Beyond the daily-reading headline
The published reading-for-fun distributions in the National Assessment of Educational Progress tell a more detailed story than a decline in the almost-every-day category alone. Between 2012 and 2025, the estimate for “Once or twice a week” moved from 22.54% to 24.30% at age nine. At age thirteen, the same category moved from 25.79% to 20.87%. Those are changes of +1.76 and −4.92 percentage points in the respective published estimates. They describe different endpoint patterns, not a tested difference between the two age groups.
This analysis retains all five response categories rather than selecting only the daily and least-frequent answers. Its object is deliberately narrow: the dated percentage estimates published by NCES and the arithmetic differences between their 2012 and 2025 endpoints, separately for each age. The item-specific denominator treatment and missing-response rates have not been established here. Consequently, these numbers are not relabelled as prevalence among all enrolled students, or as verified proportions among a precisely defined set of item respondents.
That restriction matters from the beginning, not just in a technical footnote. The comparison is informative about the composition of the official published vectors. It cannot establish that the observed directions would survive a different treatment of missing responses or changes in who answered. Nor can it identify children who stopped reading, people who moved from one frequency to another, or the causes of any change.
What the questionnaire actually measures
The question concerns reading for fun on one's own time. In both the 2012 and 2025 questionnaires inspected for ages nine and thirteen, it appears as a row in a matrix asking how often the student does each listed activity. The reading row and the five ordered answer labels match across these instruments. This direct comparison is stronger evidence of wording continuity than merely finding similarly named columns in a data download, although matching wording does not establish identical surrounding context or response behaviour.
The categories are “Almost every day,” “Once or twice a week,” “Once or twice a month,” “A few times a year,” and “Never or hardly ever.” The last label must remain intact. It does not mean that a person never reads anything. A student can read for school or other purposes while selecting that answer about pleasure reading. Similarly, the first category is almost every day, not an exact count of seven reading days per week.
“Once or twice a week” is also narrower than the everyday phrase weekly readers. The latter could include people who read almost every day. Here it is one exclusive published response category, not a combined at-least-weekly measure. The item does not identify books as opposed to every other reading format, hours spent reading, the number of books completed, literary genres, reading proficiency, or enjoyment of a specific text. None of those outcomes can be substituted for the actual wording.
The complete age-nine profile
At age nine, the published estimate for “Almost every day” was 52.55% in 2012 and 36.96% in 2025, a difference of −15.59 percentage points. The estimate for “Once or twice a week” was 22.54% and 24.30%, a difference of +1.76 points. “Once or twice a month” moved from 7.22% to 11.54%, or +4.31 points. Values here are rounded for reading; the downloadable input preserves the additional precision supplied by the official data service.
The remaining categories also have positive endpoint differences. “A few times a year” moved from 6.68% to 10.38%, a change of +3.71 percentage points. “Never or hardly ever” moved from 11.01% to 16.82%, or +5.81 points. Thus the age-nine vector has one negative category difference and four positive ones. This is a description of the published numbers, without a significance claim about any individual difference.
Keeping all five answers visible changes the interpretation of the familiar daily-reading comparison. A description confined to the first and last categories would conceal the positive differences in the three intermediate categories. Those intermediate answers are not interchangeable: once or twice a week describes a substantially different reported frequency from a few times a year. The full table preserves that distinction instead of treating every answer below almost daily as the same condition.
The complete age-thirteen profile
At age thirteen, the published “Almost every day” estimate was 27.07% in 2012 and 14.23% in 2025, a difference of −12.84 percentage points. “Once or twice a week” moved from 25.79% to 20.87%, or −4.92 points. Unlike the age-nine vector, two categories therefore have negative endpoint differences. This observation does not require merging them into a new statistic or treating the age groups as a single combined sample.
The other three estimates have positive differences. “Once or twice a month” moved from 13.97% to 17.86%, a change of +3.89 percentage points. “A few times a year” moved from 11.31% to 18.11%, or +6.81 points. “Never or hardly ever” moved from 21.86% to 28.93%, or +7.07 points. The accompanying table reports every category in the original order, including the smaller intermediate-frequency changes.
This profile cannot be adequately described as a simple replacement of daily reading by never reading. Two initial categories have negative differences, three have positive differences, and the least-frequent label includes hardly ever. More fundamentally, replacement is a potentially misleading description of separate survey snapshots. The arithmetic describes different distributions at two dates; it does not document the paths followed by individual students between those dates.
What the two profiles can be compared for
The most useful contrast is structural and descriptive. In the age-nine published vector, the once-or-twice-a-week estimate has a positive endpoint difference. In the age-thirteen vector, it has a negative one. This is a transparent comparison of the signs of two calculated quantities. It is not evidence from a statistical test that an age effect changed, that adolescents responded differently to a particular influence, or that a younger cohort became the older cohort in these data.
Neither age category is a substitute for a school grade. NAEP long-term trend samples are defined by age, and students of the same age can attend different grades. The service response includes grade-related metadata, but the analysis uses the documented age cohorts and does not convert the results into fourth-grade or eighth-grade findings. Such a conversion would change the population description without changing the observations, creating an apparent precision the source does not support.
The periods also do not describe identical calendar windows for the two ages. The 2025 wave includes age-thirteen assessment in fall 2024 and age-nine assessment in winter 2025. The shared wave label is useful for identifying the release, but it should not obscure that timing. No annual rate of change is calculated, and no intervening trajectory is inferred from the two endpoints. A thirteen-year endpoint difference is not evidence that the change proceeded steadily each year.
The unresolved denominator is a substantive limitation
The official service returns five percentage estimates for each selected age and year. Their unrounded values sum to approximately one hundred, allowing normal floating-point representation differences. That accounting check verifies the completeness of the retrieved vector. It does not reveal exactly which records were retained, excluded, reweighted, or otherwise handled because of a missing answer to this particular questionnaire row.
General NAEP questionnaire documentation discusses missing information, but the item-specific treatment and missing rates for these LTT endpoints were not established in this analysis. It would therefore be inappropriate to state that the denominator is every sampled student, every assessed student, all school-enrolled children of the relevant age, or a fully verified set of usable-item respondents. The analytic denominator remains the one embodied in the published estimates, with its item-specific handling unresolved here.
This uncertainty limits interpretation rather than invalidating subtraction. The difference between two published numbers is still reproducible. What is not established is whether a corresponding population change would have the same size or even direction under another defensible treatment of nonresponse. No sensitivity analysis to differential item nonresponse is presented. The estimates should not be described as robust to that uncertainty, and the absence of a low-response warning in a returned cell is not proof that no answers were missing.
Sampling, participation and inclusion
The broader NAEP long-term trend target population consists of students of the relevant ages enrolled in public and private schools nationwide. The assessment uses a multistage sample and weights rather than a census or a simple unweighted tally. That design context explains the origin of the published estimates, but it does not resolve the narrower item-denominator question. Children outside the school-enrolled target population must not be silently added to the interpretation.
For the 2025 reading assessment, the official methods report approximately 7,400 age-nine participants and 8,200 age-thirteen participants. Student participation was 91% and 89%, respectively. Original school participation before substitution was 89% and 85%. These are assessment and sampling participation figures. They are not response rates for the reading-for-fun row, and dividing a published category percentage by them would not produce an appropriate item-nonresponse adjustment.
Private-school original participation was 37% at age nine and 35% at age thirteen, below the standard for separate reporting of private-school results. Private-school students remain part of the national target population; the limitation must not be converted into a claim that national estimates cover public schools alone. Participation, exclusions, weighting and the composition of assessed students can affect interpretation across waves. This analysis does not demonstrate that those processes were unchanged between 2012 and 2025 or that they explain the observed category differences.
Instrument continuity is not perfect procedural continuity
The inspected 2012 questionnaires place the reading matrix at question fourteen for age nine and question sixteen for age thirteen. In 2025 it appears at questions ten and twelve. The shared matrix stem, focal reading row and ordered answer labels support comparing the same named response categories. Different question positions and surrounding questionnaire content nevertheless mean that an identical focal sentence should not be advertised as an entirely identical measurement environment.
The 2025 methods also describe a recovery procedure for assessment answers that students marked in ways the scanning device recorded as missing. Human coders reviewed qualifying booklets to recover intended responses. The documentation establishes a procedural issue in the assessment, but this analysis has not established a separate effect or treatment for the reading-for-fun survey item. It would be equally unwarranted to attribute the category changes to that procedure or to assert that the item could not have been affected.
The earlier 1984 data belong to the original assessment format without accommodations, whereas the primary 2012 and 2025 observations use the revised format permitting accommodations. A bridge study accompanied changes in 2004. For that reason, 1984 is historical context, not an additional primary endpoint here. The current comparison neither pools the original and revised formats into one unqualified series nor treats the bridge as proof that every survey-response process remained invariant.
Why the calculation uses percentage points
Each primary result subtracts the 2012 published estimate from the 2025 published estimate within the same age and response category. A change from 22.54% to 24.30% is approximately +1.76 percentage points, not a 1.76% relative increase. Keeping the unit explicit prevents a common ambiguity and avoids introducing a second measure that can appear larger simply because its starting value is smaller.
Calculations use the additional decimal precision returned by NCES, not the two-decimal values printed in the article. The exact source numbers and differences are provided in compact files. Rounding occurs only for display. This distinction explains why a displayed subtraction can occasionally differ by a hundredth from the displayed change: subtracting two separately rounded values is not always the same as rounding their unrounded difference.
No category is dropped, and no published vector is rescaled to force its displayed values to total one hundred. The calculation checks that each full-precision vector does so within a very small numerical tolerance, while retaining ordinary display-rounding residuals. Changes across the five categories likewise balance arithmetically apart from numerical representation. This is a property of the published composition, not a transition matrix or an estimate of how many children changed behaviour.
Why there are no confidence bands or significance claims
The primary tables and figures are descriptive. They do not show confidence intervals, p-values, significance stars, or a declaration that one change is reliably larger than another. Although the official service supplies comparison metadata, the exact test family associated with the retrieved comparison batch was not independently established. This article does not turn those returned labels into newly validated statistical findings.
A complete inferential analysis would also have to respect the survey design and the dependence between category estimates. Five categories that form one distribution cannot be treated as five unrelated samples. Adding marginal variances as if categories were independent would not automatically produce a valid uncertainty estimate for a combined category, a contrast between category changes, or an interaction between age and time. No such synthetic interval is calculated here.
The absence of significance claims does not mean that every numerical difference is mere error, nor does it mean that the underlying distributions are equal. It means that this article restricts its claims to the published point estimates and their reproducible arithmetic. Readers needing an inferential answer should use the official tools and appropriate design documentation, with the comparison family and denominator assumptions made explicit rather than inferred from an attractive chart.
The pandemic and other explanations remain outside the design
The interval between 2012 and the 2025 wave includes the COVID-19 pandemic, as well as many other changes in schooling, technology, family life and reading opportunities. Including an event inside an observation window does not identify its contribution to an endpoint difference. There is no untreated comparison group, individual exposure measure, or causal model in this analysis, and the two selected dates cannot separate pandemic-related changes from earlier or later developments.
The same restraint applies to screens and reading scores. The questionnaire response is not a measure of device use, and an aggregate association with assessment performance would not demonstrate that one behaviour caused the other. This article therefore does not calculate a screen effect, an achievement penalty, or an educational return to moving between categories. It also does not use the reading scores to rank the worth of students' reported leisure choices.
The results can help sharpen questions for further investigation: why do intermediate-frequency categories deserve separate attention, and how do survey-response processes affect the reported profile? They cannot answer those questions by themselves. Any follow-up would need additional documented variables, an appropriate research design and explicit treatment of nonresponse. A descriptive table should remain a starting point for such work, not a substitute for it.
A modest contribution from keeping the middle visible
NCES already publishes the categories, and earlier reporting has extensively discussed declining almost-daily reading and higher estimates for rare or absent pleasure reading. This analysis does not claim to discover a new dataset, a previously unknown question, or the first evidence of changing reading habits. Its contribution is a reproducible endpoint comparison that keeps the intermediate answers visible and confines interpretation to what the available evidence supports.
The substantive takeaway is therefore specific. The age-nine published profile has positive differences in every category below almost every day, including once or twice a week. The age-thirteen profile has negative differences in both almost every day and once or twice a week, with positive differences in the remaining three categories. That distinction is lost when all non-daily answers are compressed into one label or when the final category is misleadingly renamed nonreaders.
Readers can inspect both complete five-category tables and rerun the subtraction from the selected official inputs. The same files preserve source precision and identify the assessment wave rather than substituting the release or retrieval date. Reproducibility establishes how the reported arithmetic was obtained; it does not resolve the missing denominator metadata or prove a population explanation. Keeping both the finding and that boundary visible is the central purpose of this secondary analysis.
Age 9: all five published category changes
2012–2025 endpoint subtraction, percentage points; not an intervening time series.
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Category | 2012 (%) | 2025 (%) | Change (pp) |
|---|---|---|---|
| Almost every day | 52.55 | 36.96 | −15.59 |
| Once or twice a week | 22.54 | 24.30 | +1.76 |
| Once or twice a month | 7.22 | 11.54 | +4.31 |
| A few times a year | 6.68 | 10.38 | +3.71 |
| Never or hardly ever | 11.01 | 16.82 | +5.81 |
NCES-published estimates, 2012 and 2025; changes in percentage points. Item denominator/missing rates unresolved. No significance claims; rounded display, full precision in CSV. Source flags 0 for all primary cells, not proof of zero missingness.
Age 13: all five published category changes
2012–2025 endpoint subtraction, percentage points; not an intervening time series.
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Category | 2012 (%) | 2025 (%) | Change (pp) |
|---|---|---|---|
| Almost every day | 27.07 | 14.23 | −12.84 |
| Once or twice a week | 25.79 | 20.87 | −4.92 |
| Once or twice a month | 13.97 | 17.86 | +3.89 |
| A few times a year | 11.31 | 18.11 | +6.81 |
| Never or hardly ever | 21.86 | 28.93 | +7.07 |
NCES-published estimates, 2012 and 2025; changes in percentage points. Item denominator/missing rates unresolved. No significance claims; rounded display, full precision in CSV. Source flags 0 for all primary cells, not proof of zero missingness.
Data & downloads
Scroll horizontally to see all columns.
| Category | 2012 (%) | 2025 (%) | Change (pp) |
|---|---|---|---|
| Almost every day | 52.55 | 36.96 | −15.59 |
| Once or twice a week | 22.54 | 24.30 | +1.76 |
| Once or twice a month | 7.22 | 11.54 | +4.31 |
| A few times a year | 6.68 | 10.38 | +3.71 |
| Never or hardly ever | 11.01 | 16.82 | +5.81 |
NCES-published estimates, 2012 and 2025; changes in percentage points. Item denominator/missing rates unresolved. No significance claims; rounded display, full precision in CSV. Source flags 0 for all primary cells, not proof of zero missingness.
Scroll horizontally to see all columns.
| Category | 2012 (%) | 2025 (%) | Change (pp) |
|---|---|---|---|
| Almost every day | 27.07 | 14.23 | −12.84 |
| Once or twice a week | 25.79 | 20.87 | −4.92 |
| Once or twice a month | 13.97 | 17.86 | +3.89 |
| A few times a year | 11.31 | 18.11 | +6.81 |
| Never or hardly ever | 21.86 | 28.93 | +7.07 |
NCES-published estimates, 2012 and 2025; changes in percentage points. Item denominator/missing rates unresolved. No significance claims; rounded display, full precision in CSV. Source flags 0 for all primary cells, not proof of zero missingness.
Evidence & method
How we calculated it
- Selected official S003501 published percentage vectors; ages 9 and 13 separately; 2012R3 and 2025R3.
- Subtract 2012 from 2025 using full returned decimal precision; display two decimal places. No renormalization.
- Exact 2012 and 2025 questionnaire stems, focal row and five ordered labels inspected. Context need not be identical.
- Independent arithmetic checked source cells and differences; no validated inference or nonresponse robustness is claimed.
- Underlying assessment uses national school-enrolled public/private samples and weights; exact item-specific denominator handling remains unverified.
What the data cannot tell us
- Item-specific denominator and missing-response rates unresolved; not all-enrolled-population prevalence or verified respondent-conditional proportions.
- No confidence intervals, significance claims, cross-age interaction tests, individual transitions or causal attribution.
- Assessment participation is not item response; 2025 private-school participation and response-recovery procedures require caution.
- The two 2025 age samples have different fieldwork seasons; endpoints do not establish annual trajectories.
- 1984 original/no-accommodation format is contextual only; 2004 bridge does not establish invariant item-response processes.
Original sources
- NCES 2025 LTT reading student experiences
NCES 2025 LTT reading student experiences: Current official report and five-category labels; school-year 2024–25 wave. Exact item denominator/rates remain unresolved in this analysis.
Accessed October 30, 2026
- NCES NAEP public data service
NCES NAEP public data service: Selected data queries: type=data; subject=RED; subscale=RRPSCT; jurisdiction=NT; statType=RP:RP; variable=S003501; year=2012R3,2025R3; cohort=1 or 2; Program=LTT; UseCache=true; ExtendedEF=true. Raw retrieval included contextual 1984R1, excluded from public primary inputs.
Accessed October 30, 2026
- NCES NAEP API documentation
NCES NAEP API documentation: Documents cohort codes, row percentages and error flags; warns results can change. Dated snapshot, not a permanently current series.
Accessed October 30, 2026
- NCES 2025 LTT methods
NCES 2025 LTT methods: National school-enrolled sampling, weighting, participation, exclusions, bridge and response recovery. Assessment participation is not item response.
Accessed October 30, 2026
- NCES LTT interpretation
NCES LTT interpretation: Age-specific fieldwork timing and caution about original/revised formats. No causal or longitudinal interpretation.
Accessed October 30, 2026
- NCES 2012 age 9 reading questionnaire
NCES 2012 age 9 reading questionnaire: Exact matrix and reading row inspected; question 14, printed page 3; five ordered labels match 2025.
Accessed October 30, 2026
- NCES 2025 age 9 reading questionnaire
NCES 2025 age 9 reading questionnaire: Exact matrix and reading row inspected; question 10, printed page 3; row VC375071. Surrounding context not assumed identical.
Accessed October 30, 2026
- NCES 2012 age 13 reading questionnaire
NCES 2012 age 13 reading questionnaire: Exact matrix and reading row inspected; question 16, printed page 4; five ordered labels match 2025.
Accessed October 30, 2026
- NCES 2025 age 13 reading questionnaire
NCES 2025 age 13 reading questionnaire: Exact matrix and reading row inspected; question 12, printed page 4; row VC375071. Surrounding context not assumed identical.
Accessed October 30, 2026
- US Department of Education publications FAQ
US Department of Education publications FAQ: Federal-publication public-domain guidance, with exceptions. Selected statistical facts only; no blanket Creative Commons licence or third-party material.
Accessed October 30, 2026
- US Department of Education copyright notice
US Department of Education copyright notice: Notice explicitly limits website blanket to www.ed.gov; not asserted as blanket licensing for nationsreportcard.gov.
Accessed October 30, 2026
Cite & reuse this study
Alisson Bento. Beyond daily reading: the full NAEP pleasure-reading profiles, 2012–2025. Secondary descriptive analysis of NCES NAEP Long-Term Trend Reading published category estimates. Sources accessed October 30, 2026. https://alissonbento.com/research/naep-pleasure-reading-category-profiles-2012-2025/
Selected factual federal statistical aggregates attributed to NCES; original calculations and presentation by Alisson Bento. No blanket article licence, third-party material or official endorsement. See rights.md.
Corrections or editorial questions