Abstract
The share underweight (BMI below 18.5) in the National Health and Nutrition Survey was lined up by sex and 6 age groups at ages 20+, joining the trend table for 2004–2017 with the tables of BMI status for 2017–2023 (18 years). The paper checks how much underweight among young women moved over the years, whether the year-to-year swings are explained by small numbers, and at which ages it changed. It claims no new mathematical theorem and no new law.
First, the tables are consistent. The two 2017 tables agree, and from the 2023 table the ministry's summary figures, "the share underweight among women in their 20s and 30s is 20.2%" (68 of 337 people) and 12.0% for all women, can be obtained as they are (Section 3).
Second, for women aged 20–29 the share was 22.68% in the first 5 years (2004–2008) and 21.14% in the last 5 rounds (2017–2023), and the slope over 18 years, −1.27 points per 10 years, is only 1.23 times its standard error. Ages 30–39 are also only +0.66 times; there is no clear trend (Section 4).
Third, this is the core of the paper. The year-to-year swings at ages 20–29 are explained by small numbers. The 5 rounds of 2017–2023, 21.7, 19.8, 20.7, 19.1, and 24.4%, seem to move, but only 119–177 people were counted each time, and the standard error expected from these numbers is 3.06–3.73 points. The sum of squared deviations ÷ squared standard errors is 1.32 (4 degrees of freedom), and the swing around the 18-year line, 2.44, is also smaller than this standard error (Section 5).
Fourth, what clearly rose was ages 50–69. Ages 50–59 went from 5.44% to 10.30% (1.89 times; slope 5.66 times its standard error) and 60–69 from 6.20% to 10.00% (1.61 times; 5.75 times), followed by 40–49 at 1.24 times (2.74 times). Women's total also rose by +1.20 points (Section 4).
Fifth, 0.665–0.687 of that rise had occurred by 2011–2015. Narrowing the window to the summary's 10 years (2013–2023), the slopes for women's total and ages 50–59 and 60–69 are smaller than twice their standard errors, which does not contradict the summary's "no significant change for women" (Section 6).
Sixth, what separates them is whether the share moved or swung because few people were counted. "Underweight among young women rose (or fell)" in one year can lie within the swing of a hundred-odd people.
Scope note: This paper claims no new mathematical theorem and no new law — the shares and counts are in public tables; the paper only compares averages of two periods, slopes of straight lines, and the swing expected from the numbers counted. This paper is a structural interpretation of the shape of numbers in published aggregate data, and is not medical advice, diagnosis, or a treatment recommendation. The author is not a physician. For decisions about individual health, consult a physician or medical institution. It is a preprint that has not been peer reviewed. It does not judge body shapes as good or bad, and says neither that being underweight is good nor that it is bad. It does not go into why the shares moved (diet, work, birth year, and so on). The standard error is a guide assuming simple sampling, without the effect of the survey design. The counts for 2004–2016 are not in the tables, so the swings were compared using the counts of 2017–2023. 2012 and 2016 are national-correction years, and 2024 is not used. Prefectures are not compared. It is not one person's outlook. About the data: the National Health and Nutrition Survey (e-Stat) is used under the Public Data License (Version 2.0), and the ministry's summary under the Public Data License (Version 1.0); this paper's averages, differences, slopes, and comparisons of swings were produced by processing them. What the paper adds: showing, with 18 years of age-specific tables, that the year-to-year swings in the share underweight among women aged 20–29 lie within the swing expected from the numbers counted, and that what clearly rose was ages 50–69.
1. Notation and question
- Share underweight: the share of people with BMI below 18.5 (pregnant women excluded).
- First 5 years: 2004–2008. Last 5 rounds: 2017, 2018, 2019, 2022, and 2023.
- Slope: points per 10 years of a straight line fitted to all 18 years. Those with slope ÷ standard error above 2 are written as a clear trend.
- Swing expected from the numbers counted: from the count-weighted average share p and each round's count n, √(p(1−p)/n) is taken as the standard error, and each round's deviation divided by it is squared and summed (4 degrees of freedom for 5 rounds; the 95% point is 9.49).
Question: How much did underweight among young women move over the years? Are the year-to-year swings explained by small numbers? At which ages did it change?
Observation 1.1: M050 showed, with the share with obesity in the same tables, that women's "flat" was the cancelling of movements in opposite directions by age. With the share underweight, this paper separates single-year swings from movement over several years.
2. Survey and design
To be honest: the design was decided after seeing the tables.
| Content | |
|---|---|
| Tables | National Health and Nutrition Survey 2017, Table 94-3 (trend in the share underweight, 2004–2017); 2017 Table 18-1 and 2018–2023 Table 17-1 (BMI status) |
| Years | 18 years: 2004–2019, 2022, and 2023 (the 2020 and 2021 surveys were cancelled) |
| Yardsticks | Averages of the first 5 years and last 5 rounds, the slope over 18 years, the swing expected from the numbers counted, and the slope with the window narrowed to 2013–2023 |
| Noticed while surveying | Women aged 20–29 swung widely from year to year but did not move in averages over several years, while ages 50–59 and 60–69 nearly doubled |
| Expected and measured | I expected the 18-year slope for women aged 20–29 to be smaller than its standard error, but it measured 1.23 times (below 2, so it is not written as a clear trend) |
3. Data and checks
Calculation 3.1 The 7 retrieved files (6 tables and the 2023 summary) were fixed by SHA-256, and the following was checked.
- The trend table is Table 94-3, with the total, age-adjusted value, and 6 age groups for each sex over the 14 years 2004–2017.
- The 2017–2023 tables of BMI status have the counts and shares underweight for the 6 age groups and ages 20+ for each sex, with share = count ÷ total. The 2017 shares agree with 2017 in the trend table within 0.05.
- The 2023 summary states that the share underweight is 12.0% for women, "over these 10 years no significant change is seen for women," and "the share underweight among women in their 20s and 30s is 20.2%." In the 2023 table, women aged 20+ are 12% and women aged 20–39 are 20.2% (68 of 337 people), in agreement.
4. First 5 years and last 5 rounds, and the 18-year slope
Calculation 4.1 Share underweight among women (%)
| Age | First 5 years | Last 5 rounds | Times | Slope (points per 10 years) | Slope ÷ SE |
|---|---|---|---|---|---|
| 20–29 | 22.68 | 21.14 | 0.93 | -1.27 | -1.23 |
| 30–39 | 15.94 | 16.36 | 1.03 | +0.57 | +0.66 |
| 40–49 | 8.90 | 11.08 | 1.24 | +1.61 | +2.74 |
| 50–59 | 5.44 | 10.30 | 1.89 | +3.27 | +5.66 |
| 60–69 | 6.20 | 10.00 | 1.61 | +2.90 | +5.75 |
| 70+ | 9.58 | 9.32 | 0.97 | -0.06 | -0.13 |
| Total (20+) | 10.06 | 11.26 | 1.12 | +0.92 | +3.47 |
Calculation 4.2 Share underweight among men (%)
| Age | First 5 years | Last 5 rounds | Times | Slope (points per 10 years) | Slope ÷ SE |
|---|---|---|---|---|---|
| 20–29 | 9.06 | 10.24 | 1.13 | +0.79 | +0.87 |
| 30–39 | 3.38 | 4.10 | 1.21 | +0.52 | +1.22 |
| 40–49 | 2.40 | 2.20 | 0.92 | +0.03 | +0.11 |
| 50–59 | 3.56 | 2.98 | 0.84 | -0.19 | -0.49 |
| 60–69 | 3.16 | 3.04 | 0.96 | -0.15 | -0.50 |
| 70+ | 6.68 | 4.70 | 0.70 | -1.45 | -2.88 |
| Total (20+) | 4.44 | 4.06 | 0.91 | -0.22 | -1.68 |
Observation 4.3: For women aged 20–29 the share went from 22.68% to 21.14%, and the slope is only 1.23 times its standard error (downward). Ages 30–39 are 0.66 times. What clearly rose was ages 50–59 (1.89 times; slope ÷ SE 5.66) and 60–69 (1.61 times; 5.75), followed by 40–49 (1.24 times; 2.74). For men, the only group with slope ÷ SE above 2 in size was 70+ (0.70 times; −2.88).
Figure 1: Share underweight by age group. Light blue is the average of 2004–2008 and navy the average of the 5 rounds of 2017–2023.
5. Core — the swings of young women lie within small numbers
Calculation 5.1 Swings in the 5 rounds of 2017–2023 and the swing expected from the numbers counted (women)
| Age | Shares in the 5 rounds (%) | Counted | Expected SE | Squared deviations ÷ squared SE |
|---|---|---|---|---|
| 20–29 | 21.7, 19.8, 20.7, 19.1, 24.4 | 119–177 | 3.06–3.73 | 1.32 |
| 30–39 | 13.4, 19.3, 16.4, 14.8, 17.9 | 196–295 | 2.16–2.65 | 4.23 |
| 40–49 | 10.6, 10.5, 12.9, 9.7, 11.7 | 274–436 | 1.50–1.89 | 2.13 |
| 50–59 | 10.1, 10.4, 10.6, 9.5, 10.9 | 347–452 | 1.43–1.63 | 0.45 |
Observation 5.2: The 5 rounds for women aged 20–29, 21.7, 19.8, 20.7, 19.1, and 24.4%, seem to move, but only 119–177 people were counted, and the standard error expected from these numbers is 3.06–3.73 points. The sum of squared deviations ÷ squared standard errors is 1.32, far below the 95% point of 9.49 for 4 degrees of freedom. The swing (standard deviation) around the 18-year line, 2.44, is also smaller than this standard error. The rises and falls of single years lie within the smallness of the numbers counted.
Figure 2: Share underweight among women aged 20–29, 50–59, and 60–69, with the line fitted to ages 20–29 (dashed). The horizontal axis shows the last two digits of the survey year. Grey marks the cancelled years 2020 and 2021.
6. The summary's 10 years and when the rise happened
Calculation 6.1 Slopes with the window narrowed to 2013–2023 (9 rounds; women)
| Age | Slope (points per 10 years) | Slope ÷ SE |
|---|---|---|
| Total (20+) | +0.35 | +0.49 |
| 20–29 | +1.72 | +0.82 |
| 40–49 | +0.35 | +0.34 |
| 50–59 | +1.58 | +1.30 |
| 60–69 | +3.02 | +1.84 |
Observation 6.2: Narrowed to 2013–2023, the slopes for women's total and ages 50–59 and 60–69 are smaller than twice their standard errors. Ages 50–59 were 5.44% in 2004–2008, 8.78% in 2011–2015, and 10.30% in the last 5 rounds, and ages 60–69 were 6.20, 8.73, and 10.00%; 0.687 and 0.665 of the rise had occurred by 2011–2015. The summary's "no significant change for women" refers to this window and does not contradict the rise seen over 18 years.
7. Limitations
- The standard error is a guide assuming simple sampling, without the effect of the survey design. The real swing can be larger, so "explained by the numbers counted" is a cautious statement.
- The counts for 2004–2016 are not in the tables, so year-to-year swings were compared using the counts of 2017–2023.
- The slope of a straight line is the average direction over 18 years and does not show the shape in between (rising and then levelling off, and so on).
- 2012 and 2016 are national-correction years (used as in the trend table).
- It does not go into why the shares moved or whether body shapes are good or bad. Prefectures are not compared. It is not one person's outlook.
Conclusion
Over 18 years of the National Health and Nutrition Survey, the share underweight among women aged 20–29 went from 22.68% to 21.14%, with no clear trend. Its year-to-year swings lay within the swing expected from counting a hundred-odd people. What clearly rose was ages 50–59 (1.89 times) and 60–69 (1.61 times), mostly by around 2015.
One thing separates them — whether the share moved or swung because few people were counted.
When you hear that "underweight among young women rose (or fell)," check together how many people were counted that year, and whether it also moves in averages over several years.
Data and reproduction
| Data | National Health and Nutrition Survey (Ministry of Health, Labour and Welfare, e-Stat), 2017 Tables 94-3 and 18-1, and Table 17-1 for 2018, 2019, 2022, and 2023; Summary of Results of the 2023 National Health and Nutrition Survey (retrieved 26 September 2026) |
| Terms of use | e-Stat follows the Public Data License (Version 2.0), and the ministry's pages follow the Public Data License (Version 1.0). This paper's averages, differences, slopes, and comparisons of swings were produced by processing these and are not figures produced by the government |
| Scope | Japan; ages 20+; 2004–2019, 2022, and 2023 |
| Verification | 検算/M051.py (--zu also writes the figures). Every number in the text is taken from 検算/M051_output.txt |
| Raw data | The 7 retrieved files and SOURCES.md (URLs, retrieval dates, SHA-256, terms of use, definitions and calculation) in data/ |
| Figures | 図/M051_zu1 and 図/M051_zu2 (png, svg, English version) |
References
- Ministry of Health, Labour and Welfare. National Health and Nutrition Survey 2017, Tables 94-3 and 18-1; 2018, 2019, 2022, and 2023, Table 17-1 (in Japanese). e-Stat (accessed 26 September 2026). https://www.e-stat.go.jp/
- Ministry of Health, Labour and Welfare. Summary of Results of the 2023 National Health and Nutrition Survey (in Japanese) (accessed 26 September 2026). https://www.mhlw.go.jp/content/10900000/001338334.pdf
- Yamagishi, Y. (2026). The Share with Obesity Rose in All 6 Age Groups for Men, While Women's Age-Adjusted Share Is Flat at −0.09 Points — for Women, +2.80 at Ages 20–29 and −2.36 at 60–69, Ages That Rise and Ages That Fall Were Cancelling Out [M050]. Zenodo. https://doi.org/10.5281/zenodo.22974804
Reference
Yamagishi, Y. (2026). The Share Underweight Among Women Aged 20–29 Shows No Clear Trend over 18 Years (2004–2023), and Its Year-to-Year Swings Are Explained by Small Numbers [M051]. Zenodo.
DOI: 10.5281/zenodo.22974868
YOI: https://www.yamagishiyuuki.com/yoi/M051
ORCID: 0009-0006-8587-3450
PDF (English) · PDF (Japanese) · Date proof of the English PDF (OpenTimestamps .ots) · SHA256SUMS
The record page for this paper (in Japanese, with the PDFs in each language and the date proof): YOI M051. This paper is a preprint and has not been peer reviewed.