Operations
Part of Secondary market research: a practical guide
The secondary market research checklist that catches what audits miss
A secondary market research checklist ties each claim to a publisher, dataset version, reporting unit and license before the number reaches a decision.
What to take away
- A portal's privacy policy covers the portal. Provenance lives with the publisher, so check the dataset page, not the homepage.
- Record publisher, version, unit, geography and release date for every series you cite. A number without a version cannot be defended later.
- Treat license and redistribution rights as a separate gate from analytical fitness. Permission to read is not permission to republish.
- Reconcile classifications, currencies and nominal versus real values before comparing two sources.
- Set a revision watch on any series that feeds a live decision, because published figures get restated.
A secondary market research checklist earns its keep when every claim can be traced to a named publisher, a version and a unit. That is a different job from designing a survey or running a focus group. This one is about sources other people collected.
Portal policy is not dataset provenance
Data.gov's privacy and website policy describes the operating agency, its information practices and how it handles external links. It governs that portal. It says nothing about the method behind a cataloged table.
From Portal to Publisher
- Start at portal policy page
- Follow record to publisher
- Read dataset documentation
- Check access terms and limits
- Log publisher, dataset, version, date
Follow each record out to its publisher. Read the dataset's own documentation, access terms and known limitations. That step is how to research secondary market research properly, rather than trusting a blanket policy written for a website.
- Preserve the query, the filters and the date you ran them.
- Prefer the original publisher over an aggregator that reformatted the file.
- Log publisher, dataset, table, series, version and access date.
Metadata that makes a number defensible
Four fields do the heavy lifting: publisher, version, unit and geography. A figure missing any of them is a placeholder, not evidence.
Units fail quietly. A series reported in thousands, in nominal dollars, or per establishment will not match one reported in units, real dollars, or per business. Write the unit into the claim itself.
Currency and price basis need the same treatment. State the year, the base and whether the figure is nominal or inflation-adjusted. If you convert, show the rate and its source.
Compatibility before comparison
Two sources can both be accurate and still not compare. Check the classification first: NAICS, SIC, NACE and internal codes do not map cleanly onto each other.
Then check the frame. A count of establishments is not a count of businesses, and a household survey is not a person-level survey. Date alignment matters too, since fiscal years and calendar years rarely line up.
The federal Data Ethics Framework sets out decision points on need, impact, lawfulness, representation and transparency. It is federal guidance, not a universal standard. Adapt its questions to your sector and jurisdiction, and check that your secondary market research benchmarks actually measure decision coverage and provenance.
Reviewer must see
- Scope
- Decision and operational definition
- Source
- Original record and method note
- Analysis
- Compatible inputs and reproducible steps
- Rights
- Item-level permission and controls
Stop when
- Scope
- No actionable boundary
- Source
- Provenance is broken
- Analysis
- Result cannot be traced
- Rights
- Use is unauthorized or unclear
License and redistribution
Reading a dataset and republishing it are different acts. Check the license on the item, not the site. Look for attribution requirements, commercial-use limits, redistribution terms and any restriction on scraping or bulk download.
Retention and model-training clauses now appear in licenses that once said nothing about them. Read the current text rather than a summary of it.
The W3C Privacy Principles statement gives shared vocabulary for privacy questions in web systems and warns against pushing that work onto individuals.
Apply it to the common questions your team keeps asking, then confirm the governing law with counsel.
For anything procured or embedded, the CISA software acquisition fact sheet covers supply-chain exposure and vulnerability management. Those questions belong in a review. They are not local approval.
Watch the series after you publish
Publishers revise. A quarterly series gets restated, a base year changes, a classification gets retired. Any figure feeding a live decision needs a named owner and a check cadence.
When a revision lands, rerun the analysis that depended on it and say what changed. Retire evidence that no longer fits the definition you started with.
Common questions
Does a completed checklist prove the conclusion?
No. It produces reviewable gates. Someone qualified still has to judge whether the source fits the claim and whether the analysis holds. The checklist makes that judgment inspectable.
Who signs off?
Assign owners for the business decision, the method, the data, the rights and the publication. One person can hold two roles on a small team, but the roles should be named.
When does the checklist get rerun?
After any material change to the decision, the definition, the source, the version, the calculation or the intended use. A revision to a cited series counts as a material change.
