A post spread widely on social media with a claim along these lines:
“More than half of Japanese people earn ¥2.5 million or less a year.”
Attached was a table showing rows such as “under ¥500,000: 30.4%” and “¥2.0–2.5 million: 4.7%.” Add the brackets up to roughly ¥2.5 million and you get 51.3%. On its face, that is a powerful number.
The awkward part is that the numbers themselves are basically real.
This is not a fabricated chart. The figures match Japan’s Ministry of Health, Labour and Welfare (MHLW) 2023 Income Redistribution Survey report.[1]
The problem is not the number. The labels have gone on a little vacation.
“Households” becomes “Japanese people.” “Initial income” becomes “annual income.” And public pensions quietly leave the frame.
This is less like inventing statistics and more like product photography for information.
Put the real product on the desk.
Move the inconvenient objects to the right.
Add lighting.
Choose the flattering angle.
Place a “BAD NEWS” card in the foreground.
Click.
The product is real. The photo is unedited.
The rest of the room simply is not in the picture.
0. The five-second answer
This table does not show that “more than half of working Japanese people earn ¥2.5 million or less.”
There are five reasons.
- The unit is households, not individuals.
- The measure is not ordinary salary income but initial income before social redistribution.
- Public pensions and public pension-type benefits are not counted in initial income; they are treated separately as social-security benefits.
- The survey is heavily shaped by an older household population: 56.7% of household heads were age 65 or older.
- The cumulative share around ¥2.5 million is 51.3% for initial income, but 29.5% for redistributed income.[1]
And redistributed income is not the same as cash take-home pay either; it includes in-kind benefits such as medical and long-term care.
So this is not a ranking of Japanese workers’ salaries.
It is a statistic designed to show how household income distributions change before and after taxes and social-security redistribution.
1. First: the 51.3% figure is real
This part should not be lazily “debunked.”
MHLW’s Table 1 gives the household distribution of initial income as follows:[1]
- Under ¥500,000: 30.4%
- ¥500,000–1 million: 5.8%
- ¥1.0–1.5 million: 5.6%
- ¥1.5–2.0 million: 4.8%
- ¥2.0–2.5 million: 4.7%
Cumulative total: 51.3%.
So the problem is not “51.3% is fake.”
51.3% is real. The subtitle has wandered into a different program.
That distinction matters, because a sloppy fact-check can become just another sloppy post.
2. The biggest conversion is “households” → “Japanese people”
The original table is a household distribution.
It summarizes 3,003 households and classifies each household into an income bracket.[1]
On social media, that often becomes:
“Half of Japanese people.”
That is a different unit.
A two-person married household, an older person living alone, and a four-person family each count as one household. This is not a population share.
Then there is the word “annual income.” Many readers will naturally hear that as a worker’s annual salary.
But the report’s label is initial income.
MHLW defines initial income as the sum of employee income, business income, agricultural income, property income, home-work income, miscellaneous income and private transfers such as remittances, corporate pensions and life-insurance payments.[1]
So it is not “salary” in the first place.
And an even more important category sits outside it.
3. Public pensions live outside “initial income”
MHLW treats social-security receipts separately, including pensions, medical care, long-term care and other benefits.[1]
Add cash social-security benefits to initial income and you get “gross income.” Subtract taxes and social-insurance contributions and you get disposable income. Add in-kind benefits such as medical and long-term care and you get redistributed income.[1]
A retired household makes the point obvious.
Suppose a household has:
- Salary: ¥0
- Business income: ¥0
- Public pension: ¥2 million a year
If it has no other income counted as initial income, it can sit near zero initial income even while receiving ¥2 million in public pension payments.
A household living on ¥2 million can therefore appear in the “under ¥500,000” band in the initial-income table.
That is not a bug in this survey.
It is the point of the survey: to measure how much redistribution changes the distribution.
One nuance: corporate pensions and other private transfers can be included in initial income. “All pensions are excluded” would be a second misleading simplification.
4. The survey is also heavily shaped by older households
MHLW explicitly notes that 56.7% of household heads were age 65 or older in this survey, up from 55.2% previously, and that ageing affected the rise in the Gini coefficient for initial income.[1]
There were also 1,232 “elderly households” among 3,003 surveyed households — about 41%.[1]
Their averages make the structure clearer:
- Average initial income: ¥1.078 million
- Pensions and public pension-type benefits: ¥1.938 million
- Average disposable income: ¥2.637 million
- Average redistributed income: ¥3.384 million[1]
If you crop out everything except “initial income: ¥1.078 million,” the picture looks extremely bleak.
But on the same row sits ¥1.938 million in pensions and related benefits.
Pension: “I’m literally right here.”
Crop: “Not in today’s shot, please.”
None of this means older people have easy lives. Housing costs, assets, health expenses and household size all matter.
The point is narrower:
This table is not a valid proxy for the salary distribution of active workers.
5. The 30.4% becomes 1.2% after redistribution
The survey exists precisely to compare income before and after redistribution.
So MHLW places initial income and redistributed income side by side in the same Table 1.[1]
Households below ¥500,000:
Initial income: 30.4%
→ Redistributed income: 1.2%
Cumulative share around ¥2.5 million:
Initial income: 51.3%
→ Redistributed income: 29.5%[1]
30.4% to 1.2%.
Taking only the first number is like clipping the first ten minutes of a movie, stopping at the catastrophe, and publishing:
“Hero completely defeated.”
The social-security act has not started yet.
Again, redistributed income includes in-kind benefits such as health care, long-term care and childcare. It does not mean that ¥3.384 million in redistributed income appears as ¥3.384 million in a bank account.
The lesson is not “redistributed income is the true salary.”
The lesson is:
Read the label together with the number.
6. Is it because students are included? Partly possible, but not the main story
Students are not completely absent.
Because the survey covers ordinary households, independently living student households can potentially be included. But MHLW’s definitions exclude people who are away from the household for schooling from that household’s members, and certain single-person households living in catered dormitories or lodging houses are excluded from the survey population.[1]
So:
“The low-income side is all students”
is also wrong.
The primary structural factors visible in the official source are ageing and the treatment of social-security benefits.
The combination of 56.7% household heads aged 65+ and public pensions being outside initial income is too large to ignore.[1]
Students may be supporting actors.
A very large pension-age cast is standing center stage.
7. The source article had already explained the pension issue immediately before the table
The Diamond Online article carrying this table was published on January 8, 2026.[3]
Just before the table, it explains that the fall in initial income is likely related to ageing and that public pensions are not included in this income measure, so a household relying only on public pension can have zero pre-redistribution income in this framework.[3]
Then comes the 51.3% table.
That sequence matters.
The original article itself uses a strong framing about half of households being at or below ¥2.5 million. But it is not presenting the figure as “half of working Japanese people earn salaries below ¥2.5 million.”
Its argument is about an ageing society, households leaving the labour market, and how pre-redistribution income looks.[3]
When a social-media caption turns the table into “Japanese people’s annual income,” the most important instruction manual disappears.
The box is real.
The contents are real.
The manual was thrown away before the review.
8. This is “product photography for information”
We cannot know from the outside whether the person who posted the image actually read the source article or found the image somewhere else.
So we cannot claim:
“They knowingly hid the pension caveat.”
But if we describe only the final editing effect as product photography, it looks like this:
- Put a genuine government table in the center.
- Replace the unit “households” with “Japanese people” in the caption.
- Replace the technical label “initial income” with “annual income.”
- Move the explanation that public pensions are separate out of frame.
- Push the 2022 income-reference period into the background.
- Add “BAD NEWS.”
- Shoot.
Done.
“Half of Japanese people earn ¥2.5 million or less.”
Not one number had to be forged, yet the meaning delivered to the reader changes dramatically.
That is the interesting and dangerous part.
Statistical misinformation does not require fabricated data.
Move the subject, unit, definition, period and context out of frame one by one.
9. Is that worse than an AI hallucination?
OpenAI describes hallucinations as plausible but false statements generated by language models.[4]
With humans, an additional question appears: intent.
UNESCO commonly distinguishes:
- False information shared without an intent to deceive: misinformation
- False or misleading information spread with an intent to deceive: disinformation[5]
We do not know which applies here.
At least two scenarios are possible.
A. The source article was not properly read
Then the statistic was amplified without understanding what it measured.
B. The source article was read
Then the pension explanation immediately before the table may have been seen.
But B does not prove deliberate concealment. The person may have misunderstood it, forgotten it, saved only the image, or made some other mistake.
What can be established externally is narrower: the social caption and the meaning of the original statistic do not match.
So there is no need to inflate the ethical accusation beyond the evidence.
The information-quality judgment is easier:
If it was not checked, that is sloppy.
If it was checked and the meaning was knowingly changed, that is more serious.
AI systems are told to say “I don’t know” when uncertain.
Humans could use the same guardrail.
10. A five-second checklist for viral statistics
No advanced statistics degree is required.
Check six things.
1. Who is the subject?
Individuals, households, companies, respondents?
2. What is the unit?
People, households, percentages, means, medians?
3. What is the formal definition?
“Income,” “annual salary,” “disposable income” and “initial income” are not interchangeable.
4. What is included — and excluded?
In this case, public pensions were decisive.
5. What period does the data cover?
This is called the 2023 survey, but the income reference period was January 1 to December 31, 2022.[1]
6. What is written immediately before and after the table?
Here, one of the biggest caveats was placed directly before the table.[3]
Before doubting the number itself,
check the number’s name tag.
That prevents a surprising number of accidents.
11. The genuinely interesting “reality of Japan” is elsewhere
The dataset is actually fascinating.
56.7% of household heads were 65 or older. Elderly households had average initial income of ¥1.078 million, but the picture moved substantially once pensions, medical care, long-term care and other redistribution were counted.[1]
That is not a story saying:
“Half of Japan has become low-paid workers.”
It is a story saying:
“In an ageing society, a distribution based only on market and pre-redistribution income can look extreme, and social-security transfers substantially change the picture.”
That is a much richer fact about Japan’s population structure and institutions.
But once the table enters social media:
Pensions to the right.
Households to the left.
2022 into the background.
“Initial income” under the shelf.
Good.
51.3% in the center.
Click.
“BAD NEWS: JAPANESE PEOPLE—”
Information can mislead without lying about a single number.
Whether the number is genuine is only the first question. The subject, unit, definition, exclusions and period are part of the data too.
