Who wrote “A Visit from St. Nicholas”?

The mystery

Two days before Christmas 1823, the Troy Sentinel, a newspaper in Troy, New York, a river town just north of Albany, gave part of its third page to fifty-six lines of verse headed “Account of a Visit from St. Nicholas”. Its editor, Orville L. Holley, set a note above them: “We know not to whom we are indebted for the following description of that unwearied patron of children.”¹ The lines (545 words, beginning “’Twas the night before Christmas”) gave St. Nicholas his eight reindeer and their names, his miniature sleigh, his descent by the chimney, his pipe and his round belly. Other newspapers copied them, the Sentinel's printer, Norman Tuttle, sold them as an illustrated sheet about 1830, and for thirteen years they circulated without an author's name.

In December 1836 a New York anthology, The New-York Book of Poetry, printed the poem as the work of Clement Clarke Moore, a professor of Oriental and Greek literature at the General Theological Seminary, the Episcopal Church's seminary in Manhattan.⁶ In December 1843 a Washington newspaper credited it to a dead artist, and Moore answered in print: the lines were his, written “many years ago … not for publication, but to amuse my children”, and had reached the Troy paper without his knowledge.¹⁰ In the summer of 1844 he put the poem in his collected Poems, and until his death in 1863 he wrote out signed copies of it for friends and collectors.^{11, 12}

Poughkeepsie, some seventy miles down the Hudson from Troy, had its own paper, the Poughkeepsie Journal, and on 16 January 1828 it too printed the poem, without a name.⁴ Six weeks later Henry Livingston Jr. died there, at his farm Locust Grove, aged seventy-nine. He had been a surveyor, a major in the Revolution, and a writer of light verse for his family and the local papers, much of it in the poem's metre. No one in his house is recorded as saying anything about the poem, then or for fifty years after. In 1879 one of his daughters, by then in her seventies, wrote to a great-niece that the family had always believed the poem was their father's and had been astonished to see it claimed for Moore.¹⁴ In 1899 the family said so in public, and by 1920 the claim was in a national magazine. The family's account was that Livingston had read the poem to his children years before 1823, that a manuscript in his hand had passed to his sons and burned in a house fire in Wisconsin in the 1840s, according to the family, that a Poughkeepsie paper had printed it before Troy did, and that a governess from the Livingston household had later carried a copy to the Moore family.^{15, 16, 17} Neither the manuscript nor that printing has ever been found.

Each side has evidence that the other has not answered. For Moore there is a sequence of dated documents. The Sentinel's editor wrote in January 1829, eleven months after Livingston's death, that the author was a New York City man of “more merit as a scholar and a writer than many of more noisy pretensions”, and recalled in 1836 that he had learned within months of the first printing that the lines came from “a professor in one of our colleges”.^{3, 2} A student's diary of 1834 describes a costume made “according to the description of Prof Moore in his poem”.⁵ Moore himself supplied it to the 1836 anthology, claimed it in 1844, and copied it out in his own hand for the rest of his life. The evidence against Moore is that he said nothing for twenty years; that no manuscript from before 1823 survives for either man; and that most of what else he wrote is grave and iambic (the da-DUM of most English verse), whereas the poem is comic and anapestic (da-da-DUM), a metre he used in only four other surviving poems.

For Livingston there are two things: the family's memory and the manner of the poem. The memory was first written down in 1879, fifty-six years after the printing; the relatives who recorded it disagree on the date, on which child remembered it and on which newspaper printed it; and, as the family admitted in 1900, no document from Livingston's lifetime supports it. Livingston habitually wrote comic verse in the poem's metre, and those who have argued his case find the poem closer in manner to his verse than to Moore's. In 2000 Don Foster, an English professor at Vassar College in Poughkeepsie who had become known for identifying the anonymous author of the novel Primary Colors, made that case at length in his book Author Unknown.²⁷ In 2016 MacDonald P. Jackson, a New Zealand scholar who had spent a career on the authorship of Shakespeare's plays, tested the claim statistically: he counted the poem's common words, its speech sounds and its favourite expressions against everything each man was known to have written, and on his tests the poem scored with Livingston.²⁰ In 2017, answering Scott Norsworthy, an independent scholar who had reviewed his book on his blog Melvilliana, Jackson wrote that the documents favoured Moore and the text favoured Livingston, and that he could not reconcile the two.²²

First I collected every dated document that bears on the question, from the 1823 printing to the family's letters of 1900, and sorted them by what I could check myself: the ones I read on a scan of the original or of a period printing, and the ones I know only through someone else's transcription. Then I rebuilt the statistical comparison from scratch. Stylometry, the method Jackson used, counts a writer's habits: which common words, letter sequences and speech sounds appear, and how often. It needs bodies of text to learn from, so I built corpora of both men's verse for this report: forty-six poems by Moore (the twenty-nine in his 1844 Poems, proofread against the scan, and seventeen manuscript poems taken from the transcriptions on henrylivingston.com, a website kept by a Livingston descendant) and sixty-six by Livingston, also taken from that site's transcriptions and not checked against originals. To these I added 2,756 poems by thirty-two other poets of the period, chosen to include the best-known writers of comic verse in the poem's metre, so that every test could be tried on authors who are not in dispute.

Before measuring the poem I tested something the published studies had not tested: how often each method gives the wrong answer on poems whose author is already known. Cut into 56-line samples, the length of the poem, the two men's verse gave 132 such samples. Standard tests trained on everything each man wrote sorted them 80 to 89 percent correctly, but the word and letter-sequence tests attributed Moore's own four comic anapestic poems to Livingston every time or all but once. They had learned that Moore writes in iambs and Livingston in anapests, and were attributing by metre rather than by author. Built from the anapestic verse of both men only, the same tests sort Moore's four correctly, and the poem moves to his side. Jackson's two strongest tests reproduce on my corpora, with the poem scoring above all four of Moore's anapests, but by margins about the size of the spread of scores among 56-line samples by a single author, and one of the two tests, the speech-sound test, is shifted toward Livingston by anapestic metre itself. A further test against twenty-nine other poets of the period could not verify either man as the author at all.

So the statistical results are close to even and weak for both men, whereas the documents favour Moore. Nothing written in Livingston's lifetime, or for fifty years after it, connects him with the poem; it was printed in his own town while he was alive and nobody in his house is recorded as reacting; and the editor who printed it placed the author in New York City eleven months after Livingston died. I put Moore at about 85 percent, Livingston at about 12 and someone else at 3. The figures are a judgment; no calculation from the statistical results produced them. The section “How I got to 85” shows how each piece of evidence contributed. A Poughkeepsie newspaper from before 1823 with the poem in it, or a dated document from before 1844 in which any Livingston mentions it, would change my mind.

What this report covers

The report goes through the evidence in this order. Two claimants gives the two men's lives and the history of the dispute, with the sources cited. The paper trail sets out every dated document from 1823 to 1920 and says which I checked against a scan and which I know only from a transcription. Building the corpora describes the bodies of text I compared: Moore's twenty-nine printed and seventeen manuscript poems, sixty-six by Livingston, and 2,756 by thirty-two other poets of the period. Testing the method on known poems tries the attribution methods on poems whose author is known, which is where the effect of metre on the tests appears. The poem measured applies them to the poem itself. Jackson's tests, rerun rebuilds his two strongest tests on the new corpora. Words, metre and rhyme goes through the individual markers each side has cited, from the word “like” to the reindeer names “Dunder and Blixem”. The impostors test asks whether either man can be verified at all against twenty-nine other poets. How I got to 85 shows how each piece of evidence was weighed to reach the figure of 85 percent. The case for Livingston puts the other side at its strongest and says what would overturn the verdict. Three short sections follow, on what is new here, where this falls short, and the wider point about stylometry. Sources and files lists the documents, the scholarship and the data files, so that each step can be checked.

Two claimants

Clement Clarke Moore (1779–1863) was professor of Oriental and Greek literature at the General Theological Seminary in New York. The poem appeared under his name in The New-York Book of Poetry, an anthology of local verse edited by the writer Charles Fenno Hoffman (dated 1837, issued December 1836), among pieces Moore supplied himself.⁶ He claimed it in a public letter of 27 February 1844 and put it in his Poems that June.^{10, 11}

Henry Livingston Jr. (1748–1828) was a landowner, surveyor and occasional versifier at Locust Grove, near Poughkeepsie. He wrote light verse in anapests, and he died on 29 February 1828 without ever claiming the poem. His descendants' belief that he wrote it was first written down in a letter of 1879 (the year is assigned from internal evidence), argued in print from 1899, and given its modern form by Don Foster in Author Unknown (2000) and by MacDonald P. Jackson in Who Wrote “The Night Before Christmas”? (2016).^{14, 20} The dates the family later gave for his reading it to his children range from before 1800 to 1808. Jackson's computational tests (pairs of speech sounds across word boundaries, the rates of common words, the rates of “a” and “the”, and each man's favourite expressions) placed the poem with Livingston's verse. Two later computational studies are reported to have found for Moore: one by Tom Jerman, a retired lawyer, in a 2023 book, using JGAAP, a standard open-source authorship-attribution program; the other by Shlomo Argamon, a computer scientist at the Illinois Institute of Technology, in tests commissioned by the Bloomberg columnist Justin Fox in 2023. I know both only through secondary accounts.^{23, 24}

Each document below exists as a text file with its source, provenance and verification status in the header (they are listed under Sources). seen means I checked the text against a scan or a period printing. transcription means I know it only through a second-hand transcription, mostly on Norsworthy's blog Melvilliana, which has published most of the documents on the Moore side, or by Mary Van Deusen, a Livingston descendant whose website henrylivingston.com holds the family's letters and Livingston's verse, and have not checked it against the original.

The dated record establishes the following six points. The Sentinel's editor did not know the author in December 1823; by his own account of 1836 he learned within months that the author was a college professor, and in January 1829 he described him in print as a New York City scholar and writer. Both statements were made while Livingston was alive or within a year of his death, and neither points toward Poughkeepsie. The poem was attributed to Moore in private by 1834 and in print from 1836, in a book to which Moore supplied it. Moore claimed it in his own words in 1844 and wrote out signed copies for the rest of his life. The poem was printed anonymously in Livingston's own town while he was alive, and no document records any reaction from him or from anyone in his house. No surviving dated document written before 1879 connects Livingston with the poem (Cornelia Griswold Goodrich, a descendant writing in 1900, mentioned family letters from about 1875 that have not been located, and said the family had no proof and no manuscript). And the family's own details conflict: the reading is dated 1780–1800, 1804–05 and 1808 by different relatives, the child who remembered is his son Sidney in one letter and his son Charles in another, and the Poughkeepsie newspaper is named as the Gazette, the Eagle or the Knickerbocker.

The documents leave two things unsettled. There is no original manuscript on either side: Moore's copies are fair copies (clean copies written out for friends and collectors) made thirty to forty years after the printing, and the Livingston manuscript is known only from descriptions written fifty to ninety years after its reported loss. And the account on Moore's side of how the poem travelled from his house to the Troy paper is itself hearsay, first written down in 1844 and 1862: Tuttle wrote in 1844 that Holley had received it from Mrs Sackett, the wife of a Troy merchant, and T. W. C. Moore, a kinsman of the poet, wrote in 1862 that a relative of Moore's had copied it into her album, from which a friend of hers from Troy made a copy. The silence of the Livingston household in 1828 is an argument from the absence of a record: no document says that the family noticed the printing, and none says that it did not. It carries weight, because the poem was printed in their own town, but it is not proof: the family may have known and left no trace.

The following sections ask whether the text of the poem can outweigh this record.

Every text was prepared for this report and carries a header with title, date, source, verse form and attribution status. The printed Moore poems were proofread against the scan of the 1844 book; the Livingston poems and Moore's manuscript poems were taken from the transcriptions on henrylivingston.com and have not been checked against the originals. Moore's corpus is twenty-nine of the thirty-three poems in Poems (1844), proofread against the scan (I left out the disputed poem, two verse translations and “The Mischievous Muse”, an adaptation from the Italian of Lorenzo Da Ponte, Mozart's librettist, who taught Italian in New York in Moore's day), plus seventeen manuscript poems including the 1,804-line Charles Elphinstone, in blank verse (unrhymed iambic lines), and the “Biography of the Heart” (leaving out a translation from the Italian poet Petrarch and a variant text of “Lines written after a snow-storm”, the second Moore poem the Sentinel printed anonymously, in February 1824). Livingston's corpus is the sixty-six poems the henrylivingston.com site gives as in his hand, signed with his name or his pseudonym “R” in the years when “R” is documented as his, or attributed by his descendants; poems attributed on style alone were left out. The controls are thirty-two poets who published in English between about 1780 and 1840, chosen to include the best-known writers of comic anapestic tetrameter (four anapests to the line) of the period: William Roscoe's The Butterfly's Ball and Catherine Ann Dorset's The Peacock at Home, the two most popular children's poems of 1807 in the metre; “Old Santeclaus”, an anonymous New York children's-book poem of 1821 that put St. Nicholas in a sleigh drawn by a reindeer; the Irish poet Thomas Moore (no relation); Byron; the English comic poets Thomas Hood, Winthrop Mackworth Praed and Thomas Love Peacock; James and Horace Smith's parodies Rejected Addresses (1812); the Americans Samuel Woodworth, Fitz-Greene Halleck, Hannah Flagg Gould and Philip Freneau; and others.

The table shows an imbalance that explains most of the results that follow: Moore left four anapestic poems, 1,797 words in all, while nearly half of Livingston's surviving verse is anapestic. A test trained on everything Moore wrote learns mainly his iambic and blank-verse habits; a test trained on everything Livingston wrote learns the habits of anapestic metre along with his own.

Before measuring the poem I tested every method on poems whose author is known. Each corpus was cut into 56-line samples, the length of the poem (long poems into windows, short poems bundled by author and verse form), which gave 132 known samples, 95 by Moore and 37 by Livingston. Each sample was held out in turn, the reference profiles (the average rate of each feature across an author's samples) were rebuilt without it, and the sample was attributed to whichever man's profile it was nearer, measured by cosine delta, a standard measure in authorship studies of the distance between two texts' profiles. Statisticians call this a leave-one-out test; I call the tests themselves profile tests below. I used six feature sets: the 100, 200 and 300 most frequent words; character 3-grams and 4-grams (every run of three or four consecutive characters, spaces included); and the frequencies of phonemes (speech sounds), taken from the Carnegie Mellon pronouncing dictionary, a free machine-readable list of American pronunciations. Two rows of the table use Burrows delta, the older form of the same distance measure.

The result that matters for this question is how the tests handle Moore's four anapestic poems, and the overall accuracies of 80 to 89 percent do not show it. With whole-corpus references, the word and character tests attribute those four poems to Livingston every time or all but once. The phoneme test gets them right, but it is the weakest test overall. Removing the features most sensitive to metre (identified from control poets who wrote in both metres) does not correct the misattribution. Building the reference profiles from the anapestic verse of both men only (metre-matched references, in what follows) does correct it: Moore's anapests are then attributed correctly on every feature set, though a quarter to nearly two-fifths of Livingston's anapests are then misattributed to Moore.

The effect is not peculiar to Moore. For 105 pairs of control poets, I tested one poet's anapestic and other samples against a reference built to look like Livingston's corpus (16 anapestic and 21 other samples of the second poet). The first poet's anapestic samples were misattributed to the second poet 29 percent of the time on words and 32 percent on character 4-grams; the same poets' other verse, 16 and 15 percent. Writing in anapests roughly doubles the chance that a poem is misattributed to a rival whose reference corpus is largely anapestic. Moore's case is worse than that average, probably because his own anapestic sample is so small. The tests that place the poem with Livingston also place Moore's own anapestic poems “The Pig and the Rooster” and “Irish Valentine” with Livingston, and on those tests the poem's score is close to the scores Moore's own anapests receive, so the result does not distinguish the two men.

Figures 1 to 4 show the same statistic for every known sample and for the poem. For each sample the margin is its distance to Livingston's profile minus its distance to Moore's, in cosine-delta units, so positive means nearer Moore. Each dot is one 56-line sample, the short black bar is the group mean, and the vertical line is the poem in its 1823 text (the text as revised in Moore's Poems of 1844 lands within 0.04 of it on every test).

Several things can be read from the table. With whole corpora the poem is Livingston-like on every feature set, which reproduces Jackson's result with different features and a different distance. But on the word and character sets it sits in or near the band occupied by Moore's own anapests (within their range on the 100-word and character 4-gram sets, beyond it on the 300-word set, short of it on the character 3-gram set), and a result that Moore's own anapests share is weak evidence of authorship. With metre-matched references the poem is Moore-like on every word and character set, within the range of Moore's four anapests and past 94 to 100 percent of the sixteen samples of Livingston's anapestic verse. The phoneme set is the exception both ways: with whole corpora the poem is far from Moore's anapests, and with matched references it is a tie. Phonemes are the feature set on which the poem is farthest from Moore's verse; only two of his 95 samples are more Livingston-like.

I rebuilt Jackson's two strongest tests on the new corpora. The first scores each text by the share of its words that fall on his Livingston-favoured list (I, his, my, her, on, as, is, was, at, thy, will, day, when, me, where, while) against his Moore-favoured list (to, from, your, for, they, be, with, this, our, not, which, so, would, it, heart, of, are, we), the lists of his chapter 16. The second is his phoneme-pair test: every word is transcribed into phonemes, the last phoneme of each word is paired with the first of the next within a line, and the ten pairs most favoured by each man are found by chi-square, a standard test of which pairs differ most between the two men's verse. I re-derived the pairs on the new corpora with the tested poem left out each time, so no poem contributes to its own test. The Livingston-favoured pairs that emerged (written in the dictionary's notation, in which DH is the “th” of “the” and AH the unstressed vowel of “a”: DH DH, AY L, AH N, AH F, N AH, AH P, AH L, Z IH, N DH, UW M) overlap Jackson's (AH N, AH F, AH L, AH P, N AH, Z IH). Jackson himself noted that the AH-initial pairs are “largely determined by rates of use of the definite and indefinite articles”, that is, of “the” and “a”.

Both tests reproduce on the new corpora. The poem scores in Livingston's range on both tests (51.6 on the word test and 64.6 on the phoneme test) and above all four of Moore's anapests on both (Moore's anapests score 34.2, 36.7, 41.9 and 45.9 on the word test and 16.7, 53.3, 55.6 and 59.4 on the phoneme test). Jackson's lists handle Moore's anapests better than the profile tests above do: cut midway between the two men's means, each test places two or three of the four with Moore, depending on whether the means are taken over all verse or anapestic verse only. The poem's score is qualified in two ways. The tests spread nine to twelve points on 56-line samples, so one poem's score carries little weight: the poem's word score is 12 points above Moore's anapestic mean and 6 above Livingston's, and the two men's anapests differ from each other by 5.5 points. And the phoneme test is shifted by metre: among the 14 control poets who wrote enough in both metres to compare, anapestic verse scores 5.5 points more “Livingston” than the same poet's other verse on the phoneme test, and 12 of the 14 shift that way. The word test is not raised by metre on the controls (-1.1 points; 11 of 16 poets score slightly less “Livingston” in their anapests), which agrees with Jackson's own check on Livingston's verse. So of Jackson's two tests only the word test survives this check, and its margin for Livingston is about the size of the spread of scores among 56-line samples by a single author.

Jackson listed six Moore “favourite expressions” absent from the poem. Norsworthy answered, in the same 2017 exchange, with eight Moore favourites present in it. I counted both lists in every corpus. The table gives rates per thousand words; the tail probabilities are the chance, under a Poisson model (which treats each occurrence of a word as an independent event at a fixed rate), of seeing at least the poem's count in 545 words at the named rate.

Taken as likelihood ratios (how many times more probable the poem's counts are under one man's rates than under the other's, with a small allowance so that a word one man never uses does not give an infinite ratio), Norsworthy's eight favour Moore by a log ratio of 14 (natural logarithms, so about a million to one on the model's own terms), of which “like” alone contributes 7, and Jackson's six favour Livingston by a log ratio of 2.8. These ratios should not be combined with any other figure: word rates vary far more from poem to poem than a Poisson model allows, and the count of “like” is a count of similes, which a comic description of a person's appearance naturally produces in quantity. The words found in only one man's corpus split three and three, and the three found only in Livingston's carry more weight each, because they are missing from a corpus nearly three times larger. The table shows that the favourite-expression evidence is mixed, rather than favouring Livingston as Jackson's six words alone suggested, and that its largest single item, the density of similes, matches the rate in Moore's anapests and has no parallel in Livingston's verse.

Metre

Line-level features come from syllable counts taken from the Carnegie Mellon dictionary. The poem opens 21 percent of its lines with “And”; Moore's anapests open 24 percent that way, Livingston's 12 percent, the anapestic controls 13 percent. Forty percent of the poem's lines have eleven syllables, which means the line starts da-DUM instead of da-da-DUM; Moore's anapests 28 percent, Livingston's 46, the controls 45. Feminine endings (lines that end on an unstressed syllable) are 18 percent in the poem and 16 percent in both men's anapests. “A” runs at 51 per thousand words in the poem, 37 in Moore's anapests, 17 in Livingston's; “the” at 62, 40 and 78; “I” at 18, 18 and 8. These figures favour Moore slightly, except the proportion of eleven-syllable lines, which favours Livingston slightly; the rate of “I” reflects the poem's first-person narration and should not be given weight.

Rhyme

Of the poem's twenty-eight rhyme pairs, four occur in Livingston's 987 couplets (care/there (×2), clatter/matter (×2), around/bound (×2), belly/jelly) and three in Moore's 2,615 (bow/snow, head/dread, sight/night). Livingston's rate is higher, four per thousand couplets against one, but the shared rhymes are ordinary ones that occur across the controls as well, and only one of Livingston's (clatter/matter) is in his anapestic verse. Rhyme does not distinguish between the two men.

Dutch names and the closing wish

Jackson made the Sentinel's “Dunder and Blixem” (the names of the last two reindeer in the 1823 text, from a Dutch oath meaning thunder and lightning) a Livingston marker, on the ground that the oath was current in the Dutch-descended society of Poughkeepsie in which Livingston lived. Google's Ngram Viewer, which counts phrases by year across the books Google has scanned, has “dunder and blixum” in print in 1803, 1812 and 1821 and in most years after 1823, and “donder and blitzen” from 1820, with one isolated hit dated 1780 (these are Google's counts; I did not check them against the books). The oath was current in printed English before 1823, and both men lived among the descendants of Dutch New York. The reindeer names do not favour either man. Moore's revision of the names to “Donder and Blitzen” in 1844, and the correction of “Blixem” to “Blitzen” on the illustrated sheet that Tuttle sent him in 1844 (Seth Kaller, a dealer in historical documents who wrote a defence of Moore's authorship in 2004, attributes the correction to Moore's hand), show that Moore treated the text as his own from 1844; they are not evidence that he wrote it. The closing “Happy Christmas”, cited on the Livingston side because Livingston wrote “happy Christmas” in a letter of 1773, was ordinary usage: Google Books has it in most years from 1782 onward.

The General Impostors method, a standard technique in computational authorship studies for checking whether a candidate can be verified as an author at all, asks whether the candidate is nearer to the poem than every one of twenty-nine other poets, rather than which of the two men is nearer. In each of 400 trials a random half of the features and a random half of the impostor samples are drawn, and the candidate wins the trial if the poem is nearer to one of his 56-line samples than to any impostor sample (using the min-max distance, the measure usual in this method, on relative frequencies). The score is the share of trials won. The impostors are the twenty-nine control poets with enough verse to yield a 56-line sample, plus the other candidate. Calibration on known samples shows what a score means here (Figures 5 and 6).

Neither man is verified as the author. For each man, the poem scores at or below the level that the other man's samples reach when tested as his. The poets who score highest when each control is tried as the candidate in turn are the comic anapestic poets of the period: Byron, Peacock, Hood and the Smiths on words; the Smiths, Byron and Robert Southey on character 4-grams; the American Nathaniel Parker Willis and Gould on phonemes. On a centroid test across all thirty-four poets (each poet's average profile is computed and the poem is ranked by its nearness to each), Moore ranks between 27th and 31st and Livingston between 15th and 28th, depending on the features. This is what one would expect when a 545-word comic narrative is measured against the light verse of its own decade: most of what these features measure in the poem is shared with other comic verse of the period. The measurable difference between Moore and Livingston within that genre is small, and most of it is attributable to metre.

The figure of 85 percent is a judgment; no calculation from the statistical results produced it. This section sets out how I reached it.

The documents alone put Moore at 85 to 90 percent for me. On one side are the editor's knowledge by 1824 (on his own 1836 account), a private attribution in 1834, a print attribution supplied by Moore in 1836, a public claim in 1844 and signed copies until his death; on the other, a family memory first written down in 1879, with no manuscript, contradictory dates, and no recorded reaction to an anonymous printing in the family's own town in 1828. The remainder goes to the family being right (several of Livingston's children are reported to have believed it) and to a third author (two other men, a young clergyman and an artist, were credited in Moore's lifetime, each by someone else after the man's death).

The stylometry has to be weighed in both directions. Whole-corpus profile tests call the poem Livingston's, but they also call Moore's own anapests Livingston's half to all of the time, so the likelihood ratio for Livingston from that finding is at most about two. Metre-matched tests call the poem Moore's; they get all four of Moore's anapests right and misattribute a quarter to two-fifths of Livingston's, so the likelihood ratio for Moore is roughly three, with wide uncertainty from a base of four poems. Jackson's word and phoneme tests place the poem above all four of Moore's anapests; on the spread of Moore's samples as a whole that is a likelihood ratio for Livingston of about two on each test (larger if only his four anapests set the spread), and the phoneme figure shrinks once the metre shift measured on the controls is taken off. The simile density favours Moore by a margin I would not trust as a number, and the other content words split. The impostors test shows that all of these signals are weak, and taken together they are close to even.

So the stylometry neither raises nor lowers the documentary figure by much. I take the lower end of the documentary range, Moore about 85 percent, Livingston about 12, someone else about 3, because the stylometric designs disagree with one another and rest on four Moore poems, and because Jackson's word test still points the other way. The stylometry adds only that the internal evidence does not contradict the documents.

Several of Livingston's children, five by the family's own count in 1900, are reported to have remembered him reading the poem to them, and a daughter said so in her own hand in a letter dated to 1879, twenty years before the claim was made in public. Livingston habitually wrote comic verse in anapestic tetrameter and Moore rarely did; Livingston's verse is playful, Moore's mostly grave, and the poem reads unlike most of what Moore published. Jackson's tests, run on the largest corpora available in 2016, placed the poem with Livingston on each of his main measures, and his own concession that the external evidence “taken at face value, favours Moore” leaves room for the poem to have reached Moore's house by a copy and been claimed in good faith twenty years later, when it was already circulating under his name. This is the case at its strongest; it has substance, and without the documentary record I would accept the family's account.

Each of these points can be answered. The family memory was not written down until 1879; the written versions contradict one another on the date, the child and the newspaper; and the family said in 1900 that it had no proof. Moore did write anapestic comic verse (four poems of it survive), and the poem lies among them on metre-matched tests. The profile tests that place the poem with Livingston also place Moore's anapests with Livingston, and on Jackson's own lists the poem exceeds Moore's anapests by about the spread of scores among samples by a single author. And if Holley's 1836 statement is accurate, the poem was being called Moore's in 1824, within a year of the printing. The family's account of 1900 requires the opposite: that a copy carried into the Moore house by a governess, with no author's name on it, came to be taken for Moore's own only decades later.

Any of the following would overturn my verdict: a Poughkeepsie newspaper printing of the poem from before 1823, which the family said existed and which searches since the claim was first published in 1919 have not found; a dated document from before 1844 in which any Livingston mentions the poem; a manuscript of the poem in Livingston's hand; or an attribution method that correctly assigns at least nine in ten of both men's anapestic poems on a leave-one-out test and still places the poem with Livingston. A stylometric result that fails that last check should not be given weight. A systematic search of surviving Dutchess County newspapers (the county around Poughkeepsie) for 1800 to 1822 is the most likely way to settle the question.

This case adds no new documents; everything new in it is analysis of texts already known. Reference profiles built from the anapestic verse of both men only, validated on known poems and then applied to the poem, are the comparison Norsworthy asked Jackson for in their exchange of 2017 on Melvilliana. Jackson had reported his tests' scores for Moore's four anapests as a control, but nobody had built and validated a metre-matched design, and it places the poem with Moore. The leave-one-out validation shows that standard profile methods trained on whole corpora misattribute Moore's own anapests, and the control-corpus simulation shows the mechanism is general and measures its size. The impostors test against twenty-nine period poets is, as far as I found, new to this question, and it shows that neither candidate can be verified. Jackson's word and phoneme tests have been replicated on independently built corpora, which I did not find done before, and the metre shift of his phoneme test has been measured on poets who wrote in both metres. And the favourite-expression evidence cited by each side has been counted by the same method on the same corpora, so the two lists can be compared directly.

The documentary picture here agrees with Norsworthy's, from his discovery of Moore's public letter of 1844 and of Holley's 1836 statement to the pre-1844 attributions and his identification of T. W. C. Moore, the kinsman who sent Moore's autograph copy to the New-York Historical Society in 1862, as a first cousin once removed of Livingston, and, before him, with Kaller's and with that of Stephen Nissenbaum, a historian of the American Christmas who answered Foster in 2001. I re-read Moore's 1844 letter from the scan (the first date is smudged). The other documents I checked against scans (the 1823 printing, Hoffman's anthology issued in 1836, Moore's Poems of 1844, Arthur J. Weise's history of Troy of 1891, which quotes the Sentinel's 1829 note on the author, T. W. C. Moore's letter of 1862, and William S. Thomas's 1919 article on Livingston) read as Norsworthy and Kaller describe them; the rest I read only in transcription. Jackson's own results reproduce. Jerman's and Argamon's reported Moore-ward results are consistent with the metre-matched finding here, though I replicated neither. Foster's “all” marker (his claim that the poem's use of “all” as an intensifier, as in “all snug” and “dressed all in fur”, was a Livingston habit) fails on these corpora: the two men use the word at the same rate.

Moore's anapestic sample is four poems and 1,797 words; every metre-matched result rests on it, and a percentile computed among four samples is imprecise. Livingston's corpus comes from second-hand transcriptions on his descendant's website, and it includes the Carrier's Address of 1803 (a New Year's verse handed out by a newspaper's carriers), which Norsworthy thinks is probably by Isaac Mitchell, a Poughkeepsie newspaper editor; I did not rerun the tests with it removed. The phoneme features use the Carnegie Mellon dictionary's modern American pronunciations and skip words it does not know. The impostors test could not verify either man, so every stylometric result here rests on weak signals. I examined no manuscript, and I read most of the documents in transcription rather than against the originals (the table in The paper trail marks which); Holley's 1836 statement, on which the earliest dating of Moore's authorship rests, is one of those, and I read Jackson's book in excerpt and through page-referenced quotation; I did not read it in full. I did not re-run Jerman's JGAAP protocol or Argamon's method (repeated comparisons on random subsets, as Fox describes it), and I took the Dutch-names question only as far as Google Books.

Stylometry separates two authors reliably when both are writing the same kind of text. When the two candidates' corpora are in different forms, nothing in the method reports that difference, and it may measure the form instead of the author. Jackson's tests were carefully designed, and they still responded to metre before authorship, because nearly all of Moore's surviving verse is iambic and much of Livingston's is anapestic. The same problem arises in any authorship dispute in which one candidate habitually wrote in the disputed form and the other did not, which is a common situation. The check is to run the method on poems whose authors are known, in the same metre and at the same length, and to record how often it is wrong. In this case the statistical results are too weak to decide the question, and the verdict rests on the documentary record from Troy and Poughkeepsie between 1823 and 1879. It would be overturned by a Poughkeepsie newspaper from before 1823 containing the poem.