Household surveys remain the public face of labor-market statistics, yet a growing body of work places tax and administrative receipts beside those questionnaires. On a light table, the two sheets do not always align. This note reviews how Katharine Abraham and coauthors have used linked evidence to show where secondary earnings and nonstandard pay can fall outside what a survey respondent reports—or what a survey even asks.
Abraham’s research program, including work with John Haltiwanger, Kristin Sandusky and James Spletzer, has compared survey answers with employer and tax records for the same people. The comparisons are not a contest to crown a single “true” income number. They are overlays: each source has a definition, a coverage rule and a reporting channel. When the sheets disagree, the disagreement itself is information about measurement.
What the administrative sheet shows
Administrative files capture payments that employers and platforms remit for tax or social-insurance purposes. They are strong on formal wage and salary flows that pass through reporting entities. They are weaker—or silent—when pay is cash, informal, or structured so that it never enters the reporting pipeline the analyst can see. Abraham and colleagues emphasize that administrative coverage is not universal; it is the footprint of institutions that file forms.
For secondary earnings, that footprint matters. A second job with a formal employer may appear clearly in payroll records. A short consulting spell paid outside payroll, or activity routed through entities the link cannot see, may leave little or no administrative mark. The sheet looks complete until the analyst remembers what the filing system never receives.
What the survey sheet shows
Surveys such as the Current Population Survey ask households about jobs and earnings with a fixed questionnaire and a limited interview time. Multiple jobholding questions exist, but they compete with other topics. Short spells, irregular hours and income that respondents do not classify as a “job” can be underreported. Abraham’s linked work has documented cases where administrative earnings exceed what the survey records for the same person over a comparable window—evidence that the survey overlay is thinner in places, not that respondents are uniquely unreliable.
Method limits belong on the page: linkage samples are not always nationally representative; matching algorithms drop ambiguous cases; tax year and survey reference periods may not coincide. When Abraham and colleagues report gaps, those design choices travel with the finding. FrostLamp repeats them here so the gap is not misread as a universal undercount of secondary pay.
Why both overlays stay on the table
Policy debate sometimes treats survey undercount as a single defect to be patched. Linked research suggests a more layered picture. Administrative data miss informal flows; surveys miss or compress some formal secondary spells; both miss activity that neither filing systems nor questionnaires recognize. Stacking the sheets does not produce a third “corrected” income series by default. It produces a map of where each instrument is informative.
For readers of measurement commentary, the practical implication is modest and methodological. Claims about the size of secondary earnings should name the instrument, the year and the definition of a job or payment. Abraham’s program is useful precisely because it makes those labels visible. FrostLamp does not translate the literature into advice about work or income. The note only records how two evidence layers sit when placed under the same lamp.
Limits of this note
This summary does not re-estimate Abraham’s tables. It does not claim that administrative totals are always higher, nor that survey redesign alone closes every gap. Contested points in the literature—how to treat independent contractors, how to align reference periods, how selection into the linked sample affects results—remain contested. Primary papers and agency technical documentation remain the source of record.
FrostLamp publishes no testimonials and makes no outcome claims. The light-table metaphor is editorial scaffolding for measurement, not a product promise.