Wisdom
Consumer spending patterns shift as households move through different life stages, but whether those shifts are driven by life stage itself or simply by the income differences between groups remains an open question. This project uses data from 7,094 households — students, working adults, and elderly — to estimate the causal effect of income on discretionary spending composition, and whether that effect differs by life stage.
Research Question
What is the causal effect of a $10,000 income increase on spending composition, and does that effect differ by life stage?
This question matters because income is one of the few consumer characteristics that policy can directly change — through earned income tax credits, cash transfers, minimum wage laws, or income support programs. Understanding how much an income increase shifts spending, and whether that shift is the same for students, working adults, and retirees, gives policymakers concrete guidance about where income transfers will have the largest behavioral effects.
Why This Question
Income can be changed through policy — earned income tax credits, cash transfers, minimum wage increases, or income support programs. This means we can ask a concrete question: what would happen to a household’s spending if its income rose by $10,000? The answer may depend on life stage. A student who receives an extra $10,000 faces very different financial pressures than a retiree in the same situation. Whether the income → spending relationship differs across life stages — and by how much — is what this project estimates.
The Three Life Stages
We define three life stages based on age and dataset source:
- Students — drawn from the Kaggle Student Spending dataset; college-aged individuals with spending tracked across tuition, housing, food, entertainment, and more
- Working Adults — drawn from the Kaggle Customer Personality Analysis dataset; general adult consumers aged roughly 25–64
- Elderly — drawn from the 2024 BLS Consumer Expenditure Public Use Microdata (PUMD); households where the reference person is 65 or older
The Outcome Variable
We measure spending composition using percentage share of total spending in four broad categories: food, discretionary (leisure/entertainment), essentials (housing + transportation), and healthcare. Shares rather than raw dollars allow fair comparison across groups with very different income levels.
A Preceptor’s Table describes the ideal dataset we wish we had — every row and column that would perfectly answer our question — then contrasts it with what we actually have. The gap between them is where honest uncertainty lives.
What moment in time? The Preceptor Table refers to households as they were at the time of their survey — 2022–2024, depending on the data source. In the ideal dataset, we would observe the same household at multiple income levels over time: once at their baseline income and again after a $10,000 income shock, so we could directly measure how spending changed. In our actual data, each household is observed only once, at a single income level, so we estimate the income effect by comparing different households at different income levels rather than tracking the same household over time.
Units: One household observed at one point in time. In the ideal dataset, the same household would be observed repeatedly across all three life stages — as a student, then as a working adult, then as a retiree — so we could watch spending composition change within the same unit over time. In our actual data, each household appears only once, from whichever life stage it was in when surveyed.
Treatment: Annual income. In the ideal dataset, income would be randomly assigned — imagine a government lottery that gave different households different income amounts — so we could directly measure how spending changed as a result. In reality, we observe households with different incomes that they earned themselves, which makes it harder to isolate income’s true effect. Life stage tells us whether the income effect differs across groups — it is not what we are changing.
Quantity of interest: How much does a $10,000 income increase change the share of spending that goes to discretionary goods — and does that change differ between Students, Working Adults, and Elderly households? Income is what we are testing; life stage is the lens we use to see whether the effect differs.
Outcome: Percentage share of total spending allocated to discretionary (leisure/entertainment) goods. In the ideal dataset, “discretionary” would be defined identically across all households using a single consistent taxonomy. In our actual data, the definition differs by source dataset and required manual mapping.
Covariates: Life stage, number of children, age, family size, region, year, education, and marital status. In the ideal dataset, all of these would be verified and consistently measured. In our actual data, only life stage and number of children are available across all three sources; region is only available for BLS PUMD rows, and education and marital status are partially missing.
Data rows: The 7,094 households from three one-time surveys, each household interviewed once during 2022–2024. These are the rows we actually observe.
Preceptor rows: The same households observed at multiple income levels — once at their baseline income and again after a $10,000 income boost. Without these repeated observations we cannot directly see how a household’s spending changes when its income changes; instead we estimate the income effect by comparing different households at different income levels.
Example Preceptor Table
The table below shows what the ideal dataset would look like for three example households. Each household has two potential outcomes — discretionary spending share at their current income and at income + $10,000. The causal effect is the difference. Hatched cells are unobservable: we never see what a household would have spent at the other income level.
| Unit | Life Stage | Age | Children | Share (current) | Share (+$10K) | Causal Effect |
|---|---|---|---|---|---|---|
| Household A ($15K) | Student | 20 | 0 | 22% | ? | ? |
| Household B ($60K) | Working Adult | 42 | 2 | ? | 14% | ? |
| Household C ($30K) | Elderly | 71 | 0 | 18% | ? | ? |
| … | … | … | … | … | … | … |
We observe only one potential outcome per household — the spending they actually had at their actual income. The counterfactual (what they would have spent at $10K more) is never directly observed. We estimate the average causal effect by comparing different households at different income levels, assuming those comparisons are fair after accounting for life stage and number of children.