What Happens to a Series Between Books

Nobody has ever put a number on the mistake series authors actually write in about: the one that only exists between two books. So we measured it. Twenty two thousand public domain books were scanned with the same tool this site sells. Setting aside a second transcription of a novel the corpus already held leaves 21,737 of them, 1,220 series were identified by which books share a cast, and in 6.9 per cent of the 8,000 novel pairs inside them the two books spell one of their own invented words differently from each other. A clockwork character is hyphenated in one volume and not in the next; a compound noun is one word in book one and two words in book three. Nobody renames a character between volumes. They lose a hyphen.

The problem this measures

Every consistency tool, this one included, works on one manuscript at a time. That is the right unit for a standalone novel and the wrong one for a series, because the reader is not reading one book. They are reading five, often over five years, and they remember the spelling of the sword. The author, six months into volume four, does not.

There was no evidence about how often that actually happens. Not "how often does an author worry about it", which is constantly, but how often a finished, edited, printed series contains a word its own volumes disagree about. This is the first measurement of it we know of.

How a series was found, which is the hard part

Project Gutenberg's catalogue does not record series membership, and the titles that announce it are a minority of the ones that have it. So no series here was read off a title. Each one was measured, using the same rule the cross book check in this product applies: two books by the same author belong to the same continuity when they share a cast.

Four conditions, all of which have to hold:

• They share at least three proper character names that recur (five mentions or more) in both books.
• At least one of those shared names is a principal, twenty mentions or more, in both.
• The shared names are not ordinary English first names. A name recurring in more than five per cent of the whole corpus is excluded from the test. John recurs in 450 of these books and George in 318, and two novels sharing a John and a George is a fact about English naming rather than about the novels.
• The shared names are at least ten per cent of the smaller book's own recurring cast. Three coincidences in a cast of ninety is not a series. A genuine sequel pair in the corpus shares 39 per cent of theirs.

Two of those four rules exist because the first version of this study got it wrong. Without the common name rule, fifteen novels by one prolific Victorian realist came out as a single thirteen volume series, joined by three ordinary English first names. Without the cast share rule, an unrelated novel by the same author joined a genuine series on a handful of incidental names. Both are in the published method so anyone can disagree with either.

A separate test throws out pairs that are two editions of one book rather than two books, by title, by sharing more than 60 per cent of both vocabularies, or by sharing nearly the whole cast. It rejected 764 pairs. It is not optional: the corpus holds several novels twice, and before it was applied the largest cross book disagreement in the study was between two transcriptions of one Regency novel.

What survives is 1,220 series, 5,238 books, 748 authors, in series of up to 46 volumes. 511 of them are two book series, and at the other end sit long ones running to dozens of volumes: children's fantasy sequences, historical romances, detective and comic series.

What two books in one series actually share

This is the most useful table here and it is not the headline one. A pair of novels in a series holds a little over a quarter of its vocabulary and a little over a quarter of its recurring cast in common. Nearly three quarters of each book's cast is that book's own.

Which means the continuity surface is small and specific. It is not the whole book that has to match the last one. It is about twenty words and about six names, and the reason a series bible works is that twenty words is a list a person can hold. The reason series still drift is that nobody knows which twenty until they look.

How often the volumes disagree

Across 8,000 pairs of full length novels that share a cast, 6.9 per cent spell at least one shared invented word differently in the two books. Widen the population to every pair that shares a cast, which brings in the short story collections and the detective and comic series among them, and it is 7.4 per cent of 11,808 pairs, with 1,397 disagreements in total. Counted by series rather than by pair, 22.4 per cent of the 839 series that contain such a pair carry at least one disagreement somewhere.

Where a pair disagrees at all, the median pair disagrees about one word. This is not an epidemic. It is a small, specific, entirely findable class of error that survived professional editing anyway.

Of the 872 disagreements found between full length novels, 44.2 per cent are hyphenation, 41.3 per cent an accent, 7.7 per cent an apostrophe and 6.8 per cent capitalisation. Nobody renames a character between volumes. They lose a hyphen, or an accent.

What the disagreements actually look like

The shape is consistent and it is narrower than authors fear. Of the 872 disagreements found between full length novels, 44.2 per cent are a hyphen, 41.3 per cent an accent, 7.7 per cent an apostrophe and 6.8 per cent a capital letter; one is a changed letter. What changes between volumes is the punctuation inside a name that was invented once and then written from memory. The accents are nearly all the same foreign name set with its accent in one translation and without it in the next, which is a translator's decision as much as an author's.

The four shapes below are illustrations rather than extracts. They are four of the patterns the study found, written out with invented words: we do not publish which books are in the corpus, and a made up example shows the pattern as clearly as a real one.

The pattern worth taking from the real ones: the minority spelling is usually the later book, and it is usually the shorter, easier form. A writer four volumes in reaches for the word the way they say it rather than the way they first wrote it, and no spellchecker has an opinion about either.

Series books against standalone books

The obvious next question is whether writing in a series makes a book internally worse. The comparison has to be done carefully. Setting series books against the whole corpus would compare authors with several books here against authors with one, which is a fact about the corpus rather than about series writing. So both sides below are novels by authors with more than one book in the corpus, split by whether two of those books can be shown to share a cast.

Series novels are shorter than standalone ones and considerably more crowded: 1.3 times the cast density, and 35 named characters against 31. They carry a little more invented vocabulary, at 1.1 times; inside a single volume their inconsistency rate is 2.88 variants per 10,000 words against 2.81, a difference of under five per cent, which is no difference at all.

The interesting reading is the cast density. A series novel is doing the same work in the same words with more people in it, and it has to do that while agreeing with books written years earlier. That is the condition under which a hyphen goes missing.

What this cannot tell you

A difference between two volumes is a difference a reader of those two volumes would see. It is not proof the author wrote it that way. Every text here is a transcription of some edition, and two volumes of one series can come from different editions, set by different compositors and typed up decades apart. A hyphen is exactly the sort of thing a publisher's house style changes between printings. What the study can honestly claim is what is on the page in the two books, which is also the thing a reader complains about.

A word has to appear at least three times in both books before a difference is counted, which removes the single stray mis-scan and also removes real disagreements about words used once or twice. The counts here are floors, not ceilings.

Series membership is measured rather than catalogued, so a sequel with an entirely new cast is invisible to this method. Genuine series are still lost to the ten per cent cast share rule, because a book whose cast turns over every chapter shares a small fraction of it however true the series is. And the corpus is public domain, which means this mostly measures series written before 1930 and says nothing about a series being written now; the published methodology states that limit and every other one.

What to do about it in your own series

The practical finding is the size of the surface. About twenty invented words and about six recurring names is what two volumes actually have in common, and that is a list, not a project.

The Novel Kit's series check is the same comparison this study ran, pointed at your books instead of these: scan two or more manuscripts and it reports every term they share, every term one has and the other does not, and every name and invented term the books spell differently from each other. The series bible generator turns the shared list into the document you carry into the next volume, and compare two drafts does the same job between versions of one book.

If you have already published two books and are drafting the third, the order that works is: scan the published ones first, settle the spellings they actually used rather than the ones you meant to use, then check the draft against that. Consistency with what is in print beats consistency with your intentions, because readers reread.

Sources checked

Every figure here is generated from the Novel Manuscript Benchmark, a public dataset of public domain books scanned with the tool this site sells, on its default settings. The series list and every cross book disagreement are published below as CSV files and are regenerated whenever the corpus grows, so a later run will move these numbers and the dataset rather than this page is the authority.

No customer manuscript was involved and none could have been. Nothing in this product's storage holds manuscript text or extracted terms, which its own test suite enforces, and the corpus is public domain fiction precisely so that a study like this can exist without going near anyone's work in progress. Each book was scanned by the engine that runs a scan for a reader, so a figure here is comparable with a figure from a scan of your own.

Frequently asked questions