Mothers' Excellent Mental Health Fell 38.4% to 25.8%, 2016-2023

A mother is awake at eleven at night with the house finally quiet, running a search she has put off for months. The phrasing is some version of whether what she is feeling is the ordinary load of raising children or something that has a name and a treatment attached to it. What comes back is a statistic, repeated across a dozen sites in the same week: only about a quarter of American mothers now rate their mental health as excellent. It is a striking number, and it answers a question she did not ask.

The statistic comes from a study published in JAMA Internal Medicine in 2025, and the paper is a better piece of work than the headline treatment suggests — not because it explains why mothers feel worse, which it never attempts, but because it measures the shape of the change precisely enough to be argued with. The detail that goes missing in the coverage is where the movement actually went.

Researchers at Columbia's Mailman School of Public Health and the University of Michigan pooled eight years of the National Survey of Children's Health, a nationally representative annual survey, covering 2016 through 2023. The sample was 198,417 mothers, defined for this purpose as female biological or adoptive parents of children aged 0 to 17, weighted to represent roughly 42.1 million women nationally. Their mean age was 39.0 years and 89.8% were 30 or older. Of the mothers in the sample, 57.9% had a privately insured child, 35.7% a publicly insured child, and 6.4% an uninsured child.

The measure is a single question. Respondents rated their mental health, and separately their physical health, on a four-point scale: excellent, very good, good, or fair to poor. That is self-rated health — the respondent's own summary judgment of how she is doing — and it is worth being precise about what it is not. It is not a screening instrument of the kind a clinician would use, it carries no diagnostic threshold, and nobody in this study was assessed by anyone. Its value is that the same question has been asked the same way of a nationally representative sample every year, which is what makes a trend readable at all.

The design is pooled cross-sections rather than a cohort: different mothers answered in each survey year, so the study watches a population change rather than following individual women as they get better or worse. Regression models estimated the trends and the sociodemographic disparities, adjusting for secular changes in who mothers were across the period — the mix of ages, education levels and insurance types shifts over eight years, and a change in the composition of a population can otherwise masquerade as a change in its wellbeing.

The unadjusted movement is the clearest way in. Between 2016 and 2023 the share of mothers reporting excellent mental health fell from 38.4% to 25.8%. The share reporting good mental health rose from 18.8% to 26.1%. The share reporting fair or poor mental health rose from 5.5% to 8.5%.

Those three figures together say something the headline number cannot. Excellent lost about 12.6 points; good picked up about 7.3 of them and fair-or-poor about 3.0. The dominant movement is one rung down the scale, from a superlative to a solid answer, which is a real change and not a crisis. The smaller movement is the one at the bottom, where the band rose by more than half its starting size. Both are true at once, and a story that reports only the first understates the second while a story that reports only the second overstates the first.

After adjusting for the shifting composition of the population, excellent mental health decreased by 12.4 percentage points (95% CI, -14.0 to -10.7) across the eight years, and fair or poor mental health increased by 3.5 percentage points (95% CI, 2.6 to 4.4). Physical health moved in the same direction and far less steeply: excellent physical health fell by 4.2 percentage points (95% CI, -5.7 to -2.8), with unadjusted excellent ratings going from 28.0% to 23.9% and fair-or-poor physical health not changing significantly at all. Whatever happened to mothers over these eight years registered about three times as hard on the mental measure as on the physical one.

The last finding is two claims that are easy to collapse into one. The declines occurred broadly across socioeconomic subgroups — this is not a trend confined to poor mothers, and reading it as one would be wrong. But the levels differed sharply: self-reported physical and mental health were significantly lower among mothers who were US born, single parents, less educated, and those whose children were publicly insured or uninsured. The slope is general; the altitude is unequal.

It cannot tell anyone why. The study is descriptive by design and tests no explanation — not the pandemic, not economic strain, not social media, not the cost of childcare, not any of the causes the coverage reaches for within a paragraph. The authors are explicit that the next job is finding out, and they close by calling for investment to investigate and address the underlying causes. A trend this large with no mechanism attached is an invitation to project, and the paper supplies nothing to discipline the projection.

It also cannot separate a change in how mothers feel from a change in how mothers answer. Between 2016 and 2023 describing one's mental health as less than excellent became a more ordinary thing to do in the United States, and a shift in willingness to say so would appear in this data in exactly the same shape as a shift in wellbeing. The design offers no way to tell the two apart. That is not a reason to dismiss the finding — the adjusted decline is large and its confidence interval is nowhere near zero — but it is a reason to hold the interpretation more loosely than the number's precision invites.

And the sample is not who the phrase maternal mental health usually conjures. These are female parents of children aged 0 to 17, with a mean age of 39, not a perinatal or postpartum population; the study says nothing specific about the year after a birth. Nor does it follow anyone: different women answered each year, so no individual mother in this data got worse. The window closes in 2023, and the abstract reports mothers alone, without the like-for-like comparison with fathers that would settle whether this is something happening to mothers or something happening to everyone.

For the mother awake at eleven at night, the useful figure is not 25.8%. It is the 3.0 points at the other end of the scale, where fair or poor went from 5.5% to 8.5%, because that is the band in which the answer to her actual question is likeliest to be yes, this is worth taking somewhere. That band grew by more than half its size in eight years and it remains a small minority of mothers. The far larger group moved from excellent to good, which describes a population carrying more than it used to rather than a population in crisis.

The disparity finding is what a clinician is most likely to weigh. A decline that runs across every socioeconomic subgroup means the question is worth asking of any mother, not only the ones whose circumstances flag; the unequal levels mean the mothers who are single, US born, less educated, or whose children are publicly insured or uninsured started lower and stayed lower throughout. Those two facts point in different directions for screening and for resourcing, and the paper supports both.

What the study cannot supply is the route from a statistic to a person. In the United States, the 988 Suicide and Crisis Lifeline answers calls and texts at any hour and is not limited to emergencies, and primary care and obstetric clinicians are the usual first contact for a mother who wants a mental health question taken seriously. None of that was measured here: this study counted how mothers rate themselves, not whether they were asked, treated, or helped.

The honest reading of this paper is less alarming than the headline and more useful than it. Its headline figure — excellent mental health among US mothers falling from 38.4% to 25.8% between 2016 and 2023, an adjusted 12.4 percentage points — is real, large, and nowhere near its confidence bound, and it is mostly a story about a superlative answer being replaced by a merely solid one, because good absorbed about 7.3 of the 12.6 points that excellent gave up. The movement that carries the weight is the quieter one at the bottom of the scale, where fair or poor rose from 5.5% to 8.5%: proportionally the largest shift in the study, and the only band where a mother wondering whether what she feels is ordinary would find a different answer. That the decline ran across every socioeconomic subgroup while the levels stayed markedly unequal is the finding most likely to survive replication and most likely to be flattened in the retelling. What the study conspicuously does not have is a cause, and the speed with which the coverage supplies one — the pandemic, the phones, the economy — is a good measure of how much interpretation this number is being asked to carry. A single self-rated question asked well of 198,417 people is a thermometer, not a diagnosis. It says the temperature changed. It does not say why, and anyone who tells a mother it does is selling the explanation, not reporting it.