Published September 8, 2026
Research question
What can public HMDA fields show about applications involving manufactured homes, and where should an analyst stop?
This brief focuses on public mortgage application records and manufactured-home fields. The unit of analysis is a reported loan or application record, modified for public release. Those two statements set the boundary for every comparison that follows.
Finding
HMDA can describe reported application and loan patterns after the analyst applies the published definitions. It does not reveal every credit decision, establish why one application had a particular result, or replace a lender file.
That finding is deliberately narrow. Public datasets are useful when the question matches the collection program. They become misleading when a national or regional estimate is treated as a property record, or when a screening value is written as though it were a verified condition.
Method
The review used this procedure: choose one data year, record whether the file is snapshot or dynamic, filter construction method to manufactured home, retain the land and loan property-interest codes, and document every exclusion. We opened the publisher's program page, data browser, and Regulation C material on September 8, 2026. We recorded the file version, URL, unit, geography, reporting period, filter values, and the treatment of unavailable fields.
| Method field | What the reviewer records | Why it matters |
|---|---|---|
| Universe | Population the program intends to cover | Defines who or what an estimate represents |
| Unit | Home, shipment, household, loan, place, or record | Prevents unlike counts from being combined |
| Period | Survey year, month, fiscal year, or release date | Keeps time comparisons consistent |
| Geography | National, region, state, county, tract, or place | Shows whether a local inference is supportable |
| Status | Preliminary, revised, final, suppressed, or unavailable | Preserves the publisher's qualification |
| Extraction | Table, filter, variable, weight, and calculation | Lets a second reviewer repeat the work |
We did not estimate a new national total, rank communities, or infer an individual outcome. No property files, resident records, lender files, or site observations were used. The result is a reading guide for the published sources, not a substitute for an analysis built around a specific business question.
How to reproduce the review
Start at the first source below and write down the dataset or program name exactly as the publisher gives it. Open the technical documentation before choosing a variable. If the publisher supplies a codebook, data dictionary, sample weight, standard error, or revision note, keep it with the extract.
Next, make a one-row source log for every value. Include the direct URL and access date. Save the filter selections in words; a screenshot alone may hide them. A second reviewer should be able to reach the same table and explain why the selected unit answers the research question.
For comparisons, put both definitions side by side. Stop if the universes, units, periods, or geographies differ. A comparison may still be useful, but it must be presented as two separate measures rather than a single rate or gap. Do not convert missing, suppressed, or not-applicable values to zero.