A reader challenged me over a number I used in my recent article about American inflation and affordability.

I had compared August 2020 with August 2026. Consumer prices had risen by roughly 29 percent over that period. Average hourly earnings had increased by about 28 percent.

The comparison appeared straightforward.

Alexander, one of my readers, asked why I had started in August 2020.

He suggested February 2020 instead, the last full month before COVID-19 tore through the American labour market. Starting there changes the result. Average hourly earnings subsequently rose by more than consumer prices.

I went back to the Bureau of Labor Statistics data. His objection was legitimate.

More interestingly, I found that the six months separating those two starting points contained one of the strangest episodes in modern American labour statistics. Millions of people had disappeared from payrolls, many of them lower-paid workers. Their disappearance helped push the average hourly wage upward even though millions of Americans were experiencing one of the worst employment shocks of their lives.

My August starting point therefore deserved more scrutiny than I had originally given it.

What Happened Between February and August 2020?

February 2020 was the last full month before the pandemic severely disrupted the American labour market. Average hourly earnings for private nonfarm workers were about $28.55. The Consumer Price Index stood at 258.678.

Then COVID-19 arrived. Businesses shut down and employment collapsed. By April, the United States had lost more than 20 million payroll jobs in a single month.

Something counterintuitive happened at the same time. Average hourly earnings jumped.

The Bureau of Labor Statistics reported an increase of $1.34 in April 2020, taking average hourly earnings to $30.01. But the agency immediately warned against interpreting this as an ordinary surge in workers’ pay.

Lower-paid workers had suffered disproportionate job losses. Once many of those workers disappeared from payrolls, the remaining workforce contained a larger proportion of higher-paid employees. That pushed up the average. No mysterious national pay rise was required.

Imagine a workplace with five employees earning $10 an hour and another five earning $20. The average is $15. If the five lower-paid employees lose their jobs, the average among the remaining workers becomes $20 even though nobody received a raise.

The pandemic produced a much more complicated version of that arithmetic across the American economy. BLS later examined this composition effect in detail. Federal Reserve officials also warned repeatedly that pandemic employment changes made average wage measures unusually difficult to interpret.

By August, the Starting Line Had Already Moved

By August 2020, average hourly earnings were around $29.48. Compare that with roughly $28.55 in February.

On paper, average hourly earnings had increased by a little more than 3 percent in six months. But those were extraordinary months. Part of the apparent increase reflected a different mix of people being counted as employed.

This matters because I had used August 2020 as the starting point in my earlier article, America Says Inflation Is Falling. The Middle Class Is Asking: Then Why Can’t We Afford America?

Starting in August means beginning the comparison after some of the pandemic-related jump in average hourly earnings had already occurred. Starting in February captures it.

Neither calculation requires incorrect data. The problem appears when we treat the starting point as neutral without asking what happened immediately before it. In 2020, six months made an unusually large difference.

An Average Wage Is Not Quite the Same Thing as Workers’ Wages

Put wage growth beside inflation and the comparison looks wonderfully simple. If wages rise faster than prices, purchasing power improves. If prices rise faster, purchasing power falls.

As a broad economic relationship, that makes sense. But average hourly earnings contain another moving part: the workers included in the average.

When employment changes dramatically across different wage groups, the average can move even without corresponding changes in individual workers’ pay.

Economists therefore examine other measures as well. The Employment Cost Index, for example, controls more carefully for changes in the composition of employment. Researchers can also use measures that follow wage changes among individual workers.

The Federal Reserve has compared these measures precisely because no single wage statistic answers every question.

COVID-19 provided an extreme demonstration of why that matters. Some of the workers suffering the greatest economic damage during the first months of the pandemic stopped appearing in payroll calculations altogether. Their absence could make average wages look stronger at the very moment unemployment was exploding.

That doesn’t make average hourly earnings a bad statistic. It tells us what question the statistic can answer safely.

Workers Later Made Real Gains

Once I acknowledge the pandemic composition effect, I cannot simply use it to dismiss subsequent wage growth. American workers really did receive substantial wage increases during the recovery.

As businesses reopened, employers faced labour shortages in many industries. Competition for workers strengthened pay, particularly in parts of the lower-wage labour market.

Federal Reserve research documented unusually strong wage growth among lower-paid workers during parts of the recovery. The Fed’s July 2026 Monetary Policy Report says real wage growth was generally stronger for workers in the bottom half of the income distribution during much of the post-pandemic recovery through the first half of 2024. That advantage later weakened, although workers in the bottom quartile still recorded positive real wage growth on average over the preceding year.

Correcting one misleading interpretation shouldn’t produce another. The pandemic initially distorted average wage statistics through employment composition. Later, genuine wage increases also occurred.

This connects with another question I examined recently in Your Salary Rose 3%. Inflation Is 3%. So Why Do You Feel Poorer?

Aggregate wages can beat aggregate inflation while some households still experience deteriorating affordability. A renter and a homeowner do not necessarily encounter the same effective change in living costs. A long-distance commuter can face another pattern again. Nor do their incomes move together.

National averages remain useful. They simply cannot describe every household.

Watch the Words “Since 2020”

Alexander’s comment made me notice another problem that extends beyond this particular disagreement. We constantly encounter economic claims beginning with phrases such as “Since 2020…”

I now think that phrase deserves an immediate follow-up question: since when in 2020?

February describes America immediately before the pandemic shock. April captures the extraordinary collapse in employment. August belongs to an economy already moving through an uneven reopening.

Choosing among those months can materially change a wage comparison. The same problem appears elsewhere. Choose the bottom of a recession and subsequent economic growth can look spectacular. Start immediately before the downturn and the recovery looks different.

Inflation comparisons are particularly sensitive because the inflation surge itself developed over time. Starting before it, during it or after its peak answers different questions.

A percentage may therefore be perfectly accurate while the surrounding argument remains incomplete. That is why the baseline belongs in the analysis, not merely in the footnote.

There Is No Perfect Starting Date

I don’t think the lesson is that February 2020 is always correct and August 2020 is wrong.

February makes sense when the question concerns changes across the entire pandemic and post-pandemic period because it gives us a pre-pandemic benchmark. But another question may require another date.

If I want to examine the recovery after the initial shutdown, February may not be the most informative starting point. If I’m studying the recent inflation episode, I may want to begin closer to when inflation accelerated.

The better test is surprisingly simple. Move the starting date. Does the conclusion survive?

If changing the baseline by several months materially alters the argument, readers deserve to know. In this case, it did.

A Reader Made Me Recheck My Own Number

I could have answered Alexander by finding another statistic supporting the argument in my original article. There are plenty available.

Instead, I went back to the series he questioned. That was more useful.

My August 2020 comparison was mathematically valid, but I had not given enough attention to what had already happened to average hourly earnings between February and August. The pandemic had changed the composition of the workforce.

That doesn’t erase America’s affordability problem. Nor does it mean that wages failed to rise later. It means I was asking one statistic to carry more of the argument than it should.

There is also something useful here for those of us who write about economics. Charts encourage us to look at their right-hand edge. That’s where the newest number sits.

Perhaps we should spend a little more time looking at the left-hand edge.

That is where somebody decided the story should begin.