A housing chart published last month shows one number for January. The source now shows a different January number. Before changing the chart, determine what changed: the source estimate, the statistical method, the downloaded series, or your own processing. Those possibilities require different explanations, and treating them all as errors can confuse readers.
The Census Bureau publishes preliminary residential construction estimates and provides revision information. Its Survey of Construction methodology also describes updates to earlier estimates and seasonal factors. This guide uses those source practices as context for original fictional examples. The dates and numbers below are illustrations, not current housing results. The goal is to maintain a chart that remains understandable as its source evolves.
Keep the observation date separate from the release date
A value can describe activity during January while being released in February and revised in March. January is the observation period. February and March identify different releases containing information about that period. A dataset vintage is the version of the available series associated with a particular release or retrieval point.
Imagine a fictional January estimate initially reported as 100 units. A later release revises it to 110. Both values refer to January, but they come from different vintages. If your file records only January and the value, you lose the information needed to explain why the number changed. Retain the release date or another stable version identifier.
For a small publication, this need not involve elaborate software. Save the original download under a clear dated name and keep a short note identifying the source release. Avoid overwriting the only copy of yesterday's data when refreshing a chart. The previous version is evidence of what the publication used at the time.
A revised baseline can change the reported growth rate
Suppose a fictional February value is 121 units. Compared with the original January value of 100, it represents 21% growth. Compared with the revised January value of 110, it represents 10% growth. Both calculations are arithmetically correct, but they use different baselines.
If the current release provides February and a revised January, a current comparison should ordinarily use that consistent release context. Copying February into an old spreadsheet while leaving January untouched creates a mixed vintage result. The calculated 21% may then differ from the source's current 10% even though no formula is broken.
A reviewer should therefore inspect the inputs before debugging the percentage formula. Ask whether both periods came from the same current download or whether one was copied from an earlier article. When a source revises history, the update must consider the historical portion of the chart, not only the newest row.
Decide whether the chart shows current knowledge or historical reporting
Two chart designs can be legitimate. A current vintage chart shows the source's latest available estimates for all displayed periods. A historical reporting chart shows values as they were first published or as they appeared on specified release dates. These designs answer different questions and should not share an ambiguous caption.
The first design helps readers understand the series using the current source record. The second can help evaluate what information was available to observers at an earlier time. A retrospective article asking what analysts knew in February may need the February vintage, even if a later revision produces a different January estimate.
Choose the design before assembling values. If the chart mixes current and historical vintages for a deliberate reason, explain that reason visibly. Otherwise, maintain a consistent policy. A chart should not become a patchwork merely because some rows were downloaded recently and others survived from an old workbook.
Separate a source revision from a publisher correction
Imagine that the original source said 100, but your article accidentally displayed 1,000 because of a misplaced zero. Fixing that is a publisher correction. Now imagine the article correctly displayed 100 and the source later revised it to 110. Updating that is a source revision. Readers deserve to know which event occurred.
A correction note can identify the original mistake, the corrected figure, and whether the interpretation changed. A revision note can identify the new source release and explain that earlier estimates were updated. Avoid language that implies the source made a mistake unless the source itself identifies an error. Preliminary estimates can change through an expected statistical process.
The distinction also matters internally. Track the cause in an update log so that future editors do not repeat a processing error or blame a normal revision. A publication that records only changed value loses useful information about why its work changed and how to improve its workflow.
Revisions can affect the shape of a whole chart
A source may update more than the most recent month. Changes to seasonal factors or other documented methods can alter a stretch of historical values. If the chart imports only the newest observations, its line can stop matching the source even though each old value was once published correctly.
For a fictional example, suppose twelve monthly values were previously flat, but a later consistent vintage changes several of them slightly. The visual trend and the highest month may change. A headline written around the old peak should be reviewed along with the chart. Updating numbers without revisiting the prose can leave an article internally inconsistent.
Check annotations, summary cards, downloadable tables, alt text, and social preview text if they repeat affected findings. Not every small revision requires rewriting an entire article. The appropriate scope depends on whether the revised values change a material claim. A short checklist makes that decision deliberate rather than accidental.
Do not assume every difference is a revision
Before attributing a mismatch to the source, compare the series definitions. One download may be seasonally adjusted and another unadjusted. One may be an annual rate and another a monthly count. Geography, construction stage, units, and coverage may differ. Those are series mismatches, not necessarily revisions of the same observation.
Imagine a fictional chart labeled completions while its refreshed download contains starts. The values might look plausible enough to escape a quick visual check. A release date alone will not catch that mistake. Preserve a series identifier and a plain language description with every dataset so that a refresh can verify identity before replacing values.
Also check whether the source distinguishes preliminary, revised, and final values through flags. Keep those flags in the working file even if the public chart uses a simpler note. Removing metadata too early makes it harder to diagnose a mismatch and easier to mistake a status change for a new observation.
An update record that is actually useful
A practical record includes the source URL, release date, retrieval date, series identifier, observation period, original value, new value, and reason for the update when known. It can also record which chart or article used the earlier value. This is enough to connect a changed input with its public consequences.
For the fictional January change from 100 to 110, the record might note that a later source release revised the estimate and changed the February growth comparison from 21% to 10%. That sentence identifies both the data event and the editorial effect. It is more useful than a note saying data updated.
Do not invent a cause when the source does not provide one. You can report that the value changed between specified releases without claiming to know whether late responses, seasonal updates, or another process caused that individual revision. General methodology explains possible mechanisms, not necessarily the cause of every observed change.
Handle an old article with care
An archived article may document what was reported at the time rather than serve as a continuously updated dashboard. In that case, silently replacing its historical values can undermine the record. A visible update note or a link to the current series may be preferable, depending on the publication's archive policy.
A living dashboard has a different promise. Readers usually expect it to reflect the chosen current vintage. Its update process should refresh compatible historical values and identify the latest release. The publication can preserve old downloads behind the scenes while presenting the current series clearly.
State the policy in ordinary language. For example, explain that the chart uses the latest available source estimates, including revisions to prior periods. Or explain that an article preserves values available on its publication date. Either statement helps readers understand why a historical article and a current dashboard may show different numbers for the same month.
Check whether the conclusion survives
After refreshing the data, recompute the measures that support the headline. Does an increase remain an increase? Does the highest value remain the highest? Does a comparison remain meaningful after uncertainty is considered? A revision that barely changes a table might still matter if the original claim depended on a narrow difference.
For the fictional change from 21% to 10%, the direction stays positive but the magnitude changes substantially. The article should not keep a phrase such as more than one fifth simply because the chart was corrected. Prose, numbers, and captions must describe the same vintage.
Keep the original chart export with the original data when the chart supports a published claim. A screenshot alone cannot reproduce the calculation, while a data file alone may not show the wording readers saw. Together, the archived inputs and presentation make it possible to explain a later change accurately.
A reliable update finishes when the reader can tell what period was measured, which version is shown, and whether the interpretation changed. Preserving vintages makes that possible. It allows a housing publication to improve its current picture without losing the history of how that picture developed.