Problem-First Empirical Economics
Start with a consequential question, find data patterns, then choose the math
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 94%
Problem-First Empirical Economics begins by choosing a question that matters to people's lives, such as underemployment, pain, insecurity, or wage stagnation. Researchers define the population and practical outcome before selecting a mathematical technique. They inspect data for recurring patterns, contradictions, structural breaks, and relationships that conventional explanations miss. Rather than collecting only confirming examples, they search for evidence that could overturn their preferred account. Methods are then chosen because they clarify the problem, not because they are prestigious or technically elaborate. Finally, results are translated back into effects on ordinary people and policy choices. Blanchflower presents this as an alternative to economics that becomes excessively mathematical and loses sight of what the analysis is meant to improve.
Origin
Danny Blanchflower credited empirical economist Richard Freeman with teaching him to examine data and later advised young economists to focus on important human problems.
Core principles
- 01Important human problems should drive research design
- 02Patterns in data can expose failures in prevailing theory
- 03Mathematics is a tool rather than the research objective
- 04Ordinary people's well-being is a meaningful outcome
- 05Good researchers actively seek disconfirming evidence
How to run it
- 1
Choose the Human Problem
Identify an outcome that materially affects people's well-being and frame a specific answerable question. Name the population bearing the consequences.
Pro tip Ask what people outside the profession would care about changing.
Watch out Do not begin with a favored technique and search for a convenient application.
- 2
Map the Relevant Measures
List the observable variables that represent the problem, including imperfect proxies and overlooked dimensions. Document how each measure can mislead.
Pro tip Include subjective measures such as pain or anxiety when the outcome requires them.
Watch out A convenient measure may omit the mechanism that matters most.
- 3
Inspect Patterns Before Modeling
Plot trends, distributions, subgroup differences, and historical relationships. Look for turning points or contradictions that require explanation.
Pro tip Compare multiple demographic groups and time periods before aggregating.
Watch out Do not discard a puzzling pattern merely because it conflicts with established theory.
- 4
Seek Disconfirmation
Specify what evidence would show that the emerging explanation is wrong and search for it directly. Compare competing mechanisms rather than accumulating only supportive correlations.
Pro tip Write the disconfirming tests before finalizing the model.
Watch out Confirmation bias can turn abundant data into false confidence.
- 5
Apply Proportionate Methods
Use enough mathematics and statistical structure to answer the question credibly. Keep the connection between assumptions, estimates, and the human outcome visible.
Pro tip Prefer transparent robustness checks over unnecessary complexity.
Watch out Technical sophistication cannot rescue a poorly framed question.
- 6
Return to Well-Being
Explain what the result means for the affected people and which decisions it should inform. Identify unanswered questions and evidence needed next.
Pro tip Express findings in concrete outcomes such as hours, wages, security, health, or opportunity.
Watch out Do not end with statistical significance while leaving practical significance unstated.
In the wild
A conventional analysis expects low unemployment to generate strong wage growth, but the predicted acceleration does not appear. The researcher examines desired hours, insecure contracts, temporary workers, and bargaining power, revealing that underemployment remains elevated and better explains wage weakness.
→ The investigation reframes the Phillips curve around underemployment rather than forcing the data into the old unemployment model.
Survey data initially suggest that children make adults less happy. Researchers add families' difficulty paying bills and discover that the relationship reverses after accounting for the high cost of raising children.
→ The revised model distinguishes the happiness associated with children from the financial stress associated with caring for them.
Common mistakes
Starting with the Mathematics
A method-first project can produce polished equations while losing sight of the consequential question it should answer.
Hunting Only for Confirmation
Searching exclusively for supportive data exaggerates confidence and leaves alternative explanations untested.
Forgetting the Ordinary Person
Research becomes less useful when aggregate outcomes are not connected to the people experiencing insecurity, pain, or lost opportunity.
Is it for you?
Best for
It is best for economists, analysts, students, and policy researchers choosing questions and designing empirical investigations.
Not ideal for
It is not ideal for purely theoretical work whose explicit purpose is developing mathematical foundations without immediate application.
From the transcript
“focus on the patterns in the data”
“Focus on an important question”
“Don't lose sight of that.”
From the episode
Danny Blanchflower Discusses Labor Economics (Podcast)
Danny Blanchflower