitations — every publicly available dataset has measurement and coverage limitations relevant to the research question

Resources

Foundationstart here

Depthgo deeper

What a verifier looks for

Confirm the dataset is publicly available — request the source link
Review the model justification — confirm it addresses why this model is appropriate for the question structure
Check identification assumptions — confirm they are stated specifically for this model and data

You'll sign in first, then come straight back here.

Estimate the model and interpret results with diagnostics

5 weeks

Proof required

What gets checked

Coefficient estimates are reported with standard errors and confidence intervals — not just p-values
Diagnostic tests are run and interpreted — not just mentioned as passed
Practical significance is discussed: is the effect size economically meaningful, not just statistically significant?

Common mistakes

Reporting p-values without effect sizes — econometric results are often statistically significant but economically trivial; both must be addressed
Running diagnostics without interpreting failures — finding heteroscedasticity and doing nothing about it undermines the inference
Not making code and data publicly available — reproducibility is a basic standard for quantitative social science

Powstik Guide

A 500-word results write-up with estimates and diagnostics.

Steps

Report coefficients with standard errors.
Add confidence intervals.
Run and report diagnostic tests.
Interpret results in plain terms.

Template

| Variable | Coefficient | SE | 95% CI |
|----------|-------------|----|--------|
Diagnostics run and results:
Interpretation:

What gets sent back

Under 500 words.
Sources uncited.
Hypothetical instead of real.

Resources

Foundationstart here

Depthgo deeper

Masteryfor the dedicated

What a verifier looks for

Access the code repository — confirm the estimation code is present and runs on the dataset
Review diagnostic test results — confirm they are interpreted, not just listed
Check the practical significance discussion — confirm it addresses economic meaningfulness alongside statistical significance

You'll sign in first, then come straight back here.

Defend methodology and results under economist challenge

1 week

Proof required

What gets checked

Reviewer is an economist or applied researcher with econometrics experience
At least three specific identification or methodology challenges are documented
At least one analysis update is committed to based on the session

Common mistakes

Choosing a reviewer who approves the approach without probing identification assumptions — identification is where most applied econometrics breaks down
Not being able to defend the instrument choice or causal claim under challenge — if the identification assumption cannot be defended in real time, the causal claim was unjustified
Not updating any analysis based on the challenge — a legitimate identification critique requires a response

Powstik Guide

Your identification strategy, challenged by an economist.

Steps

Write your causal claim in one sentence.
State the assumption it rests on.
List the best argument against that assumption.
Book an economist or applied researcher.
Ask for three identification challenges.

Template

Challenge session log
Date:            Duration:
Reviewer name:
Reviewer role and credentials:

Challenge 1 (identification or exogeneity):
  What they asked:
  My answer:
  Did it hold?
Challenge 2:
  What they asked:
  My answer:
  Did it hold?
Challenge 3:
  What they asked:
  My answer:
  Did it hold?

Update I am committing to:

What gets sent back

A reviewer who approves without testing identification.
A causal claim you can't defend live.
No update after the session.

Resources

Foundationstart here

Depthgo deeper

Masteryfor the dedicated

What a verifier looks for

Confirm the reviewer is an economist or applied researcher with econometrics experience
Review Q&A notes — confirm at least three identification or methodology challenges are documented
Check that at least one analysis update is documented based on the session

You'll sign in first, then come straight back here.