Skip to content

2022 survey update · Participation, not membership

Book-club participation and audiobook listening: a 2022 survey update

Among weighted joint-valid 2022 SPPA adult responses, past-year book-club or reading-group participation was 13.1043% for audiobook listeners and 3.3770% for nonlisteners. This descriptive update reports the full joint distribution, sampling uncertainty and unresolved missingness; it does not measure formal membership or causal effects.

Published
Share
Share this page
EmailCite & reuse this study
Key findings as text
Reusable imagesBook-club participation and audiobook listening: a 2022 survey update — 13.1043%: Club participation among observed audiobook listeners. Weighted joint-valid domain; 597 respondents.Download sharing image (JPG) ↓

Key findings

13.1043%

Club participation among observed audiobook listeners

Weighted joint-valid domain; 597 respondents.

3.3770%

Club participation among observed nonlisteners

Weighted joint-valid domain; 2,494 respondents.

9.7274 pp

Conditional participation difference

Audio yes minus no; sampling SE 2.1480 pp. Not a causal effect.

7.5463%

Nominal joint weight in incomplete target pairs

204 of 3,295 selected for both sections; item MCAR not established.

A later-wave update, not a first discovery

The 2022 Survey of Public Participation in the Arts allows a specific question about reading-related activity: among selected adults with valid answers to both questions, how does reported book-club or reading-group participation differ between those who listened to any audiobook and those who did not? The weighted conditional estimates are 13.1043% and 3.3770%, respectively. This is an association between two activities recalled over the preceding twelve months. It is not a measurement of formal club membership, a preference for one reading format, or the effect of joining a club. Keeping that narrower question visible is essential to interpreting the result.

This analysis is a later-wave update to an established research topic. The National Endowment for the Arts' 2020 report, How Do We Read? Let's Count the Ways, already examined 2017 participation in book clubs and reading groups by reader type. Its Figure 12 and related discussion include audiobook listeners, while Appendix B-4 supplies adjusted modeling. The present binary distinction between any audiobook listening and no listening is not identical to that report's reader-type classifications. Consequently, the new calculations do not establish a 2017–2022 change, reproduce an identical historical contrast, or reveal an association that earlier official research had overlooked.

What the two questions actually measure

The audiobook variable, PEC1Q17A, records whether the selected person listened to any audiobooks during the preceding twelve months. Its scope expressly includes recordings, broadcasts and web streaming. The question does not require finishing an audiobook, naming a title, or reporting a minimum number of listening hours. Unlike adjacent questions about some reading activities, this item's wording does not impose a work-or-school exclusion. A positive answer therefore identifies reported listening within that broad activity definition, not leisure-only listening or sustained engagement with a complete book. A negative answer identifies reported non-listening, not a dislike of audio.

The participation variable, PEMCQ1I, concerns taking part in any book clubs or reading groups during the same recall period. It does not ask whether the person held formal membership at the interview date, paid dues, joined online, met in a library, or attended a minimum number of meetings. A separate frequency item follows, but frequency is not the outcome analyzed here. Neither question establishes that an audiobook was the club's chosen book. The combination describes two activities of the same selected subject, without providing a link between the specific books, occasions or social settings involved.

Why a joint sample matters

Separate percentages of audiobook listeners and club participants cannot determine how often the activities occur in the same people. A format margin and a participation margin leave the overlap unknown. The 2022 public-use files make the overlap observable because some selected subjects received both relevant sections of the survey. Audiobook listening belongs to Core 1, whereas the club question belongs to Module C. The analysis therefore uses the intersection of those assignments rather than treating either section's full sample as the denominator for a joint calculation. Responses from different selected people are never combined to manufacture an individual relationship.

The survey's randomized structure assigned one of two cores and additional modules. The released item patterns recover 6,578 Core 1 subjects and 6,615 Module C subjects, with 3,295 in their intersection. This agrees with the technical appendix's allocation of 1,651 subjects to Core 1 with Modules A and C, and 1,644 to Core 1 with Modules B and C. Agreement establishes the relevant release-level assignment structure. It does not establish that every selected subject answered both target items, and it does not make a missing answer equivalent to a negative answer.

From assignment to the observed-response denominator

The starting public-use frame contains 13,251 adults aged eighteen or older with positive supplement weights. The retained subjects are civilian; the survey concerns the civilian noninstitutionalized population. The construction does not impose an extra completed-interview restriction, because positive-weight records include some incomplete interviews and other response statuses. Requiring a completed-interview flag would alter the supplied analytical frame. Within the 3,295-person joint assignment, 3,091 subjects have a yes or no answer to both target questions. Those 3,091 are the complete-valid-item domain used for the primary association.

The distinction between a survey's target population and this observed-response domain is substantive, not a footnote. The weights address features of the sample design and its adjustment process, but they do not prove that item nonresponse is ignorable for this particular conditional relationship. The primary percentages directly describe the weighted joint-valid responses. Extending them to all eligible adults requires additional assumptions about response behavior. Random assignment to survey sections explains planned omissions; it does not prove that refusal, uncertainty or failure to answer within an assigned section is unrelated to club participation or listening.

The primary conditional comparison

Among the 597 joint-valid subjects who reported audiobook listening, 81 reported participation in a book club or reading group. Among the 2,494 who reported no audiobook listening, 92 reported participation. Applying the survey weights gives conditional participation estimates of 13.1043% for the listening group and 3.3770% for the non-listening group. These weighted estimates are not simply the raw fractions of positive answers. The unweighted counts are reported to make the response support visible, while the weighted percentages reflect the prespecified survey calculation within the observed domain.

The difference, defined in advance as the listening estimate minus the non-listening estimate, is 9.7274 percentage points. Its supplied-replicate standard error is 2.1480 percentage points, and its approximate normal 95% interval is 5.5173 to 13.9374 percentage points. The corresponding conditional intervals for the two proportions are 9.0563% to 17.1524% and 2.4081% to 4.3458%. These intervals quantify sampling variation under the specified calculation. They do not include an allowance for every possible reporting error, unmeasured confounder or nonresponse mechanism, and should not be presented as an unconditional guarantee about the population difference.

Reading the entire two-by-two distribution

Conditional percentages answer a specific question within each listening group. The complete two-by-two distribution answers a different question: how is the entire joint-valid weighted domain divided between the four answer combinations? Reported listening and participation together account for 2.4508%; listening without participation accounts for 16.2511%; participation without listening accounts for 2.7454%; and neither reported activity accounts for 78.5528%. All four shares have the same joint-valid denominator and sum to 100% before presentation rounding. Publishing all four cells prevents the discussion from being reduced to a single favorable contrast.

The two views also explain why the larger conditional participation estimate among listeners does not mean that most club participants must be listeners. The groups being conditioned on have very different sizes. Likewise, the large neither-activity cell does not imply that its subjects do not read: this analysis does not classify printed-book or ebook use. Someone can read extensively without reporting an audiobook, and someone can participate in a reading group without reporting audio use. The joint table cannot support an exhaustive taxonomy of readers, a preference ranking or a claim about mutually exclusive reading formats.

Correct weighting across survey sections

The Census documentation specifies a factor of four when combining one core and one module. The computation multiplies PWSUPWGT and each of the 160 supplied replicate weights by that same factor for the joint domain. This assignment multiplier is not chosen to force an agreement with a published result. It represents the documented cross-section sampling treatment. A common multiplier cancels in the numerator and denominator of each conditional proportion, but keeping it explicit prevents confusion with the factor for a single section or with calculations of population totals, which are not reported here.

The weight fields have four implied decimal places. The sufficient-total files preserve the source numerical units multiplied by the assignment factor, rather than displaying those large stored values as numbers of people. Dividing every such weight by ten thousand would change the scale of totals but not the ratios or their replicate estimates. This scale distinction matters for reproducibility: an analyst should not interpret a raw stored weight sum as an estimated population. The public reproduction works with like-scaled numerators and denominators and produces percentages, percentage-point differences and their sampling measures, not headcounts.

Replicates, covariance and the limits of an interval

For every supplied replicate, the analysis recomputes both the numerator and denominator of each proportion. It then recomputes the listening-minus-non-listening difference within that same replicate. The variance formula is four divided by 160 times the sum of squared differences between replicate and full-sample estimates. This procedure retains the covariance between the two conditional proportions. Adding their variances as though the estimates were independent would produce a different standard error for the contrast. The common assignment multiplier and the numerical four in the variance formula perform different jobs and must not be applied interchangeably.

The supplied weights should not be described as having a universally ordinary balanced-repeated-replication construction. Census documentation describes successive-difference replication in self-representing strata and a modified half-sample procedure in non-self-representing strata. The official four-over-160 evaluation corresponds to the stated Fay one-half mean-squared-error configuration, but that computational equivalence is not a reason to erase the construction details. All relevant replicate denominators were positive. Approximate normal intervals are useful sampling summaries here; they remain conditional on the observed-response analysis and do not transform the association into evidence that listening causes participation or that participation causes listening.

What the missing responses leave unresolved

Of the 3,295 subjects assigned to both sections, 204 lack a valid pair of answers. Together they carry 7.5463% of the nominal joint domain's base weight. The coarse breakdown is 69 with valid listening information but missing participation, sixteen with valid participation but missing listening, and 119 missing both. Their weighted shares are 1.9797%, 1.0947% and 4.4719%, respectively. The remaining 3,091 complete pairs carry 92.4537% of the joint weight. These categories describe the analytical missingness pattern; they do not explain why the answers are missing or establish that missing people resemble those observed.

The dictionary distinguishes don't-know, refusal and no-answer codes, while a blank code also appears elsewhere in the file. None is converted to no. The target items have no blank code within the recovered joint assignment, but that empirical finding does not justify treating blanks outside the assignment as substantive responses. In particular, subjects who were not selected for the relevant section cannot be used to enlarge the negative-response group. Missingness diagnostics are reported alongside the principal estimate so that readers can see the unanswered portion rather than mistaking a complete-case calculation for complete observation of everyone selected.

Bounds for one part of the missingness problem

A limited sensitivity calculation addresses missing club answers among subjects whose listening answer is known. For the 608 observed listeners, eleven lack a valid participation answer. Assigning every missing participation answer first to no and then to yes gives a weighted interval from 12.9681% to 14.0078%. Among the 2,552 observed non-listeners, 58 lack a valid participation answer; the analogous interval is 3.2981% to 5.6343%. These are worst-case identification bounds for those observed-listening domains. Their endpoints follow from alternative assignments of the unanswered club outcomes, not from a normal sampling approximation.

The bounds do not address the 135 joint-assigned subjects whose listening answer is missing, including the 119 missing both items. Those people cannot be allocated to the listening groups from the observed information alone. Consequently, the displayed bounds are not confidence intervals and are not an all-adult bound for the main contrast. Even an apparently separated pair of observed-listening bounds cannot establish a population-wide ordering under unrestricted missingness. Reporting their narrower purpose allows the sensitivity exercise to be useful without implying that it resolves selection into the observed-listening or complete-valid domains.

Self-report and proxy-report sensitivity

The survey allows a proxy to answer when a selected person is unavailable. A same-subject pair of responses therefore need not be a pair personally reported by that subject. The primary joint-valid sample includes 2,367 self-reported records, 722 proxy-reported records and two with blank proxy status. The primary analysis retains all of them in line with the supplied survey response frame. The two unknown-status records are not silently labeled self-reports. This distinction matters because a proxy may have different knowledge of private listening and social activities, but the present data do not independently measure the accuracy of either reporting mode.

Restricting the calculation to the self-report domain gives participation estimates of 13.3873% among listeners and 3.3072% among non-listeners, a difference of 10.0801 percentage points. Its replicate standard error is 2.0787 percentage points. The self-report analysis contains 483 observed listeners and 1,884 observed non-listeners. Its similarity to the primary result is a descriptive sensitivity finding, not proof that proxies introduce no bias. Self-report availability may itself be selective, and removing proxies changes the response domain. Neither analysis is declared a corrected population estimate simply because its result resembles the other.

What can be reproduced and what cannot be inferred

The national aggregate inputs retain the four cell totals for the full sample and each replicate, separately for the primary and self-report domains. They are sufficient to reproduce the reported conditional proportions, full joint distribution and covariance-preserving difference without releasing respondent identifiers. Coarse missingness totals and the observed-listening bound inputs accompany them. Original public-use files remain separate from this compact analytical package. The reproduction accepts explicit input and output locations; it does not depend on an author's private directory. Independent computation from the original files and clean-folder aggregate reproduction check the arithmetic through different paths.

Those checks establish reproducibility, not the truth of every behavioral interpretation. They cannot identify why particular adults listen or join groups, resolve absent print and ebook classifications, or prove item-response assumptions. The analysis plan was fixed locally before these association calculations; it was not an externally registered protocol with an independently immutable history. The contribution is a transparent 2022 update with explicit joint denominators, uncertainty treatment and missingness limits. It leaves the broader formal-membership and all-format question unanswered rather than filling that gap with library provision, event attendance, social-media following or an unsupported causal story.

Past-year club participation by audiobook listening, 2022

Weighted conditional participation among joint-valid adult responses, 2022 SPPA; approximate 95% sampling intervals.

Full chart and source
Among joint-valid adult responses in the 2022 Survey of Public Participation in the Arts, weighted past-year book-club or reading-group participation was 13.10 percent among audiobook listeners and 3.38 percent among nonlisteners. Approximate 95 percent sampling intervals were 9.06 to 17.15 and 2.41 to 4.35 percent. Intervals do not include uncertainty from item missingness.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Primary conditional participation estimates
Audiobook listeningValid nClub yes nParticipation (%)SE (pp)95% lower (%)95% upper (%)
Any5978113.10432.06539.056317.1524
None2,494923.37700.49432.40814.3458

Joint-valid adult responses only. Approximate 95% sampling intervals exclude item-missingness uncertainty.

Participation means book clubs or reading groups during the previous 12 months. Audio means any listening in the same recall period. The intervals reflect supplied-replicate sampling uncertainty, not item-missingness uncertainty. Census/NEA SPPA 2022. Original analysis and figure: Alisson Bento. No agency endorsement.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Data & downloads

Scroll horizontally to see all columns.

Primary conditional participation estimates
Audiobook listeningValid nClub yes nParticipation (%)SE (pp)95% lower (%)95% upper (%)
Any5978113.10432.06539.056317.1524
None2,494923.37700.49432.40814.3458

Joint-valid adult responses only. Approximate 95% sampling intervals exclude item-missingness uncertainty.

Scroll horizontally to see all columns.

Complete two-by-two weighted distribution
AudioClub participationnJoint share (%)SE (pp)
AnyYes812.45080.3899
AnyNo51616.25110.9567
NoneYes922.74540.4037
NoneNo2,40278.55281.0081

Primary joint-valid denominator 3,091. All four national cells retained. Percentages sum 100% before rounding.

Scroll horizontally to see all columns.

Conditional contrast and self-only sensitivity
Domain / metricn denominatorEstimate (% or pp)SE (pp)95% lower95% upper
Primary / Listening minus non-listening (pp)3,0919.72742.14805.517313.9374
Self-report only / Club participation among listeners (%)48313.38731.99379.479717.2950
Self-report only / Club participation among nonlisteners (%)1,8843.30720.49132.34424.2702
Self-report only / Listening minus non-listening (pp)2,36710.08012.07876.005914.1543

Difference=audio yes minus no. Self-only is a selected response domain, not a proxy-bias correction. Approximate sampling intervals only.

Scroll horizontally to see all columns.

Missingness within nominal joint assignment
Validity patternnWeighted share (%)
Both valid3,09192.4537
Audio valid; club missing691.9797
Audio missing; club valid161.0947
Both missing1194.4719

Denominator 3,295. No negative code was converted to No. Weighted shares use all nominal joint base weights.

Scroll horizontally to see all columns.

Observed-audio club-missingness bounds
AudioObserved audio nClub missing nLower (%)Upper (%)
Any6081112.968114.0078
None2,552583.29815.6343

Identification bounds, NOT confidence intervals. 135 missing-audio cases remain outside these observed-audio domains.

Evidence & method

How we calculated it

  1. One explicitly labelled 2022 update; prior 2017 NEA analysis credited. Binary any-audio grouping differs from prior reader types; no trend estimate.
  2. Local plan fixed before association calculations, not externally registered or immutable historical preregistration.
  3. Positive-supplement-weight adults 18+; recover Core 1 and Module C assignments from nonblank item patterns and reconcile 3,295 joint assignments. Primary 3,091 valid pairs; no unknown/nonselected answer becomes No.
  4. Factor 4 on PWSUPWGT and all 160 replicates for one core plus one module; source weights have four implied decimals, common scaling cancels in ratios.
  5. Each replicate recomputes ratios and contrast; variance=(4/160)sum squared deviations. Actual Census mixed successive-difference/modified-half-sample replication; covariance retained.
  6. Prespecified full four-cell distribution, coarse missingness, observed-audio worst-case club bounds and self-only sensitivity. Sampling intervals and missingness bounds are separate.
  7. Independent raw-file and aggregate computation reconciled estimates; default public script reproduces compact national sufficient-total arithmetic using explicit paths.

What the data cannot tell us

  • Formal membership, preference, meeting format, exclusive formats and causal effects are not measured. No link connects a listened-to book to a club selection.
  • Complete-response population interpretation requires unverified response assumptions; random module allocation does not prove item MCAR.
  • Observed-audio missing-club bounds exclude 135 missing-audio cases and are not all-adult contrast bounds or confidence intervals.
  • Primary includes proxies; self-only sensitivity is selected and does not establish absence of proxy bias.
  • No 2017–2022 trend or first-discovery claim; 2022 binary grouping differs from prior official reader-type analysis.

Original sources

  1. Census 2022 SPPA public-use data and replicate weights

    Public-use 2022 person file and supplied 160 replicate weights; national derived aggregates only. Raw fields have four implied decimals; aggregate stored weight scale cancels in ratios.

    Accessed October 30, 2026

  2. Census July 2022 CPS technical documentation

    Attachment 3 and Attachment 16 Table 1 specify joint factor 4; Attachment 7 pp 7-9/7-25 and questionnaire pp 11/32 define PEC1Q17A and PEMCQ1I. Proxy interviews allowed; PUNXTPR3 identifies response mode.

    Accessed October 30, 2026

  3. Census person-level supplement variance instructions

    March 7, 2012 instructions linked by 2022 documentation: module factor on base and replicates, recomputed statistics, variance 4/160; mixed successive-difference/modified-half-sample construction.

    Accessed October 30, 2026

  4. NEA 2022 SPPA technical appendix

    Table A-2 core/module allocation; pp 11–12 weights, missingness assumptions and equivalent Fay .5 MSE configuration. Published MCAR assumption is not empirical proof.

    Accessed October 30, 2026

  5. NEA 2020 How Do We Read? Let’s Count the Ways

    Prior 2017 analysis, pp 33–34 Figure 12 and Appendix B-4, already relates participation to reader types. Current 2022 binary any-audio grouping is not identical; no first-discovery or trend claim.

    Accessed October 30, 2026

  6. NEA 2023 Arts Participation Patterns in 2022

    Official 2022 highlights: the audiobook-listening share printed on p. 12 (18.5%) and the book-club share printed on p. 15 (roughly 5%) were used as coding checks, not as validation of missingness assumptions or joint conditional inference.

    Accessed October 30, 2026

  7. Census Open Data policy

    Public data access and reuse/replication framework, inspected 2026-10-30; no licence inferred for third-party report artwork, photographs, logos or questionnaire layout.

    Accessed October 30, 2026

  8. Census citation guidance

    Attribution guidance for original analysis of public Census statistics; no agency endorsement implied.

    Accessed October 30, 2026

  9. NEA public data resources

    Public access to arts datasets/analysis; source attribution retained. Not a blanket licence for third-party assets.

    Accessed October 30, 2026

Cite & reuse this study

Alisson Bento. Book-club participation and audiobook listening: a 2022 survey update. Original analysis of U.S. Census Bureau and National Endowment for the Arts 2022 Survey of Public Participation in the Arts, with reference to NEA 2020 analysis of 2017 reading types. https://alissonbento.com/research/book-club-participation-audiobook-listening-2022/

Original statistical analysis and figures: Alisson Bento. Underlying public-use statistics: U.S. Census Bureau / National Endowment for the Arts, 2022 Survey of Public Participation in the Arts. Prior 2017 analysis credited to NEA 2020 How Do We Read? Let's Count the Ways. No official endorsement. Census Open Data and citation policies support analytical reuse and replication with attribution; NEA offers public arts datasets. These are not a blanket licence for agency logos, report photography, artwork, third-party prose or questionnaire layout. No such assets, full source reports, microdata or respondent identifiers are included. Only original national aggregates, code and concise source descriptions are supplied. No licence for the whole article is invented.

Corrections or editorial questions

Pass it on.

Book-club participation and audiobook listening: a 2022 survey update | Alisson Bentoalissonbento.com

WhatsAppFacebookXTelegramRedditThreadsEmailSMS