Economics, Business & Finance: Identification: naming the obvious alternative

What you'll be able to do: Name the boring alternative explanation for what looks like a causal story, which is what "identification" means in this subject, and the difference between a correlation and an answer.

Why this is the technical core of the discipline

Correlation is cheap. A rising minimum wage next to a falling employment rate, a marketing campaign next to a sales spike, a stimulus package next to a recovery, every one of these pairs can be read as cause and effect, and every one of them has at least one boring alternative explanation that produces the same pattern without the causal story being true.

Identification is the economist's word for the specific thing in your data or your research design that lets you tell the causal story apart from the alternative. It is not a synonym for "evidence." A correlation is evidence for a claim; it is evidence for several claims at once, and identification is what narrows that down to one.

The classic cases, and what solved them

Minimum wage and employment. A state raises its minimum wage and its unemployment falls afterward, does the raise not cost jobs, or does the state raise wages precisely when its labour market is already tightening (reverse causality: the strong economy causes both the wage confidence and the falling unemployment)? A well-known empirical approach compared adjacent counties across a state border, one that raised its minimum wage and one that didn't, on the logic that neighbouring local economies share most other shocks (weather, regional demand, industry mix) so a difference in outcomes is more plausibly attributable to the wage change itself. That's an identification strategy: a specific design feature meant to rule out the alternative story, not just more data on the same correlation.

Education and earnings. People with more education earn more, does education cause the higher earnings, or do people who would have earned more anyway (more ability, more family resources, more connections) also happen to get more education? This is "ability bias," and it's why studies exploiting something that shifts education for reasons unrelated to ability, a change in compulsory schooling age, distance to the nearest college, are taken more seriously than a raw correlation between years of schooling and income.

Stimulus and recovery. An economy recovers after a stimulus package, does the stimulus cause the recovery, or would the economy have recovered on a similar timeline regardless, with the stimulus getting credit for a business cycle turning on its own? Comparing the recovery's speed and shape against previous recoveries, and against countries that didn't run a comparable stimulus, is an attempt at identification; citing the recovery alone is not.

The exercise

Here's a claimed cause-and-effect story: [describe it, the outcome, the proposed cause, the evidence offered]. Before accepting it: name two or three plausible alternative explanations that would produce the same evidence without the causal story being true. For each alternative, tell me what specific piece of evidence, a comparison, a timing check, a natural experiment, would let me tell it apart from the causal story.

The output that matters is the second half: not just "here's a confound" but "here's what data would distinguish the confound from the real effect." A list of possible confounds with no way to adjudicate between them is not identification either, it's just a longer list of things you don't know.

Across the disciplines

Microeconomics. Ice cream sales and drowning deaths both rise in summer, a classic case where a third factor (heat) drives both, with no causal link between them at all. The exercise is spotting when your own analysis has the same structure with less obvious variables.

Macroeconomics. Inflation and unemployment moved together in some decades and apart in others, any single episode is compatible with several competing macro stories, which is part of why macro disagreement (method 07) runs as deep as it does; a single business cycle rarely identifies anything on its own.

Business and marketing. Sales rose the week of a marketing campaign, was it the campaign, seasonality, a competitor's stockout, or a price change that happened the same week? A/B testing exists specifically to solve this problem: holding everything else equal between a treated and an untreated group is what an experiment buys you that a before-and-after comparison can't.

Finance. A fund manager beat the market last year, skill, or luck, given that some fraction of any large group of managers will beat the market by chance in any given year purely by the law of averages? Distinguishing the two requires looking at performance across many periods and many managers, not one good year.

Management. A new process was introduced and productivity rose, the process, or the fact that whichever team got picked to pilot it was already the stronger team (selection effect)?

Pitfalls

1. Stopping at "there could be other explanations." That's true of nearly every correlation and gets you nowhere. Name the specific alternative.
2. Naming an alternative and not saying what would distinguish it. A confound you can't test for is not yet identification, it's a caveat.
3. Treating a natural experiment as automatically clean. Adjacent counties, policy discontinuities and similar designs are stronger than raw correlation, not proof, they can still share confounds the researcher didn't think of.
4. Being satisfied by a story merely consistent with the evidence, rather than one the evidence could have contradicted. If no possible outcome would have changed your mind, you weren't testing anything.
5. The tell: you've never had a causal claim you found persuasive get undermined by naming the obvious alternative, either you're unusually careful already, or you've never actually tried.

Try this today

Find a causal claim from recent economic or business news, a policy that "caused" an outcome, a campaign that "drove" sales. Name the most plausible alternative explanation, and say specifically what evidence would let you tell the two apart.