Skip to content

Original research · Alisson Bento

When an Author Field Is Not Enough: Contributor Filters in 103,963 DOAB Book Records

A book-record selection can exclude an explicitly credited volume simply because its rule accepts authors but not editors. In a fixed Directory of Open Access Books metadata export, accepting either field retains 102,658 records rather than 58,808. The additional 43,850 records constitute 42.18% of the selected frame—not a measured improvement in website search, but a consequential difference between two precisely defined metadata filters.

Published
Share
Share this page
EmailCite & reuse this study
Key findings as text
Reusable imagesWhen an Author Field Is Not Enough: Contributor Filters in 103,963 DOAB Book Records — 58,808: Records retained when an author field is required. 56.57% of all 103,963 selected book records have a populated author field.Download sharing image (JPG) ↓

Key findings

58,808

Records retained when an author field is required

56.57% of all 103,963 selected book records have a populated author field.

43,850

Additional records when editor fields also qualify

These editor-positive, author-negative records represent 42.18% of the full 103,963-record frame, not a 42.18% increase over the author-required selection.

102,731

Records retained when any of three contributor fields qualifies

98.81% of the selected frame; accepting the other-contributor field adds 73 records beyond the author-or-editor rule.

1,184

Records with more token occurrences than distinct exact tokens

Literal semicolon-separated strings repeat within or across contributor fields in these records. This is not a count of duplicate people or erroneous credits.

The selection rule changes the collection you see

Suppose a reading-list compiler starts with book metadata and keeps only records containing an author credit. That rule sounds like a straightforward way to retain books with named responsibility. It is more specific than that, however: it retains records with a populated author field. An editor credit elsewhere in the same record does not satisfy the rule. The distinction matters whenever the intended selection includes books represented through editorial rather than authorial responsibility. Here, the question is not whether those books deserve inclusion, but how many records a particular field rule allows through.

In the selected DOAB export, requiring the author field retains 58,808 of 103,963 book records. Allowing either the author or editor field retains 102,658. The difference is 43,850 records with an editor field populated and an author field unpopulated. Nothing about this calculation requires recognizing a person's name, deciding whether two names identify the same person, or evaluating a book's contents. It is a direct comparison of populated cells. That independence from name parsing makes the central result especially clear: the selection changes because the accepted roles change.

For a reader assembling a subject bibliography, or a library preparing a downstream collection, the useful lesson is to state the intended role coverage before applying a filter. A list intended specifically to study author-attributed records can legitimately require that field. A list intended to include both authored and edited books needs a different rule. This study supplies the size and composition of that difference for one fixed export. It does not show that DOAB's website excludes edited books, that any reader failed to find a title, or that either selection is universally preferable.

What the snapshot contains—and what its date means

The evidence comes from the complete official DOAB CSV response retrieved on 30 October 2026 at 08:37:57.351428 UTC. That response contained 114,618 export rows. The analysis selects rows whose type field, dc.type, is exactly book and whose containing-book field, oapen.relation.isPartOfBook, is empty after trimming surrounding whitespace. These conditions yield 103,963 book records. The containing-book condition is part of the frame definition, not a claim that every remaining row represents a distinct intellectual work or that every excluded row lacks value for readers.

The retrieval timestamp pins the acquired file; it does not date the books or establish the provider's exact database state at that instant. No provider generation timestamp was verified. Accordingly, references here to the snapshot mean this particular retrieved response, not a census of everything DOAB has ever held or every open-access book available on that date. The file's recorded size and SHA256 fingerprint identify the source bytes underlying the counts. A later response from the same official URL could differ without contradicting any result reported for this one.

Within the selected frame, exact UUID and handle strings are nonempty and unique. That is useful for distinguishing these records, but it does not settle whether separate records describe different editions, manifestations, or versions of the same intellectual work. All denominators in this article therefore refer to book records. A specialist collection may reasonably retain multiple editions, while another bibliography may consolidate them. Neither decision has been imposed here. Readers can interpret the filter results without assuming that a record total is also a total of unique books in every bibliographic sense.

Eight profiles make the missing author field interpretable

Three fields, each either populated or unpopulated, produce eight possible profiles. The first table presents all eight in binary author/editor/other order rather than ranking them by frequency or importance. A populated cell means that its trimmed text is nonempty. An unpopulated cell means only that this representation contains no text in that field. This deliberately narrow definition prevents a common interpretive jump: absence from a particular metadata cell is not evidence that no corresponding contributor exists in the book, its front matter, or another catalogue.

The largest profile contains 55,673 records with an author field alone populated, accounting for 53.55% of the complete selected frame. The next contains 43,404 records with an editor field alone populated, or 41.75%. These two profiles explain much of the difference between the filters, but not all of it. Another 446 records have editor and other-contributor fields populated without an author field. Together, those 43,404 and 446 records form the 43,850 additions obtained when editor credits become eligible. The composition is visible rather than hidden inside a single headline total.

The smaller profiles are necessary to complete the interpretation. There are 351 author-plus-other records, 2,578 author-plus-editor records, and 206 with all three fields populated. Another 73 have only the other-contributor field populated, while 1,232 have all three fields empty. These are mutually exclusive categories of records, unlike the overlapping filter totals discussed later. The profile table therefore answers a different question from the retention table: it describes how fields coexist within records, including combinations that would be indistinguishable if the analysis reported only whether any contributor text appeared.

Editor-positive is evidence, not a failed author credit

An editor-positive record makes a positive assertion of editorial responsibility in the exported metadata. It should not automatically be described as defective because the author field is blank. An edited volume can be represented through its editor without the book-level record also assigning authorship. This is an interpretive possibility consistent with the role distinction, not an inspected explanation for every record in this frame. The study did not examine the books' title pages or compare their credits with the exported cells, so it cannot certify whether an individual assignment is complete or appropriate.

For collection building, the immediate question is whether editorial responsibility belongs within the intended scope. If a reader wants book-level works relevant to a topic, excluding editor-positive records merely because they lack an author cell can narrow the candidate set in an unintended way. If the task instead concerns records explicitly classified under author responsibility, the narrower selection may be exactly right. The empirical finding does not decide between these purposes. It shows that treating them as equivalent would obscure 43,850 records in this snapshot, a substantial difference that warrants an explicit choice.

The other-contributor field requires a separate decision. Its populated text cannot silently be recast as authorship, because the field may represent different kinds of contribution and this analysis does not classify them. Allowing it as an alternative eligibility condition means accepting a broader category of recorded responsibility, not proving the presence of an author or editor. Similarly, the 1,232 all-empty profiles should remain described as records without text in these three cells. Calling them uncredited books would turn a limited observation about metadata into an unsupported statement about the books themselves.

Three nested filters, three different denominators

The retention comparison proceeds through three nested selections. First, keep records with a populated author field. Second, keep records with a populated author or editor field. Third, keep records with any populated author, editor, or other-contributor field. Each broader rule includes every record retained by the preceding one. Their totals—58,808, 102,658, and 102,731—must therefore not be added together. The meaningful additions are the differences between adjacent rules: 43,850 when editors qualify, then 73 when other contributors also qualify. The second table keeps these totals and increments separate.

Using the complete frame as denominator, the three rules retain 56.57%, 98.74%, and 98.81%, respectively. The editor-positive additions represent 42.18% of all 103,963 selected records. That is the most direct measure of how much of this frame changes status between the first two rules. A different question is what proportion of the broader author-or-editor selection consists of those additions: the answer is 42.71% of 102,658. Neither percentage expresses the increase relative to the smaller author-required total. Keeping the denominator attached to each statement prevents an apparently small wording change from altering its meaning.

Relative to the 58,808 records retained initially, the 43,850 additions constitute a 74.56% increment. That ratio is legitimate when describing expansion from the narrower selection, but it is not interchangeable with the 42.18% full-frame share. For a library explaining a selection policy, reporting both the retained count and the chosen denominator is more informative than saying only that coverage increased. These are exact descriptive comparisons of a fixed frame, with percentages rounded as stated. They need no manufactured confidence interval: the uncertainty lies in extending their meaning beyond the observed export, not in sampling the records counted here.

Counting credit strings without pretending to count people

Field presence answers the main selection question, but it does not describe how much contributor text each retained record contains. A second calculation therefore counts literal tokens within the three role fields. The recipe is explicit: split on literal semicolons, trim each resulting component, discard empty components, and preserve repeated strings. Commas remain inside tokens. Thus, a comma within a name is not treated as a boundary between contributors. The resulting unit is a token occurrence under this recipe, not an independently verified person, identity, or bibliographically complete credit.

This distinction matters because the inspected official guide does not guarantee that every semicolon identifies exactly one creator. A token may be useful for comparing the structure of exported fields without being a reliable unit for estimating the number of people who made a book. Two nonempty fields contain an empty semicolon component; dropping such components prevents punctuation alone from increasing the token count. That cleaning step belongs to the declared counting rule. It does not establish that all remaining components have been correctly interpreted as personal names or that any credit has been checked against its source.

For readers, these counts are most useful as descriptions of record representation. They indicate whether a profile tends to carry one literal credit string or several, and how far a profile's observed maximum extends. They cannot establish how many collaborators worked on a volume, how responsibility was shared, or whose contribution should determine a reading-list entry. Crucially, none of these parsing limitations changes whether a trimmed field is populated. The large retention difference between the author-required and author-or-editor rules therefore remains independent of the more interpretively restricted token analysis.

What the exact-token union reveals about repetition

The companion count takes the union of distinct exact trimmed token strings across all three role fields within each record. It does not change case, merge aliases, expand initials, reconcile identifiers, or decide whether differently written strings refer to one person. If the same exact string appears more than once, the occurrence count preserves each appearance while the union includes that string once. This comparison asks how literal repetition affects the numerical summary. It does not produce a corrected count of people, because exact textual sameness and personal identity are different kinds of evidence.

There are 1,184 records in which token occurrences exceed the exact-token union. Repetition can occur within a field or across roles. The calculation alone does not say whether a repeated string is an accidental duplicate, an appropriate credit in more than one capacity, or a coincidence involving distinct contributors with the same written name. Treating every difference as an error would exceed the evidence. Conversely, equal occurrence and union counts do not certify that all contributors are distinct people: different strings can refer to the same individual, and this analysis deliberately leaves such possibilities unresolved.

The union is best read as a sensitivity check on a particular representational choice. If a summary changes when exact repeated strings are collapsed, the reader knows that literal multiplicity contributes to the result. If it does not change, the summary is insensitive to that narrow operation, not validated as a census of creators. This is especially useful when comparing mixed-role profiles, where one string may appear in more than one role. The analysis preserves those role assertions in the underlying occurrence count while showing what happens when exact strings are counted only once per record.

Read the token summaries within their role profiles

The profile-conditioned medians distinguish a typical record from a profile's extreme case. Author-only records have a median of one token occurrence and an exact-token-union median of one, yet their maximum occurrence count is 113. Editor-only records have corresponding medians of two and two, with a maximum of 38. These summaries describe the distribution of the exported text under the stated recipe. They do not establish that edited books involve more people than authored books in general, nor that the author-only record with the largest count represents a verified group of 113 distinct creators.

Mixed profiles show why both count definitions are worth retaining. Editor-plus-other records and author-plus-editor records each have a median of five occurrences but a median exact-token union of four. Records with all three role fields populated have medians of six and five, respectively. These are separately calculated distribution summaries, not a claim that every record at one median occupies the corresponding position in the other distribution. They demonstrate that repeated exact strings affect the central summaries in these profiles. They do not identify which credits overlap or determine whether any overlap is appropriate.

The maxima reinforce the need to keep composition and scale separate. Editor-plus-other records reach 154 token occurrences, author-plus-editor records reach 147, and all-three-role records reach 47. The profile with more populated roles does not therefore have the largest observed maximum. Other-only records have medians of one and one and a maximum of 14; author-plus-other records have medians of two and two and a maximum of 32. These compact summaries help a reader compare representations without turning an extreme observation into a typical case or treating a role combination as a ranking of bibliographic complexity.

Why the manual-entry limit does not explain the counts

The August 2025 DOAB Publishers' Guide supplies relevant submission context. On pages 6 and 11, it describes a maximum of 20 creators for manually entered works and expressly states that ONIX submissions have no such limit. This is a distinction between submission routes in a dated guide, not a universal ceiling on every exported record. The guide predates the October 2026 retrieval, but its date does not establish which instructions governed any individual record when it was entered or subsequently revised. Its role here is to bound interpretation, not to reconstruct ingestion histories.

The selected frame contains 63 records with exactly 20 token occurrences and 463 with more than 20. Those observations cannot be translated into counts of truncated books or manual submissions. The analysis did not examine ingestion routes, original submissions, earlier record versions, or complete creator lists. Moreover, the measured unit is a literal token occurrence, whereas the guide describes creators. Even a conspicuous count at a documented limit would not by itself connect those units or establish the route through which a record reached the catalogue. A count above 20 likewise does not prove ONIX ingestion.

For someone inspecting a metadata export, the practical value of the guide is therefore caution against an easy but unsupported causal story. A documented limit can suggest a question for a different investigation, but it cannot answer that question through coincidence with a token count. Establishing truncation would require evidence connecting a record's exported credits with its submission history and fuller source credits. None of that evidence is supplied here. The defensible result remains the observed number of records at and above the token threshold, accompanied by the guide's explicit distinction between manual entry and ONIX.

Using the findings in a reading list or library selection

A useful application begins by translating a collection's purpose into an eligibility statement. If the purpose is to assemble book records with either authorial or editorial responsibility explicitly represented, the author-or-editor rule matches that wording more closely than an author-only rule. In this frame, the choice retains 102,658 rather than 58,808 records. That is evidence about candidate-set construction, not evidence that every additional record belongs in a particular subject collection. Subject relevance, edition choice, language, and other selection requirements remain separate questions that this contributor-field analysis does not answer.

A second application is to preserve role information after selection. Accepting an editor field as a reason to retain a record need not mean displaying its editor as an author. Likewise, allowing an other-contributor field can widen eligibility without changing the meaning of that credit. This separation between admission to a list and description within it keeps the analytical rule intelligible to readers. It also makes the 73 other-only records visible as a distinct decision rather than hiding them inside a broad claim that every retained record has an author or editor.

A third application is to explain exclusions accurately. A record rejected by an author-required rule has failed that rule; it has not thereby been shown to lack named responsibility or to be unsuitable for reading. The profile table gives a precise vocabulary for describing what happened. For the 43,850 editor-positive additions, the decisive fact is an accepted alternative role field. For the 1,232 all-empty profiles, the decisive fact is the absence of text from the three inspected cells. Neither result measures actual discoverability, accessibility, reading quality, or the success of a reader's search on DOAB.

The contribution is a reproducible comparison, not a universal verdict

The original contribution is the combination of a fixed-frame role-filter comparison with a complete eight-profile description and a bounded sensitivity analysis of literal credit strings. Existing DOAB documentation supplies the export context, metadata reuse terms, and submission guidance; the field-selection rules and calculations here are Alisson Bento's analysis of that material. The work does not claim to introduce metadata auditing or string comparison as new methods. Its value lies in showing exactly how a familiar-looking requirement changes the records retained in a specific, identifiable source.

The headline result has a firm local meaning: accepting editors alongside authors adds 43,850 records, or 42.18% of the selected frame. It has no demonstrated universal omission rate attached to it. Another export, a different containing-book condition, a deduplicated work-level collection, or a different catalogue could yield different counts. Nor does the result justify a publisher league table or metadata-quality score. A field profile is not a verdict on whether a book's credits are correct, and a broad filter is not automatically superior to a narrow one designed for a different question.

For readers and libraries, the lasting distinction is between a book's responsibility as represented in separate fields and the rule used to admit that record into a collection. Making the rule explicit permits a reviewable choice: author required, author or editor, or any of the three contributor fields. Reporting the profiles then reveals what each choice leaves behind. The study stops at that evidence boundary. It does not turn metadata into biography, infer missing creators from empty cells, or equate retention with successful discovery. It gives a concrete, reproducible basis for choosing and explaining a contributor filter.

Which contributor fields are populated?

DOAB metadata · Book-record snapshot retrieved 30 October 2026

Full chart and source
Which contributor fields are populated? DOAB metadata · Book-record snapshot retrieved 30 October 2026. Frame: 103,963 book records. None of the three: 1,232; Other only: 73; Editor only: 43,404; Editor + other: 446; Author only: 55,673; Author + other: 351; Author + editor: 2,578; All three: 206. Profiles refer to populated author, editor and other-contributor fields. An empty field is not proof of absent real-world credit. Records are not unique works.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Contributor fields and literal-token profiles
Author field populatedEditor field populatedOther-contributor field populatedBook recordsShare of frame (%)Median token occurrencesMedian exact-token unionMaximum token occurrences
NoNoNo1,2321.19000
NoNoYes730.071114
NoYesNo43,40441.752238
NoYesYes4460.4354154
YesNoNo55,67353.5511113
YesNoYes3510.342232
YesYesNo2,5782.4854147
YesYesYes2060.206547

All 103,963 selected book records. Yes/No means nonempty/empty after trimming, not actual presence/absence of a credited person. Semicolon token occurrences preserve repetitions; exact-token union compares trimmed strings without person resolution. Percentages use the full book-record denominator and are rounded to 2 decimals.

Profiles refer to populated author, editor and other-contributor fields. An empty field is not proof of absent real-world credit. Records are not unique works. DOAB source metadata: CC0 · Original analysis: Alisson Bento

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

Accepting editor fields retains more records

DOAB metadata · Book-record snapshot retrieved 30 October 2026

Full chart and source
Accepting editor fields retains more records. DOAB metadata · Book-record snapshot retrieved 30 October 2026. Frame: 103,963 book records. Author field required: 58,808 (56.57%); Author OR editor: 102,658 (98.74%); Any of the three fields: 102,731 (98.81%). Nested hypothetical CSV filters; not observed DOAB search-interface behaviour. The broader filters add 43,850 and then 73 records. Do not sum retained totals.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Retention under nested hypothetical CSV field filters
Hypothetical CSV field filterRetained book recordsShare of frame (%)Additional records vs previous filter
Author field required58,80856.57Not applicable
Author OR editor field102,65898.7443,850
Any author, editor or other-contributor field102,73198.8173

The filters are nested analytic selections, not observed DOAB search-interface behaviour. All shares use 103,963 book records. Added records compare each filter with the immediately preceding filter; the first row has no preceding filter.

Nested hypothetical CSV filters; not observed DOAB search-interface behaviour. The broader filters add 43,850 and then 73 records. Do not sum retained totals. DOAB source metadata: CC0 · Original analysis: Alisson Bento

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

Data & downloads

Scroll horizontally to see all columns.

Contributor fields and literal-token profiles
Author field populatedEditor field populatedOther-contributor field populatedBook recordsShare of frame (%)Median token occurrencesMedian exact-token unionMaximum token occurrences
NoNoNo1,2321.19000
NoNoYes730.071114
NoYesNo43,40441.752238
NoYesYes4460.4354154
YesNoNo55,67353.5511113
YesNoYes3510.342232
YesYesNo2,5782.4854147
YesYesYes2060.206547

All 103,963 selected book records. Yes/No means nonempty/empty after trimming, not actual presence/absence of a credited person. Semicolon token occurrences preserve repetitions; exact-token union compares trimmed strings without person resolution. Percentages use the full book-record denominator and are rounded to 2 decimals.

Scroll horizontally to see all columns.

Retention under nested hypothetical CSV field filters
Hypothetical CSV field filterRetained book recordsShare of frame (%)Additional records vs previous filter
Author field required58,80856.57Not applicable
Author OR editor field102,65898.7443,850
Any author, editor or other-contributor field102,73198.8173

The filters are nested analytic selections, not observed DOAB search-interface behaviour. All shares use 103,963 book records. Added records compare each filter with the immediately preceding filter; the first row has no preceding filter.

Evidence & method

How we calculated it

  1. Use the complete official DOAB CSV response retrieved on 30 October 2026 at 08:37:57.351428 UTC: 114,618 export rows, 230,316,814 bytes, SHA256 cddd2660be7cbb14cc88eeb854e31e1346767508d69dd74c1d46e5aba9e4c5e6. The retrieval timestamp identifies the acquired snapshot; no provider generation timestamp was verified.
  2. Select exact dc.type == book and an empty oapen.relation.isPartOfBook field after trimming whitespace. This defines a complete descriptive frame of 103,963 records, not a deduplicated set of intellectual works.
  3. Classify each record by trimmed nonempty versus empty author, editor, and other-contributor cells. Tabulate all eight mutually exclusive profiles and calculate three nested field-presence selections: author; author OR editor; author OR editor OR other contributor.
  4. For literal-token occurrences, split each of the three contributor fields only on semicolons, trim components, discard empty components, retain commas, and preserve repetitions. For the exact-token union, count distinct exact trimmed strings across those fields within each record, without identity resolution or case folding.
  5. Calculate profile-conditioned descriptive summaries and fixed-frame retention totals. Table percentages use 103,963 records and round to two decimal places. Alternative ratios in the prose explicitly identify their different denominators. No inferential confidence intervals are assigned to these complete descriptive counts.
  6. Use the official metadata documentation for feed and reuse context and the August 2025 Publishers' Guide for manual-entry and ONIX context. The source had already been used for other studies; this article makes no claim of outcome-blind preregistration.

What the data cannot tell us

  • The analysis tests hypothetical CSV field filters, not DOAB website queries, actual search recall, reader outcomes, or accessibility.
  • Results describe one retrieved export and the stated selection conditions; they are not a census of all books, a universal omission rate, or a verified exact historical database state.
  • Unique nonempty UUID and handle strings identify selected records but do not resolve editions, manifestations, or duplicate intellectual works.
  • Empty contributor cells establish only missing text in those fields. The 73 other-only and 1,232 all-empty profiles do not establish the absence of authors, editors, or other creators from the books.
  • Semicolon tokens are literal trimmed strings, not validated people or authoritative credits. Exact-token unions neither resolve identities nor diagnose erroneous repetition.
  • The manual-entry cap does not apply to ONIX submissions. No ingestion route, original submission, earlier version, full creator list, truncation, or policy violation was established for any record.
  • The analysis does not evaluate the correctness of role assignments, classify other-contributor roles, assess subject relevance, or create book, publisher, or metadata-quality rankings.
  • The retrieval date is distinct from the August 2025 guide date and from book publication dates. No claim of prospective, outcome-blind registration is made.

Original sources

  1. DOAB — official metadata CSV snapshot

    DOAB — official metadata CSV snapshot: Retrieved 2026-10-30T08:37:57.351428Z: 230,316,814 bytes; SHA256 cddd2660be7cbb14cc88eeb854e31e1346767508d69dd74c1d46e5aba9e4c5e6. The 114,618 export rows yield 103,963 selected book records. No provider generation timestamp or work/person deduplication was established.

    Accessed October 30, 2026

  2. DOAB — metadata access and reuse documentation

    DOAB — metadata access and reuse documentation: Identifies the CSV feed and states CC0 availability for metadata, not book full text or this article. Frozen HTML SHA256: a18c635b5950db92a7186c3974c7f2ea32b5f3d30900d83b789507e240e084ed.

    Accessed October 30, 2026

  3. DOAB — Publishers’ Guide, August 2025

    DOAB — Publishers’ Guide, August 2025: Pages 6 and 11 describe a 20-creator manual-entry cap and no ONIX limit; pages 14–16 discuss CSV export and identifiers. These passages do not establish any record's ingestion route or truncation. The guide is cited, not redistributed or relabelled CC0. Frozen PDF SHA256: ed3b61e386c3267ba041843e71d9bb69f7915a9c564e25225882895c003175ee.

    Accessed October 30, 2026

Cite & reuse this study

Alisson Bento. 2026. “When an Author Field Is Not Enough: Contributor Filters in 103,963 DOAB Book Records.” Original research. https://alissonbento.com/research/doab-creator-role-filtering-2026/

DOAB's metadata documentation states that its metadata are available under CC0. That dedication concerns the source metadata, not the books' full text, the Publishers' Guide PDF, or this complete article. The field-filter definitions, calculations, interpretation, and original prose presented here are Alisson Bento's original contribution; underlying DOAB metadata and third-party documentation retain their respective provenance. Only compact aggregates and selected metadata-derived inputs are redistributed. No blanket article licence, book-text permission, or institutional endorsement is asserted.

Corrections or editorial questions

Pass it on.

When an Author Field Is Not Enough: Contributor Filters in 103,963 DOAB Book Records | Alisson Bentoalissonbento.com

WhatsAppFacebookXTelegramRedditThreadsEmailSMS