What the library can establish.
Each record is a published finding with a named measure, population, and stated limit on generalization.
The library keeps the 38 repeated series and eight historic snapshots separate. Each finding retains its definition, population, and limits of inference.
The repeated library contains national estimates produced by seven federal systems, one Historic system and one annual survey that a professional association commissions from a polling firm. The historic mode contains selected non-series records whose actual scope and intended generalization can be reconstructed from public sources. Every display states which kind of claim its evidence can support.
The library is not a synthetic prevalence estimate, a ranking of instruments, a causal explanation for change, a census of every relevant dataset, or evidence that national coverage automatically creates valid national inference.
Four evidence states
Comparable series. At least two published national observations retain a sufficiently stable construct and threshold, instrument, population, recall window, survey mode, and field frame. A chart is allowed, but documented survey breaks interrupt the line.
Study snapshot. A data-producing study supplies national statistics but no repeated like-for-like series. It gets a profile of producer, population, threshold, recall period, fieldwork, and primary table, but no temporal axis.
Historic snapshot. A record from 1840–2004 materially changed how mental health was counted or generalized. It receives an inference class, population frame, finding register, boundary statement, archival rationale, and linked source record, but no connected time axis.
Comparison example. A historically important headline figure is useful for teaching non-comparability but does not qualify for the repeated library. It remains in the comparison ledger and is excluded from the 509-estimate total.
Producer-only provenance
A named organization must have generated the evidence. Designing the measure or sample, fielding interviews or examinations, registering records, processing or weighting responses, compiling jurisdictions, or directly producing the national analysis qualifies. Funding alone, hosting a copy, publishing a report, or curating another organization’s statistics does not. Publisher-only collections are excluded even when their summaries are useful.
How definitions and claims expand.
This framework distinguishes a change in diagnostic eligibility from a change in the population covered by a public claim. It records the kind of change without estimating its contribution to prevalence.
Broader eligibility within a category
Vertical expansion. Haslam uses this term for the inclusion of less severe or less extreme cases. The comparison identifies the threshold, duration, impairment requirement, and exclusions before and after a change. Removing the DSM bereavement exclusion is an eligibility-rule change; it is not literally a lower symptom score. A symptom-screen threshold and a clinical diagnosis also remain different case rules.
A claim reaches more populations
Population or domain expansion. The archive follows when an established crisis claim is applied to groups such as young children, LGBTQ youth, college students, Asian Americans amid pandemic-era racism, or post-Roe patients. A new population alone is not evidence of Haslam’s horizontal concept creep: that term concerns a qualitatively new kind of phenomenon or context. Each interpretation rests on the population, speaker, wording, date, and evidence of the claim.
What the selected measures leave out
Measurement selection. This library displays selected household-survey estimates and vital-statistics records. A survey question about treatment use differs from an insurance record of an authorized encounter. No claims-based series is joined to a symptom-screen series. The selected chart measures do not directly establish the effects of housing cost, caregiving, working conditions, discrimination, or isolation; that does not imply that the underlying surveys never collect related covariates.
Reading these comparisons. The six checks ask what is measured and what counts as a case; which questions or instrument are used; who is represented; what period respondents recall; how responses are collected; and who can enter the sample. A rising count can reflect changes in lived conditions, eligibility, reporting, or measurement. This archive does not identify the causal share of any one explanation.
Framework source. Nick Haslam (2016), Concept Creep: Psychology’s Expanding Concepts of Harm and Pathology.
Depression across the DSM editions
The chapter follows selected depression descriptions and adult episode rules through eight publications, using the manuals and the association’s revision notes. Because DSM-I and DSM-II distribute depression across several categories, the sequence does not translate their patients into a single modern diagnosis or estimate how many additional people each revision would include. An episode is also distinct from a disorder: the full diagnosis considers episode history, mania or hypomania, other conditions, and exclusions.
DSM-I, 1952. The original manual’s printed pages 24–25 distinguish manic-depressive and psychotic depressive reactions; pages 33–34 describe depressive reaction within psychoneurotic disorders. The chapter paraphrases their explanations of loss, guilt, and personality, rather than applying a later checklist to them. American Psychiatric Association, DSM-I, pp. 24–25 and 33–34.
DSM-II, 1968. Printed pages 36–38 distinguish the depressed type of manic-depressive illness and psychotic depressive reaction, while page 40 describes depressive neurosis in relation to conflict or loss. These are descriptions with clinical distinctions, without a common symptom-count or duration rule. American Psychiatric Association, DSM-II, pp. 36–40.
DSM-III, 1980. Printed pages 213–214 separate criterion A, low mood or loss of interest, from criterion B, at least four accompanying symptoms for two weeks in adults; loss of interest also appears in the accompanying list. The chapter therefore preserves that structure instead of calling it the later five-of-nine rule. The manual excludes uncomplicated bereavement and describes dysthymic disorder separately. American Psychiatric Association, DSM-III, pp. 213–214 and 220–223.
DSM-III-R, 1987. The revised criteria place nine symptoms in one list and require five within the same two-week period, including depressed mood or loss of interest or pleasure. The source used here is a credited reproduction of the APA criteria in the Institute of Medicine’s 1994 report, rather than an independently reconstructed checklist. Reducing Risks for Mental Disorders, table 5.3, p. 89.
DSM-IV, 1994. Criterion C adds an explicit distress-or-impairment requirement to the major depressive episode criteria. Criterion E retains a qualified bereavement exclusion: an episode can be diagnosed after a death if it lasts beyond two months, or if marked functional impairment, preoccupation with worthlessness, suicidal ideation, psychotic symptoms, or psychomotor retardation is present. The chapter names examples of these exceptions without treating recent bereavement as an absolute bar. American Psychiatric Association, DSM-IV, p. 327.
DSM-IV-TR, 2000. First and Pincus, who worked on the revision, describe an update to the explanatory text, research, and codes, with a small number of criteria corrections. The major depressive episode rule and its bereavement exclusion remain, so the chapter distinguishes revised descriptions from a changed threshold. First and Pincus (2002), The DSM-IV Text Revision.
DSM-5, 2013. The APA’s change register states that the core episode symptoms, two-week duration, and distress-or-impairment requirement are unchanged, while the bereavement exclusion is removed and dysthymia and chronic major depression are combined under persistent depressive disorder. Removing the exclusion allows a diagnosis after a loss when the full criteria are met; it does not make ordinary grief a depressive disorder. American Psychiatric Association, Highlights of Changes, pp. 4–5.
DSM-5-TR, 2022. The APA’s major depressive disorder revision note explains the clarification to criterion D, which specifies how at least one qualifying depressive episode relates to schizoaffective and other psychotic disorders. This is a clarification of that exclusion, with the episode’s symptom count and duration preserved. The separate addition of prolonged grief disorder is not a renaming of depression. American Psychiatric Association, Major Depressive Disorder revision note.
Edition context. The APA’s DSM History supplies the broader development sequence, including DSM-IV’s evidence review and coordination with ICD-10 and the scope of DSM-5-TR’s text reviews. The edition names in Measurement lead to the corresponding Timeline publication records.
Anxiety across the DSM editions
This trace follows generalized anxiety and its earlier descriptive categories, rather than treating every anxiety disorder as one condition. The DSM-I and DSM-II labels also cover experiences later distinguished as panic disorder. Chapter reorganizations therefore mean something different from a changed symptom or duration requirement.
DSM-I and DSM-II. DSM-I describes anxiety reaction on printed page 32, while DSM-II describes anxiety neurosis on page 39. Neither passage supplies the later fixed duration rule. Sources: DSM-I, p. 32 and DSM-II, p. 39.
DSM-III through DSM-IV. DSM-III’s adult criteria require one month and symptoms from three of four groups; DSM-IV requires six months, difficult-to-control worry, and three of six accompanying symptoms in adults, or one in children. The intervening six-month DSM-III-R rule is documented in the Semi-Structured Assessment for the Genetics of Alcoholism interview manual, section P, which explicitly identifies the diagnostic systems it implements. Sources: DSM-III, pp. 232–233; SSAGA interview manual, generalized anxiety section; and DSM-IV, pp. 435–436.
DSM-IV-TR and DSM-5. The text revision retains the core DSM-IV rule; the rationale for text revisions is documented in First and Pincus (2002). The APA’s 2013 changes register, pp. 5–7, documents the separate chapters for obsessive-compulsive and trauma-related disorders and the revised placement and age coverage of separation anxiety.
DSM-5-TR. The six-month duration and accompanying-symptom requirements are checked against the DSM-5-TR criteria reproduced in Veterans Affairs Canada, Anxiety Disorders, pp. 3–4 (2025). Only the reproduced diagnostic criteria support this reading; the Canadian benefit-entitlement rules elsewhere in that document are not DSM criteria.
ADHD across the DSM editions
The trace begins with the absence of a separately named attention-deficit or hyperkinetic diagnosis in DSM-I. It does not retrospectively classify the manual’s broad childhood categories as ADHD. Subsequent labels differ in their emphasis on activity, attention, age, and qualifying symptom combinations.
DSM-II and DSM-III. DSM-II’s brief hyperkinetic-reaction description appears on printed page 50. DSM-III’s attention deficit disorder criteria distinguish forms with and without hyperactivity, with onset before seven and a six-month duration. Sources: DSM-II, p. 50 and DSM-III, pp. 43–44.
DSM-III-R and DSM-IV. The eight-of-fourteen DSM-III-R list is credited to the original APA manual in the SUNY Cortland reproduction of the 1987 criteria. DSM-IV’s two lists and three subtypes appear in DSM-IV, pp. 83–85. Its onset rule concerns symptoms causing impairment before age seven; the chapter preserves that distinction when describing the later change.
DSM-IV-TR and DSM-5. The text revision retains the core symptom, duration, and age rules. The APA’s 2013 ADHD fact sheet documents the shift to symptoms before age twelve, a five-symptom minimum from age seventeen, examples for adults, and presentations in place of subtypes. These are selected changes, not a complete diagnostic checklist.
DSM-5-TR. Koutsoklenis and Honkasilta (2023), section 4, directly compare the 2013 and 2022 texts and report unchanged ADHD criteria alongside revised discussion of sex and gender. This is a critical perspective article; the trace uses its textual comparison, without adopting the authors’ wider conclusions about ADHD.
DSM-II category count: 182
The DSM-I total of 106 and the DSM-IV total of 297 follow Surís, Holliday and North (2016), which is a different published history from the one behind the two middle figures, so the four are historical tallies from two sources rather than one series. Kawa and Giordano (2012), in the section on DSM-III, explicitly report 182 DSM-II mental-disorder categories and 265 in DSM-III. This is a published historical tally, not a new independent recount. The original 1968 manual’s Section 2 (printed pp. 5–13) distinguishes coded entries, subdivisions, and administrative terms; its ten principal diagnostic groups are a different level of counting. The number should not be interpreted as prevalence or as a standardized measure of diagnostic breadth across editions.
Verification sources. Kawa and Giordano, published historical analysis · American Psychiatric Association, DSM-II (1968), original manual scan.
DSM-5: what the total counts
Blashfield et al. (2014, Table 1, p. 32; pp. 41–42) report 541 DSM-5 categories, including miscellaneous administrative entries, while 151 categories have diagnostic criteria. Their earlier-edition totals are 128, 193, 228, 253, 383, and 383 for DSM-I through DSM-IV-TR. These differ from the historical tallies used elsewhere in this chapter, including Kawa and Giordano’s 182 for DSM-II, so the table presents source-specific tallies rather than a standardized growth series.
The previous DSM-5 figure of 237 has been withdrawn because its counting convention could not be established. More specifically, 541 and 151 count different units; neither rescues an inference that the diagnostic domain contracted to 237. The authors attribute much of the increase in their total to administrative categories, whereas changes in eligibility require examining individual diagnostic rules. Blashfield et al., The Cycle of Classification.
From the source to the chart.
Every estimate passes through the same sequence. A chart cannot resolve an unclear source or definition.
The questions guiding selection.
For repeated evidence: national U.S. prevalence, symptoms, distress, suicidal behavior, care use, and mortality produced by federal statistical systems or preserved from an auditable national probability study. One commissioned survey of self-rated stress stands beside them, because its sponsor made a national crisis claim from it and its core question never changed. For historical evidence: records that changed how a U.S. population was counted or generalized from 1840 through the first national diagnostic surveys.
The original source.
An official table, archived study record, or primary paper supplies the finding. Peer-reviewed historical analysis explains how a study was conducted or received without replacing an available original. Public source links accompany the record.
The basis for inclusion.
A finding needs an identifiable producer, population and measure, together with enough documentation to explain who was sampled, whom the result was intended to represent, and how far it can actually be generalized or compared.
The published finding.
The archive uses the published total, rate, percentage, sample description, or documented judgment about validity. It does not average subgroups, fill missing years, estimate values from unlabeled plots, recalculate survey weights, or substitute a secondary headline.
The meaning of the number.
Each finding identifies its study or statistical system, what was measured, the case rule, population, period, collection method, unit, comparable segment, source, and scope of inference. Historical records also explain their limits and why they were selected.
Comparability over time.
An observation connects to an earlier one only when their definitions and methods support comparison. Survey redesigns, changed recall periods, and incompatible collection methods create separate segments or snapshots.
Checks before publication.
Checks look for duplicate estimates, missing definitions, implausible units or values, incorrect comparison groups, broken source links, and differences between the source inventory and the chart totals.
Each source must meet its admission criteria.
Repeated series and historical records have different requirements. Neither admits a finding merely because it is influential or numerically dramatic.
- Is the estimate national? State estimates and local studies answer different questions.Required
- Is the source authoritative and public? The finding must come from a federal statistical publication, vital record, or archived national probability study with a traceable published result. The one exception is a commissioned survey that has asked one question in one form every year, deposits its data in a public archive, and is shown with its sampling stated, and Stress in America is the only series admitted under it.Required
- Is the value explicitly published? A labeled table, data row, or official chart value must exist.Required
- Can the construct be stated precisely? The label alone is insufficient; threshold, population, and recall window must be recoverable.Required
- Can comparability be judged? Survey mode, redesigns, and fieldwork changes must be documented well enough to show where comparisons are defensible.Required
- Are there two like-for-like observations? If not, producer-generated statistics may still enter as a study snapshot.Series only
Systematic exclusions
- Demographic subgroups by default. Each measure covers its specified broad national population.
- State, county, campus, health-system, and convenience samples in the repeated library. A historically consequential local probability study may enter the historic mode only with its external-validity boundary displayed.
- Proprietary polls in the repeated library, unless a poll meets the exception for commissioned surveys above; a bounded report may still appear as a clearly labeled snapshot.
- Pre/post-redesign splices that an agency warns should not be compared.
- Values that require unverified reconstruction, visual digitization, subgroup aggregation, or microdata weighting not independently reproduced here.
These exclusions protect interpretability, and they are not claims that the omitted populations or datasets matter less.
Six questions for a historical finding
Every value carries its definition with it.
The source record preserves enough context to reconstruct why an observation was admitted and how it should be read.
The minimum evidence record joins the number to the decisions that produced it, so a percentage that lacks that record is not admitted to the dashboard.
Transcription rules
- The estimate is the published national total for the stated population, rather than the most dramatic subgroup.
- Published precision is retained. A whole-number YRBS table remains whole-number data; it is not padded with invented decimals.
- Confidence intervals appear only when the official release provides them for the selected estimate.
- Pauses in data collection remain gaps. They are not shown as zeroes or filled with estimated values.
- Grouped survey cycles stay intact. “2017–March 2020” is not silently converted into a single-year 2019 estimate.
- The source note identifies the table or page whenever the official publication makes it identifiable.
- Links lead to the agency’s stable report page wherever available, rather than a secondary summary.
- Age-adjusted rates, modeled disorder estimates, and survey weights are not recalculated here; the chart shows the official published result.
Historical transcription keeps unlike evidence unlike
A historic finding may be a count, rate, percentage, sample size, design omission, or validity judgment. Each retains its published form. Counts and sample sizes receive no percentage bar; qualitative findings such as “invalid” or “omitted” receive no invented numeric scale. Obsolete source language appears only where naming the original construct is necessary and is identified as historical terminology.
When observations can be connected.
Names can remain stable while the thing measured changes. Comparability depends on the definition, population and collection method behind the label.
Connecting two points asserts that comparison is defensible. The archive checks six dimensions before connecting observations.
Continuity rule: a material change in any dimension triggers review. The result may be a new segment, a different series, or a snapshot.
Pre-2020 estimates and the 2021+ multimode design are retained but never connected. The suicidal-behavior measures use only the 2021+ segment, while the GAD-7 series begins in 2024, when the instrument entered the survey.
The modified GAD-2 and PHQ-2 reference window changes from seven days to two weeks. From January 9, 2024 the survey collects in continuous four-week cycles, and the three series fall by 9 to 14 points across that change, so the chart marks both breaks rather than treating the survey as stable.
The combined 2017–March 2020 cycle is separated from the Aug 2021–Aug 2023 estimate after fieldwork resumed under changed procedures.
The library begins with the 2019 design. GAD-7 and PHQ-8 appear in 2019 and 2022; two observations support comparison, not a stable trend. Children’s medication for mental health has three published estimates, and in 2021 most child interviews were conducted at least partly by telephone.
Current parent-reported diagnoses are not joined to pre-2016 NSCH results or equated with direct youth interviews.
The consistent national YRBS items form the 2013–2025 series. The survey moved to tablets in 2023, and CDC publishes its trends across that change. The separately fielded online 2021 ABES figure remains a comparison example.
The age-adjusted rate supports population-adjusted comparison. The recorded count shows scale of loss. They remain separate series and units.
Question Q605 kept its wording and its 1–10 scale from 2007 to 2025, so the nineteen published means form one line. Each sample, however, comes from people who joined The Harris Poll’s online panel, and the weights changed during the run: earlier surveys adjusted for the propensity to be online, the 2023 survey took its internet-use weight from a Pew Research Center benchmark, and the 2024 survey added political party. Because the mean held at 5.0 across those changes, the line shows a level that stayed between 4.8 and 5.1 from 2012 to 2025, yet it is not evidence that one year’s 4.9 differs from another’s 5.0.
ECA standardized five sites into a national projection; NCS and NCS-R used different national probability frames, age ranges, diagnostic systems, and instruments. Their 28.1%, 29.5%, and 26.2% estimates remain adjacent but unconnected.
The NSDUH comparison break
The gap marks a change in survey design. Estimates on either side cannot be read as a year-to-year change in the underlying population.
Definitions behind the five comparisons
The chapter’s five comparisons illustrate changes in measurement rules. They are not five estimates of the same outcome, and the additional comparison examples do not enter the repeated-series inventory.
The first two comparisons use the 2024 National Survey on Drug Use and Health (NSDUH), covering the U.S. civilian, noninstitutionalized population aged 18 or older. Those four estimates are retained in the evidence library.
Serious psychological distress uses a Kessler-6 score of at least 13. The month and year estimates are published in SAMHSA’s detailed tables 6.89B and 6.88B. This symptom screen does not establish a psychiatric diagnosis.
Major depressive episode is defined from reported symptoms. Severe impairment adds a Sheehan Disability Scale score of at least 7 in one or more adult role domains. See the 2024 annual report, figures 49–50 and table A.37B.
The cutoff and instrument examples use 2019 NHIS household interviews of the U.S. civilian, noninstitutionalized adult population. GAD-7 ≥10 is approximately 6.1%, calculated by adding CDC’s rounded moderate and severe categories. The GAD-2 estimate uses the first two GAD-7 items; valid-response denominators can differ because the scales require different items to be answered.
The final comparison is a meta-analysis, not a national survey or a time series. Levis and colleagues report pooled estimates of 24.6% (95% CI: 20.8–28.9) for PHQ-9 ≥10 and 12.1% (9.6–15.2) for SCID major depression. Differences varied across studies; these results do not supply a universal correction factor for screening estimates.
When a category includes more conditions
In the same 2019 NHIS short-screen estimates, 8.1% of adults screened positive for anxiety symptoms, 6.5% for depression symptoms, and 10.8% for either. “Anxiety or depression” includes people meeting either screen’s cutoff, including those meeting both.
The combined estimate is not the sum of the two separate percentages: that would count people who meet both definitions twice. Read the CDC estimates and definitions.
Different evidence keeps different units.
The categories group related measures for reading; their values are never pooled.
Modeled disorder prevalence, reported current diagnoses, and validated symptom-screen thresholds.
Navigation family · never pooledNonspecific distress, self-rated stress, reported symptoms, suicidal thoughts, plans, attempts, and treated injury.
Navigation family · never pooledMental-health service use, including children’s medication for mental health, and recorded suicide mortality, shown as rates or counts.
Navigation family · never pooledSurvey prevalence and service-use estimates retain their published percentage unit. A percent is never placed on the same scale as a death rate.
The NVSS age-adjusted suicide rate keeps its official denominator and is labeled on the axis, in the source ledger, and in the download.
Annual suicide counts are displayed separately because population growth and age structure make them unsuitable substitutes for rates.
Published 95% confidence intervals are retained. Where the extracted official table does not supply one, the ledger shows N/A.
The vertical scale is padded around the selected series and fully labeled, although it is not a shared comparison scale across different constructs.
The Stress in America mean is drawn on its full 1–10 axis rather than on one padded around its values, because on a bounded rating a padded axis would turn a tenth of a point into a slope.
A single report shows its study design and related estimates from the same population and fieldwork period, without a time axis.
Before the finding, the mode names the population frame and whether the result is representative, administrative, projected, restricted, local, population-specific, or invalid.
Multiple findings may share one register when their methodological difference is the lesson. They receive no connecting line, shared scale, or implied trend.
Points within a segment may connect. A survey redesign or incompatible field frame creates a visible gap and a prose explanation.
The chart’s summary reports the latest value, period covered, and number of observations. The definition, comparison limits, and sources sit beside it.
Checking the evidence records.
The source inventory and the evidence library must agree on the number of series and observations.
- The number of estimates shown for each measure matches its stated total.
- Each measure has only one estimate for each period, so none is counted twice.
- Every estimate identifies its period, value, comparable segment, and official source.
- Each of the 38 series belongs to one reading category and one of the nine systems.
- Percentage, rate, count, and rating units remain explicit in labels, axes, ledgers, and selected-series downloads.
- Series and observation totals agree with the source register: 15 + 19 + 4 series and 136 + 216 + 13 + 4 + 38 + 32 + 48 + 3 + 19 observations.
- All eight historic snapshot records have unique identifiers, chronological periods, inference categories, population frames, boundary statements, archive rationales, displayed findings, and public source links.
- The study snapshot contains producer-generated numerical estimates, identifies the data producer, states the field frame, and links to the producing study’s table.
- Study snapshots require statistics from the organization that produced them.
- Study and historic snapshot records remain outside the 38-series and 509-estimate totals; the Historic system is counted inside them and carries a visible break before 1996.
What validation cannot prove
Data reconciliation can catch duplicate periods, absent values, unit drift, descriptions without a matching record, and incorrect totals, but it cannot prove that an agency’s construct is the best measure of mental health, that survey nonresponse is harmless, or that changes in a published series reflect only changes in the population. Those are substantive limitations, not data-cleaning errors.
Nine systems and the records that preceded them.
The inventory records where the 509 repeated observations, one producer-generated study snapshot, and eight historic snapshots come from, while keeping the Historic series system distinct from the historic snapshot register.
| System | Series | Estimates | Coverage represented | Principal source | Governing caution |
|---|---|---|---|---|---|
| NSDUH | 16 | 136 | 2004–2025, measure-dependent; 2020 omitted | SAMHSA detailed tables | Do not bridge the 2021 multimode redesign to 2019 or earlier. |
| Household Pulse | 3 | 216 | Apr 2020–Sep 2024 published field periods | NCHS national data | Reference period changes from seven days to two weeks in July 2021, and collection moves to four-week cycles in January 2024. |
| NHIS | 4 | 13 | 2019–2024, measure-dependent | NCHS symptom report · care series · children’s medication: NCHS Data Brief 381, NCHS Data Brief 472, MMWR QuickStats, October 27, 2023 | Use the redesigned 2019+ frame; rotating measures yield sparse series, and the children’s medication item counts use rather than appropriateness. |
| NHANES | 1 | 4 | 2013–Aug 2023 grouped cycles | NCHS Data Brief 527 | Break before post-pandemic fieldwork and administration changes. |
| YRBS | 6 | 38 | Biennial 2013–2025 | CDC trends report, p. 54 · Youth Health in Focus technical notes, 2025, Table 46 | The poor-mental-health item begins in 2021; do not splice ABES. Tablets replace paper in 2023, and the 2025 overall response rate is 30.6 percent. |
| NSCH | 4 | 32 | Annual 2016–2023 redesigned survey | HRSA/MCHB brief via NCBI | Parent-reported current diagnosis; no comparison to pre-2016 design. |
| NVSS | 2 | 48 | Annual 2001–2024 | CDC suicide data | Recorded count and age-adjusted rate stay separate. |
| Historic | 1 | 3 | 1957–1996; break before the 1996 qualified extension | ICPSR harmonized pair · NORC GSS data table | Connect 1957–1976 only; the vernacular item is not diagnostic prevalence. |
| Stress in America | 1 | 19 | Annual 2007–2025, fielded between June and September | APA reports for 2007–2012, 2013, 2014–2015, 2016–2017 and 2021 and 2018–2020 · Harris Poll topline tables for 2022, 2023, 2024 and 2025 | Opt-in online panel with changing weights; a self-rating on a 1–10 scale, not a symptom screen or a diagnosis. |
| Total | 38 | 509 | Totals exclude the study snapshot, eight historic snapshots, and comparison examples. | ||
Added-series table references. NSDUH suicidal thoughts, plans, and attempts use Tables 9.62B, 9.63B, 9.64B, 9.78B, 9.79B, and 9.80B. K6 serious psychological distress uses Tables 6.88B and 6.89B. GAD-7 anxiety symptoms use the 2024 annual report, Figures 45 and 46 and Tables A.34B and A.35B, and the 2025 annual report, Figures 49 and 51 and Tables A.32B and A.34B.
Data producers. NIMH topic pages and the Mental Health America report archive are excluded because they curate or repackage statistics generated elsewhere. The NCS-R values below use the producing study team’s table; the Stress in America series credits The Harris Poll, which recruits, fields and weights each survey, beside the American Psychological Association, which commissions it.
Study snapshot source register
| Study | Displayed statistics | Data producer | Field frame | Source record | Governing caution |
|---|---|---|---|---|---|
| NCS-R | 15 past-year disorder prevalence estimates | Harvard Medical School research team + University of Michigan Survey Research Center | Adults 18+ · contiguous U.S. households · 2001–2003 face-to-face CAPI | NCS-R prevalence table | Parallel disorder estimates inside one study; not a time trend. |
Historic snapshot register
The register is curated for methodological value to this archive. It begins with the earliest nationwide count located, distinguishes actual national probability from projection or administrative coverage, and closes where structured diagnostic surveys meet the chapter’s modern evidence period.
Americans View Their Mental Health is a repeated measure under the Historic system: 1957–1976 is the connected replication pair, while 1996 appears after a visible break in comparability as a qualified extension.
| Period | Record | Displayed finding | Inference class | Principal source | Governing caution |
|---|---|---|---|---|---|
| 1840 / 1844 | Sixth Census and Jarvis audit | Nationwide disability schedule; racial comparison invalid | National enumeration | Census schedule · Jarvis critique | Coverage cannot repair classification error. |
| 1880 / 1890 | Decennial design break | Physician schedules supplied 17.2% of the 1880 reported count; omitted in 1890 | National enumeration | 1890 Census analysis | Do not connect counts across the collection change. |
| 1940–1945 | Selective Service screening | Adjusted 1.846 million neuropsychiatric rejections; 38.2% of IV-F findings | Restricted administrative cohort | Army medical history | Men examined for service do not represent U.S. adults. |
| 1952 / 1962 | Midtown Manhattan Study | 23.4% marked, severe, or incapacitated; n=1,660 | Local probability study | Historical analysis | The estimate generalizes to the study community, not the nation. |
| 1955 | Patients in mental hospitals | 631,503 patients; 389.1 per 100,000 | Public and private mental hospitals, including veterans’ hospitals; excludes Alaska and Hawaii | Statistical Abstract, table 94 | This broader total includes the 558,922 state-and-county residents reported in the Timeline. Occupancy measures institutions, law, capacity, and duration, not untreated prevalence. |
| 1979–1992 | National Survey of Black Americans | 2,107 participants in the opening wave; four-wave panel | Population-specific national probability | ICPSR record | Supports inference to Black adults within its frame, not the total U.S. population. |
| 1980–2003 | ECA → NCS → NCS-R | 28.1%, 29.5%, and 26.2% 12-month any-disorder estimates | Projection plus national surveys | NCS-R comparison | Different samples, ages, diagnoses, and instruments prohibit a firm trend. |
| 2001–2004 | NCS-A | 49.5% lifetime any assessed disorder; 22.2% severe impairment or distress | National youth probability survey | NCS-A paper | Any assessed disorder and severe need are not interchangeable headlines. |
How new evidence is added.
Each new release is checked against the previous measure and survey design before it is added.
- Publication and date.
The publication must replace or extend the current source, with its release date recorded.
- Definition and method.
The instrument, case threshold, population, recall period, collection method, and sampling design are compared with the preceding observation.
- Published national estimate.
The estimate retains its published value, precision, period, confidence interval, and source note; the archive does not derive a replacement value.
- Comparability over time.
The comparison determines whether the estimate joins an existing segment, begins a new segment or measure, or remains a separate snapshot.
- Explanations alongside the number.
The definition, comparison limits, sources, coverage, and account of what is latest are reviewed with the estimate.
- Historical corrections.
When a newly available table, archival correction, or stronger primary record changes a historical finding, its scope, limits, reason for inclusion, and sources are revised together.
- Consistent totals.
Totals are recalculated for each measure, statistical system, reading category, and the library as a whole.
- Final checks.
Before publication, the records are checked for duplicates, missing information, inconsistent units or comparison groups, and errors in links or totals.
Limitations that remain
Broad totals can move differently from age, race, gender, income, disability, geography, or other subgroup estimates. The present library does not expose those differences.
The library inherits agency decisions about questions, thresholds, respondent frames, survey modes, exclusions, and publication schedules.
GAD, PHQ, K6, and Household Pulse screeners identify symptom thresholds; modeled disorder prevalence and reported diagnoses are different constructs, and a 1–10 stress rating sets no threshold at all.
Stress in America inherits a professional association’s choice of questions and a polling firm’s panel and weights, and The Harris Poll states its precision as a Bayesian credible interval rather than the margin of error a probability sample reports.
Self-report, parent report, school-based surveys, household interviews, examination surveys, and death certificates do not produce interchangeable evidence.
The latest year differs by system. “Latest” always means latest within the selected series, not a common national reference date.
Household Pulse contributes many field periods, while other systems publish annually or biennially. Observation count measures cadence, not evidentiary importance.
The eight snapshot records were selected for their value to the archive’s argument about counting and generalization, and they are not a census of every historical survey, institution table, or epidemiological study.
Historic census labels, military classifications, impairment ratings, vernacular self-reports, and DSM interviews are presented in context. None is translated into another era’s diagnostic vocabulary.