Methodology & coding notes

Methodology

The dataset comes from parsing the embedded JavaScript in the Guardian's interactive feature page (May 2026). Every voter's full ranked top-10 is in that source, including books that didn't make the top 100. The parser extracts (voter, book, rank, position, commentary) tuples and pivots them into structured CSVs. The full pipeline — build scripts, analysis code, interactive site — is in the project's git repo.

Voter gender research

Each of the 172 voters is coded for gender — F, M, NB, or unknown. The coding was done in two passes. The first pass drew on general knowledge of contemporary literary figures (most of the voters are well-known novelists, critics, and Guardian journalists). The second pass was a careful validation: every voter was checked against an authoritative public source — typically the lead paragraph of their Wikipedia entry, with explicit attention to pronouns; failing that, their Guardian author-profile page, publisher bio, or personal website. The two-pass process caught three errors in the first-pass coding: Guy Gunaratne (M → NB; he/they), Lucas Rijneveld (NB → M, per his post-2022 use of he/him in English), and Olivia Laing (F → NB; they/them per Wikipedia).

Caveats about voter gender coding. Even with a validation pass, this kind of coding has real failure modes; the resulting CSV is an informed best reading, not ground truth:

• For lesser-known voters (a handful of Guardian staff and early-career critics) there is no Wikipedia entry and no publicly-stated pronoun usage that could be found. Coding for these voters relied on contextual cues — bio photographs, third-person references on publisher pages, named partners or spouses — and was marked as medium confidence. None of them stood out as ambiguous, but one or two could be wrong.
• The validation pass is a snapshot. A voter's publicly-stated identity at the time of coding may not match their identity today; pronouns and gender identity can change over time. Lucas Rijneveld is the clearest example (NB in 2020, M-identified in English since 2022).
• One voter — Sandra Newman — has publicly identified as gender-nonbinary in interviews, but Wikipedia and her publishers continue to use she/her. She is coded as F here and flagged as a borderline case in the source CSV. A reader could reasonably argue for NB instead.
• Yael van der Wouden is intersex and discussed this in her Women's Prize acceptance speech but accepts she/her pronouns and publicly identifies as a woman; coded F.

Net of these caveats, the voter-gender file represents a sincere best effort, and the headline statistics about gendered voting patterns are robust to plausible single-voter recoding — no individual voter's gender determines the F-versus-M ballot split.

Each book is coded along three axes:

• Subject score (−3 = very male-coded subject matter, +3 = very female-coded). The most subjective code in the project. It combines protagonist gender, the public-vs-domestic balance of the setting, the presence of violence or war, and overall thematic concerns. Half-integer values are used where a book sits between two integer categories.
• Canonicity score (−3 = deep traditional canon, +3 = recent diverse / expanded canon). Combines age of the book with author identity. Pre-1900 Western male authors anchor the canonical end; recent post-colonial or trans/queer voices anchor the idiosyncratic end.
• Publication year — objective, looked up.

Authors are also coded for gender, separately from voters. 479 distinct authors across the dataset are coded (75 for the top 100; 404 for the additional authors that appear only in long-tail picks). Authors who use pen names (George Eliot, George Orwell, Elena Ferrante) are coded as the gender the author publicly identified as.

For the chart's binary F-vs-M math, the two non-binary voters are not in the denominator or the marginal curves. They appear normally in voter lists and rankings — they just don't enter the percentage math. The Y-axis is labelled Voter gender* with vertical "All male" / "All female" endpoint labels; the asterisk footnote spells out the F ÷ (F + M) calculation and the non-binary voter handling.

Reading the four headline statistics

The four percentages at the top of the main page are parallel asymmetry measures, not complementary halves of a single whole. Each is computed against its own group's total ballots:

• "27% of male voters' ballots go to female authors" has all male-voter ballots in the denominator (684 ballots).
• "49% of female voters' ballots go to male authors" has all female-voter ballots in the denominator (1,003 ballots).

The two stats describe each gender's cross-the-line voting behaviour independently. Adding them together (27% + 49% = 76%) is not meaningful — they're different bases. Within each voter gender the complements do sum to 100% as expected (male voters: 27% to F authors + 73% to M authors = 100%; female voters: 49% to M authors + 51% to F authors = 100%). The same logic applies to the two subject-coded statistics.

Coding assumptions and limitations

Most of the codings in this analysis are judgment calls, and they are open to reasonable disagreement. It's worth being frank about which ones.

Subject score is the most subjective code. It compresses several distinct features of a book — protagonist gender, setting, themes, violence, narrative voice — into a single number on a stereotype gradient. The same book could legitimately be coded ±1 in either direction depending on which features you weight most. Some specific examples of the judgment calls made:

• Wolf Hall (−0.5): Cromwell at Tudor court is male-coded subject matter, but Hilary Mantel's intimate close-third prose technique — the famous "He, Cromwell" — sits in a stylistic lineage closer to Virginia Woolf than to Bernard Cornwell. The score was pulled from −1 to −0.5 to reflect that the prose reads less male than the plot summary.
• Anna Karenina (+1.5): Anna's marriage tragedy is the title plot, but Konstantin Levin's farm-life is structurally equal in page-count. Half-step.
• Pride and Prejudice (+3): A pure Regency marriage plot. Unambiguously at the top of the female-coded scale.
• Beloved (+2): Motherhood, slavery, ghost of an infant daughter; female protagonist; domestic setting; slavery's violence keeps it from +3.
• Catch-22 (−3): WWII bomber squadron, all-male crew, absurdist war. Maximum male-coded.
• The Vegetarian (+2.5): Yeong-hye refusing meat; body horror centered on a woman; family interpersonal violence. Female-coded but the family-violence element keeps it from +3.
• Moby-Dick (−3): All-male whaling crew, obsession, sea. Maximum male-coded.

Canonicity score is similarly judgment-based. It conflates two correlated factors — age and author identity — into a single number, which works for most books on this list but breaks down for cases like Anita Brookner (canonically venerated but not in the deep canon) or Toni Morrison (very canonical now but the canon adopted her relatively recently). Each book's canonicity score reflects where the book currently sits in the literary establishment's received wisdom, which is not identical to how venerated it was at publication.

The long-tail subject coding was done by a research agent working from plot summaries and brief author research — not as carefully as the top-100 codings, which were hand-done with individual rationale per book. Coding consistency across the long tail is therefore weaker than within the top 100. For the headline patterns this matters less (the signal at the top of the list dominates), but for any individual long-tail book, the score should be taken as approximate rather than precise.

Author gender for the 404 new authors was also agent-coded. Famous and confidently-known authors are reliable; obscure single-name authors with no clear public gender are coded as "unknown" rather than guessed.

The "voter pool" is the Guardian's curated list of contributors, not a random sample of readers. The patterns here describe the gendered reading habits of a specific community of literary professionals — novelists, critics, academics, Guardian journalists. Generalising to "what men read" or "what women read" outside this community would overreach. The voter pool is also 59% female, which itself shapes which books aggregate to the top 100 and which fall out as honourable-mentions.

← back to the chart