Seasonally Adjusted vs. Not Seasonally Adjusted Data: SA, NSA, and SAAR

Economic activity often changes predictably with the calendar. Retail hiring may rise before the winter holidays, construction may slow in cold weather, and factory schedules may shift during model changeovers. These repeated influences can make one month look stronger or weaker than the neighboring month even when the underlying pace has changed little.

Seasonal adjustment estimates the part of a series associated with recurring seasonal effects and removes it from the published adjusted series. Agencies may also account for predictable calendar effects such as different weekday counts or movable holidays when their methods call for it. The aim is to make month-to-month or quarter-to-quarter movement easier to interpret, not to erase real economic activity or produce a perfectly smooth trend. The Bureau of Labor Statistics describes seasonal adjustment as removing recurring seasonal influences; the Census Bureau explains that it can help reveal nonseasonal features of a time series. See the [BLS overview of seasonal adjustment]({source:blsSeasonalAdjustmentOverview}) and [Census Bureau seasonal adjustment FAQ]({source:censusSeasonalAdjustmentFaq}).

Seasonal adjustment is not the same as inflation adjustment. A seasonally adjusted series can still be measured in current dollars, and a real, inflation-adjusted series can still be presented with or without seasonal adjustment. Check both dimensions when reading a price, output, income, or spending statistic.

Seasonality is not the same as the business cycle. A predictable summer slowdown may recur each year, while a decline that lasts several quarters may reflect broader economic activity or another shock. Seasonal adjustment cannot classify every movement perfectly, so read the release alongside its technical notes and other indicators.

SA means seasonally adjusted. NSA means not seasonally adjusted; releases may also call it unadjusted or original data. The two versions usually describe the same economic series but answer different comparison questions.

The adjusted number is an estimate based on the series and the agency's method, not a second survey. It may move differently from the unadjusted number because the seasonal effect estimated for one month can be larger than for another. BLS says neither version is universally better: unadjusted data can suit comparisons of seasonal hiring in the same part of different years, while adjusted data are often more useful for recent month-to-month movement. The [BLS guide to using adjusted and unadjusted CPI data]({source:blsCpiSeasonalDataUse}) gives a series-specific example; check each release's own notes before generalizing.

The adjustment changes the comparison by estimating the calendar pattern for each period. Consider a deliberately simple, hypothetical series that uses an additive seasonal effect. In Month A, an observed value of 100 includes a seasonal effect estimated at 15, leaving an adjusted value of 85. In Month B, an observed value of 108 includes a seasonal effect estimated at 25, leaving an adjusted value of 83.

The unadjusted series rises by 8, while the adjusted series falls by 2. That is possible because the usual seasonal lift is estimated to be larger in Month B. In this example, the change in the adjusted series asks whether the observed rise exceeded the expected calendar pattern. The figures are invented solely to illustrate the arithmetic; they are not readings for any published indicator.

This subtraction is only a teaching example for an additive adjustment. Many series use multiplicative factors, in which the adjusted value is calculated by dividing the original value by a factor; methods may also combine seasonal, weekday, or moving-holiday effects. A change in direction is therefore not, by itself, evidence that the agency made an error or that the economy reversed sharply. The Census FAQ notes that opposite directions between adjusted and unadjusted series are common and should be interpreted with the seasonal factor and the rest of the release in view. {source:censusSeasonalAdjustmentFaq}

For a published series, a small adjusted decline alone does not establish a clear turning point. Sampling error, rounding, revisions, and uncertainty in the estimated factors can all affect a short-term reading. Displayed decimal places show how a value is reported, not how certain the underlying estimate is.

<!-- learn:illustration -->

Seasonal factors are estimated from the history of a particular time series. Statistical procedures look for patterns that recur at similar times of the year while distinguishing them from longer-run movement and irregular variation. A series may be represented as seasonal, trend-cycle, and irregular components; the published adjusted series generally removes the estimated seasonal component while retaining nonseasonal movement and irregular shocks.

The pattern is not necessarily a fixed percentage that an analyst subtracts every year. Some series fit better with additive effects, while others use multiplicative factors whose size depends on the level of the series. If relevant, a method may also estimate trading-day effects or a moving holiday such as Easter. The Census Bureau maintains X-13ARIMA-SEATS for seasonal adjustment and describes how it uses diagnostics to evaluate residual seasonality and stability. That describes the tool and method family, not a guarantee that every federal release applies identical settings. Consult the methodology for the specific series. {source:censusSeasonalAdjustmentFaq}

The calendar is only one influence on the estimate. A sudden strike, unusual weather, a pandemic disruption, sampling error, or a business-specific shock can remain in the adjusted series as irregular movement. Some agencies model identified outliers or calendar events separately, but seasonal adjustment does not make an observation free from measurement uncertainty. A smoother chart should not be mistaken for a direct measurement of an unobserved trend.

Some releases adjust a headline total while publishing selected components without seasonal adjustment. If component series use different adjustment bases, adding the displayed values may not reproduce the published total. Check the status and aggregation method for each series before summing or comparing them.

SAAR means seasonally adjusted annual rate. It combines two separate ideas: seasonal adjustment estimates and removes recurring calendar effects; the annual-rate convention scales a period's adjusted level or growth pace to an annual comparison. Do not read the acronym as a claim that the current month's pace will last for a year.

For example, the Census Bureau presents certain monthly building-permit estimates at a seasonally adjusted annual rate. If a hypothetical adjusted monthly pace is 8,000 permits, its SAAR is 8,000 multiplied by 12, or 96,000. The annual-rate figure helps compare the current month with an annual total. It does not say 96,000 permits were issued in that month, nor does it forecast that 96,000 will be issued over the next year. This example follows the [Census Bureau's building-permit methodology]({source:censusBuildingPermitsMethodology}) and uses invented values.

Quarterly GDP growth at an annual rate is a related but distinct convention. BEA compounds a quarterly percentage change over four quarters; it does not simply multiply the observed change by four. A seasonally adjusted GDP level shown at an annual rate, an annualized quarterly percentage change, and an annual forecast are three different things. BEA explains the quarterly growth calculation in its [annual-rate FAQ]({source:beaQuarterlyAnnualRates}). For the rate-comparison details, see how to read real GDP growth rates.

Yes. Seasonal factors are estimates, and new observations can change the estimated recurring pattern. Agencies may rerun the adjustment as new data arrive, revise earlier adjusted observations on a scheduled basis, or update adjusted values when the underlying unadjusted data themselves are revised. The frequency and history affected differ by program.

For example, BLS recalculates seasonal factors for its seasonally adjusted CPI series each year and may revise up to five years of those adjusted indexes. That is a description of the CPI program, not a universal revision rule for every U.S. data set. BLS also reassesses whether individual CPI series still exhibit enough seasonality to warrant adjustment. The [CPI seasonal-adjustment page]({source:blsCpiSeasonalDataUse}) describes its current process and revision window.

When reproducing a chart, record the series identifier, seasonal-adjustment status, release date, and data vintage. A real-time chart created from the values available at the time may differ from a chart downloaded later, even though both use the same series name. Keep the original value and the adjusted value distinct when the question is about the seasonal pattern or the agency's later revisions.

A revision does not automatically overturn an earlier interpretation, but it can matter when the initial movement was small or close to a turning point. When comparing a past report with the latest download, state whether the data were revised and preserve the version used for each comparison.

Use seasonally adjusted data when the question concerns the latest movement from the previous month or quarter and the agency publishes a suitable adjusted series. Removing the expected seasonal pattern can make a new turning point easier to see. It does not prove that the remaining change is statistically meaningful; short-term estimates can still be noisy.

Use NSA data when the recurring seasonal cycle is itself the subject. For example, comparing retail hiring from October through December across several years can show how the holiday season changed. A matched-month year-over-year comparison also reduces the effect of a stable seasonal cycle, but it may still be affected by a holiday that shifts dates, different numbers of weekdays, or changes in seasonal behavior. Follow the measure's methodology and use the version that matches the question.

Do not compare SA with NSA observations as though the gap were an economic gain or loss. Match the same series, units, frequency, population or sector coverage, and adjustment status. For inflation questions, also distinguish the consumer-price measures in CPI vs. PCE vs. the GDP deflator. For consumer spending, retail sales vs. PCE explains why two related releases can differ for reasons beyond seasonal adjustment.

Start with the exact series label and its source. Confirm whether the displayed value is seasonally adjusted, not seasonally adjusted, or unavailable in one of those forms. Then check what the number measures: a level, dollar change, percentage change, rate, index, or count. A table heading may also specify a monthly or quarterly frequency and an annualized presentation.

Next, match the comparison window to the question. Adjacent-month SA movement describes change after estimated seasonal effects; same-month NSA comparisons retain the seasonal cycle; a year-over-year percent change covers a full year; and SAAR restates a period pace at an annual rate under a specified convention. These labels are not interchangeable. A housing-start SAAR, GDP annualized growth rate, and monthly CPI percent change can all use annual language while applying different calculations.

Finally, read the source's technical note for revisions, holidays, seasonal status, and unusual events. When one report is meant to show the broad cycle, avoid treating one adjusted monthly change as a definitive turn. When a report is meant to compare recurring calendar periods, do not discard the NSA seasonal pattern. In short, identify the series, choose SA or NSA for the comparison, state the frequency and units, and preserve the data vintage.

If SA and NSA move in opposite directions, first confirm that they refer to the same series, frequency, units, and comparison window. Then check the estimated factor, moving holidays or weekday counts, any unusual event, and revisions to the source data. Explain the different question each version answers instead of labeling one of them an error.