# Who wrote "A Visit from St. Nicholas"?

Published page: https://historical-mysteries.netlify.app/night-before-christmas/report

In one line: Clement Clarke Moore wrote it, with my confidence at about 85 percent (Henry Livingston Jr. 12, someone else 3); the evidence label is Probable; the verdict agrees with the traditional attribution.

## The mystery

Two days before Christmas 1823 the Troy Sentinel, a newspaper in Troy, New York, printed fifty-six lines of verse headed "Account of a Visit from St. Nicholas", with a note from its editor, Orville L. Holley, saying he did not know who wrote them. The poem (it begins "'Twas the night before Christmas") gave St. Nicholas his reindeer, his sleigh and his descent by the chimney, and it circulated for thirteen years without an author's name. In December 1836 a New York anthology printed it as the work of Clement Clarke Moore, a professor of Oriental and Greek literature at the General Theological Seminary in Manhattan. In 1844 Moore said in print that he had written it years before to amuse his children, and he put it in his collected Poems. The Poughkeepsie Journal had printed the poem anonymously in January 1828, six weeks before Henry Livingston Jr., a local landowner, surveyor and writer of light verse in the poem's metre, died there; nobody in his house is recorded as saying anything about it. From 1879 his descendants wrote, and from 1899 said in public, that he had read the poem to his children years before 1823, that his manuscript had burned in a house fire in Wisconsin, and that a Poughkeepsie paper had printed it before Troy did. Neither the manuscript nor that printing has been found.

## Why it is a mystery

For Moore there is a sequence of dated documents. The Sentinel's editor wrote in January 1829 that the author was a New York City scholar, and recalled in 1836 (a statement I know only in transcription) that he had learned within months of the printing that the author was "a professor in one of our colleges". A student's diary of 1834 describes a costume made "according to the description of Prof Moore in his poem". Moore supplied it to the 1836 anthology, claimed it in 1844 and wrote out signed copies until his death in 1863. On the other side, he said nothing for twenty years; no manuscript from before 1823 survives for either man; and most of his other verse is grave and iambic, whereas the poem is comic and anapestic, a metre he used in only four other surviving poems. For Livingston there are two things. The family's memory was first written down in 1879, fifty-six years after the printing; the relatives who recorded it disagree on the date, the child and the newspaper; and, as the family said in 1900, no document from his lifetime supports it. The poem's manner matches the comic anapestic verse he habitually wrote. Don Foster, an English professor at Vassar College, argued the case for Livingston in his book Author Unknown (2000). MacDonald P. Jackson, a New Zealand scholar of Shakespearean authorship, tested the claim statistically in 2016: on his counts of common words and speech sounds the poem scored with Livingston. 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.

## How I judged it

I collected every dated document from 1823 to 1920 and marked which I checked against a scan and which I know only from a transcription. Then I rebuilt the stylometric comparison (stylometry counts a writer's habits: which common words, letter sequences and speech sounds appear, and how often) on corpora built for this report: 46 poems by Moore (the 29 in his 1844 Poems proofread against the scan, and 17 manuscript poems taken from the transcriptions on henrylivingston.com, a website kept by a Livingston descendant), 66 by Livingston (also from that site's transcriptions, not checked against originals), and 2,756 by 32 other poets of the period. Every method was tested first on poems whose author is known. Standard tests trained on everything each man wrote sort those known poems 80 to 89 percent correctly, but the word and letter-sequence tests attribute Moore's own four comic anapestic poems to Livingston every time or all but once, because the tests learn the metre along with the author. Tests built from the anapestic verse of both men only sort Moore's four correctly and put the poem with Moore: inside the range of his four and beyond 94 to 100 percent of the sixteen 56-line samples of Livingston's anapestic verse on every word and letter-sequence feature set (speech sounds are a tie, and four Moore poems are a thin base). Jackson's two strongest tests reproduce on the new corpora, with the poem above all four of Moore's anapests, by margins about the size of the spread of scores among 56-line samples by a single author. Against twenty-nine other poets of the period, neither man can be verified as the author at all.

## Verdict

Clement Clarke Moore wrote it. Moore is the probable author: the evidence clearly favours him over Livingston but does not prove it, and the percentage moves when the same evidence is weighed differently. My confidence is about 85 percent: Moore 85, Henry Livingston Jr. 12, someone else 3. Those figures are a judgment, conditional on the weights set out in the report's section "How I got to 85"; no calculation from the numbers produced them. The stylometry is close to even and weak in both directions, so the verdict rests on the documents: nothing written in Livingston's lifetime, or for fifty years after it, connects him with the poem, and the editor who printed it placed the author in New York City eleven months after Livingston died.

## How this stands to earlier views

The traditional attribution is to Moore. Livingston's descendants have disputed it since 1899, Don Foster since 2000 and Jackson since 2016; Jackson's position in 2017 was that the external evidence favours Moore, the internal evidence favours Livingston, and the question may never be settled. My verdict agrees with the traditional attribution. What it changes is the reading of Jackson's numbers: his Livingston-favouring results are real, but they are close to what any anapestic poem by Moore produces on tests trained on his mostly iambic output.

## What is new here

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 2017 exchange on Melvilliana. Jackson reported his tests' scores for Moore's four anapests as a control; nobody had built and validated a metre-matched design. It places the poem with Moore on every word and character feature set, on a base of four Moore poems.

The leave-one-out validation of standard profile methods (each known sample held out in turn and attributed by the rest, using the most frequent words or runs of consecutive characters) is new to this question. Trained on whole corpora they misattribute Moore's own anapests (0 to 2 of 4 correct). Jackson's own lists do better (2 or 3 of 4 with Moore, depending on the cut), but the poem lies beyond all four on both of his tests. A simulation on 105 pairs of control poets shows the mechanism is general: a poet's anapestic samples are misattributed to a Livingston-like reference about twice as often as the same poet's other verse (29 against 16 percent on words, 32 against 15 percent on character 4-grams).

A General Impostors verification (a standard check of whether a candidate can be verified as an author at all, against many other writers) using 29 period poets is, as far as I found, new to this question. Neither candidate is verified, and the poem's nearest neighbours are the comic anapestic poets of the period (Byron, Peacock, Hood, the Smiths).

Jackson's word-list and phoneme-pair tests are replicated on independently built corpora, which I did not find done before. Both reproduce (poem 51.6 on the word test against Moore 33.7 and Livingston 49.5; 64.6 on the phoneme test against 49.9 and 61.8). On 14 control poets who wrote in both metres, anapestic verse scores 5.5 points more "Livingston" on the phoneme test; the word test is not shifted by metre.

Norsworthy's eight Moore expressions and Jackson's six are put on one likelihood scale, using the chance under a Poisson model of a count at least as high as the poem's. "Like" occurs 8 times in the poem's 545 words (14.7 per 1,000) against 8.4 per 1,000 in Moore's anapests and 1.2 in Livingston's. Norsworthy's eight favour Moore by a log likelihood ratio (natural logarithm) of 14, driven by "like"; Jackson's six favour Livingston by 2.8; words found in only one man's corpus split three and three. Taken together, the expression evidence is mixed and does not favour Livingston.

What this agrees with: Norsworthy's documentary discoveries (the 1844 New York American letter, which I re-read from the scan, where the first date is smudged; Holley's 1836 statement, read only in his transcription; the pre-1844 attributions; T. W. C. Moore as Livingston's first cousin once removed), and Seth Kaller, a dealer in historical documents (2004), and the historian Stephen Nissenbaum (2001) on the record. Foster's "all" marker (his claim that the poem's use of "all" as an intensifier was a Livingston habit) fails on these corpora (the two men's rates are equal). Tom Jerman, a retired lawyer (2023 book), and Shlomo Argamon, a computer scientist at the Illinois Institute of Technology (2023 tests for Bloomberg), known to me only through secondary accounts, are reported to point the same way as the metre-matched result; I replicated neither. Jackson's own numbers reproduce.

## Where this falls short

Moore's anapestic sample is four poems and 1,797 words, and every metre-matched result rests on it. Livingston's corpus comes from second-hand transcriptions on henrylivingston.com and includes the 1803 Carrier's Address (a newspaper's New Year's poem), which Norsworthy thinks is probably by Isaac Mitchell, a Poughkeepsie newspaper editor; I did not run a purged corpus. The speech-sound features use the Carnegie Mellon pronouncing dictionary's modern American pronunciations and skip unknown words. The impostors test could not verify either man, so every stylometric statement concerns weak signals. I examined no manuscript and read most documents in transcription (flagged in docs/documents.md), including Holley's 1836 statement on which the earliest dating of Moore's authorship rests; 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 protocol in JGAAP (an open-source authorship-attribution program) or Argamon's random-subset method, and I took the Dutch-forms question only as far as Google Books Ngrams. The confidence figures are a reasoned judgment; no calculation from the numbers produced them.

## Files

- `report.html`: copy of the published report (self-contained; figures are inlined as SVG).
- `figures/fig_margins_mfw300_both.svg`, `figures/fig_margins_mfw300_anapestic.svg`, `figures/fig_margins_char4_both.svg`, `figures/fig_margins_char4_anapestic.svg`, `figures/fig_margins_char3_both.svg`, `figures/fig_margins_char3_anapestic.svg`, `figures/fig_gi_mfw200.svg`, `figures/fig_gi_char4.svg`: standalone copies of the figures, all made for this report (the report uses the first four margin figures and the two impostors figures, numbered 1 to 6 in order of appearance). The six margin figures, including Figures 1 to 4, are drawn from `data/results/margins_cosine.json`, which `code/calibrated_margins.py` produces; Figures 5 and 6 are drawn from `data/results/gi_calibration.json`, which `code/analysis.py` produces; `code/figures.py` draws all of them.
- `docs/documents.md`: dated documentary timeline with verification flags; `docs/scholarship.md`: annotated bibliography and claims register; `docs/sources/`: about 90 source texts and 3 scans, each headed with URL, provenance and verification status.
- `corpus/moore/`, `corpus/moore_mss/`, `corpus/livingston/`, `corpus/control/`, `corpus/visit/`: the texts, each with a MANIFEST.md; `corpus/visit/visit_collation.md` collates the 1823, 1830, 1837, 1840 and 1844 witnesses.
- `code/`: `stylo.py`, `phon.py`, `meter.py`, `analysis.py`, `replicate.py`, `run_replicate.py`, `calibrated_margins.py`, `confound_sim.py`, `metre_shift_phone.py`, `figures.py`, and `code/report/` (report source and builder).
- `data/results/`: every number in the report as JSON with logs (`validate.json`, `margins_cosine.json`, `gi_calibration.json`, `attribute_1823.json`, `attribute_1844.json`, `replicate_full.json`, `markers.json`, `rhymes.json`, `metre_shift.json`, `confound_sim.json`) and the figure SVGs; `data/ngrams/google_ngrams_1780_1860.csv`.

To regenerate: `python3 code/analysis.py`, `python3 code/run_replicate.py full`, `python3 code/calibrated_margins.py`, `python3 code/confound_sim.py`, `python3 code/metre_shift_phone.py`, `python3 code/figures.py`, then `python3 code/report/build.py`.