Economic reporting problems: how to read releases

Economic numbers arrive in the news almost daily, and the sentences around them often say more than the figures do. Most economic reporting problems are small: a percentage point becomes a percent, a slowing rate becomes falling prices, an advance estimate is read as a final count. Each of these has a fix, and the fixes are short.

What to take away

The releases behind the headlines

Most of the numbers that reach the news come from a short list of agencies. The Bureau of Labor Statistics publishes the monthly Employment Situation — the jobs report — along with the Consumer Price Index, the Producer Price Index and the Job Openings and Labor Turnover Survey. The Bureau of Economic Analysis publishes gross domestic product, including the advance estimate, and the PCE price index. The Census Bureau publishes the retail sales report and other business surveys, and the Federal Reserve publishes industrial production. Weekly unemployment insurance claims come from the Labor Department.

Two of those measures are aggregates, and aggregates carry a weighting rule. Real GDP and its components use chain-weighted indexes, so the parts do not add up exactly to the total the way fixed-weight figures would. A contribution presented as if it summed is an arithmetic claim the data do not support.

The words that carry the claim

A missing word changes the number. Annualized, real, average and preliminary each do work, and dropping one turns a true figure into a false sentence. Annualized means a rate converted to a twelve-month basis, real means adjusted for price changes, average means an aggregate divided by the population it covers, and advance or preliminary means a first pass, with revisions still to come.

Slower inflation is not falling prices

A slowing inflation rate gets written as prices coming down, which is not the same thing. The Consumer Price Index can rise more slowly, hold flat or fall, and those are three different sentences. Report the rate and the index direction separately. The distinction is worked through in translating inflation to household context.

Percentage points are not percent

A rate moving from 4 percent to 5 percent is not a 1 percent increase. It is a 1 percentage point rise, which is a 25 percent increase against the starting rate — divide the change by the base, then say which of the two you mean. Pick the expression that answers the question the number is meant to answer.

Periods that do not belong on one chart

A monthly rate, a quarterly annualized rate and a year-over-year rate are not directly comparable. Annualizing a monthly rate means compounding twelve months, which is not the same as multiplying the monthly rate by twelve. Convert only where the method allows, name the method, and show the calculation. Separate panels or a table with explicit periods is the safer layout.

Revisions that disappear below the fold

When a large revision changes the recent pattern, it belongs in the opening, not in a footnote. The jobs report and the GDP estimates both work this way: an early figure is published, then revised as more complete data arrive. Keep an old vintage only to show how estimates evolved. Health desks handle this with status labels; compare health evidence and alert types.

Dollar sales are not units sold

Higher dollar sales do not mean more goods purchased until someone adjusts for price. State the release's own adjustment. The Census Bureau's Advance Monthly Sales report labels its estimates as adjusted for seasonal, holiday and trading-day effects but not for price changes. The figures on that page change monthly; what holds is the measurement distinction.

Read the category definition too. Retail-establishment data group sales by the primary business of the establishment, which can differ from the product a reader pictures. A store category sells many kinds of goods, and one product moves through several categories. Do not convert one classification into the other without data behind it.

Check the current and prior vintages before describing momentum. An advance estimate can be superseded by a larger sample, and a benchmark can rewrite a longer history. A chart that silently mixes vintages invents a change at the join.

Average earnings are not one worker's raise

An average increase applied to every worker or household is a different claim from the one the release makes. Name the employee population, whether the measure is hourly or weekly, the composition effects, and the price deflator.

The BLS real earnings technical note explains that its constant-dollar series combine Current Employment Statistics earnings with CPI deflators, using different CPI series for specified employee groups. That is a defined aggregate, not a payroll record for each worker.

A release describes; it rarely explains

One category moving in the jobs report or the CPI is not evidence that policy, weather or sentiment caused it. Keep description and explanation in separate sentences. A causal claim needs additional data, timing, a mechanism and research. Policy attribution has its own trap list; see government decision reporting problems.

The forecast is one comparison, not the verdict

A jobs report or an inflation reading called good or bad only because it beat or missed consensus tells the reader almost nothing. Report the forecast as one comparison among several: the measure's level, its trend, its revision history, its uncertainty, and what it means for households or businesses.

Charts and examples that outlive their release

A chart that refreshes itself while the text describes an older release is two articles in one file. Archive the data used for the conclusion. The St. Louis Fed's ALFRED keeps vintages of many series, and the Bureau of Labor Statistics and the Census Bureau run their own retrieval tools; record the release the numbers came from, not just the day they were pulled.

A hypothetical budget placed beside government data reads as official unless it is labelled. Mark it as an example, publish the inputs and the arithmetic, and say plainly that it does not estimate a typical household.

Common questions

Is beating a forecast the main story?

Not usually. Forecasts matter to markets, but readers also need the measure, the trend, the revisions and the household or business context. A miss against consensus can sit beside a level that is still rising.

Can seasonally adjusted and unadjusted data appear together?

Yes, with a stated reason and clear labels. Do not subtract one from the other or compare them as though they were a single series. They answer different questions about the same month.

Does a revised estimate mean the agency made a mistake?

Not by itself. Many programs publish an early estimate and fold in more complete data later, which is the design rather than an error. Report the size and direction of the revision and what it changes about the recent pattern.

When can a release support causal language?

When its design or supporting evidence identifies a cause, not merely simultaneous movement. Otherwise attribute the claim to the source making it and say what evidence would settle the question.