The Odds Against the Earl

In 2004 Ward Elliott and Robert Valenza, advisers to the Claremont Shakespeare Clinic, put the odds against Shakespeare producing the Earl of Oxford’s poems by chance at 400,000 times worse than for his own least typical poem block, and 150 trillion times worse by a second measure. This page reproduces both from the Clinic’s own published test scores; the second comes out far stronger than printed, 8.8 × 10²¹. Then it runs a test proposed in 2009 and not run by either side, Shakespeare’s earliest plays scored against ranges from his later ones. As its proposers predicted, the earliest blocks drift outside those ranges (17 of 23 fall outside at least one), but none reaches Oxford’s count of mismatches, a gap the test is shown able to catch in 21 of 23 planted copies. The second measure gives his own play verse, set against his poems (a shift of genre and of date at once), odds up to 6.4 × 10²², and his 96 baseline blocks can certify only about 10.2 to 1 of the 398,000. Whether the odds measure authorship remains open; who wrote Shakespeare is not tested here.

Everything below is computed in your browser from two files this page transcribed from the claimants’ own appendix: the test scores the Clinic printed for every block of verse it measured. Numbers the claimants printed are marked as printed and cited to their page; every other number is the page’s own computation from those scores, and says so. Nothing here re-measures a line of verse. The check at the bottom recomputes the page against itself while you read.

I · the claim, at full strength

Four hundred thousand to one

The idea that Edward de Vere, seventeenth Earl of Oxford (1550 to 1604), wrote the plays and poems of Shakespeare was first argued at length by J. Thomas Looney in 1920. This page does not test it. It tests a number put against it: the odds computed by the Claremont Shakespeare Clinic, a project in which students at Claremont McKenna College, advised by Ward E. Y. Elliott, professor of American political institutions, and Robert J. Valenza, professor of mathematics and the humanities, built computer tests of Shakespeare’s style from 1987 onward. Their 2004 paper in the Tennessee Law Review says in its opening paragraph:

Our internal-evidence stylometric tests provide no support for Oxford. In terms of quantifiable stylistic attributes, Oxford’s verse and Shakespeare’s verse are light years apart. The odds that either could have written the other’s work are much lower than the odds of getting hit by lightning.

Elliott and Valenza, “Oxford by the Numbers”, Tennessee Law Review 72 (2004), p. 323

And they state the headline in numbers:

His poems now have seven Shakespeare rejections in fifteen tests, far too many to look like Shakespeare to us or to our computer, which calculates the Discrete odds against so many rejections arising by chance from Shakespeare as 400,000 times worse than those for his own most discrepant block. The Continuous composite odds are 150 trillion times worse.

Elliott and Valenza (2004), p. 376

What was measured is sixteen short poems, 3,042 words, that the scholar Steven May assigned conclusively to Oxford in 1980. Elliott and Valenza set two specimens side by side in a later paper and asked whether they sound like one person (“We think not”):

Help man, help beasts, help birds and worms, that on the earth do toil;

Help fish, help fowl, that flocks and feeds upon the salt sea soil,

Oxford, “Fram’d in the Front of forlorn hope”, published 1576, as printed by Elliott and Valenza, The Oxfordian XII (2010), citing May (1980), pp. 27 to 28

O God, Horatio, what a wounded name,

Things standing thus unknown, shall I leave behind me!

Hamlet, V.ii, as printed in the same paper

How the odds are made

Each test is a Shakespeare profile: a range his own 3,000-word blocks of verse fall inside, such as grade level 10 to 14, or 7 to 25 per cent of lines with feminine endings. A sample outside the range gets a “rejection”. The discrete composite counts rejections and asks how likely that many would be by chance, “assuming that rejections occur randomly at a fixed rate” of 2.6 per cent per test (p. 349). The continuous composite measures how many standard deviations the sample sits from Shakespeare’s mean on every test, adds the squares, and turns the total into a probability (p. 350). The claimants describe both in words and print no formula; the formulas this page uses (a binomial tail, and a chi-square tail with one degree of freedom per test) are a reconstruction that reproduces their printed values, shown in the check below. Both odds are then ratios: Oxford’s composite against the composite of Shakespeare’s own most discrepant poem block, the first 3,000 words of Venus and Adonis.

The claim is stronger than printed

The discrete odds reproduce: 6 of 14 rejections (grade level, feminine endings, enclitics, proclitics, BoB7 and modal distance) at 2.6 per cent give 7.749E-07, exactly as printed on p. 424, and the ratio to Venus 1 is 3.98 × 10⁵, which the claimants round to 400,000. The continuous odds do not merely reproduce; they grow. Oxford’s continuous composite error, recomputed from his fourteen printed scores and the printed Shakespeare means and standard deviations, is 11.962 against the printed 11.9588 (a difference of 0.03%). The probability that error implies is 1.7 × 10⁻²³ (1.8 × 10⁻²³ from the printed error), which the appendix prints as “<1.0000E-15”: a floor below which the tables print no number (in 2010 the claimants describe such odds as “too low to compute”), not a value. The printed “150 trillion” is exactly the Venus block’s 1.50E-01 (1.4955E-01 in the appendix) divided by that floor (1.5E+14). Unfloored, the formula as reconstructed here gives 8.8 × 10²¹, about 58.6 million times the printed figure. Half of the squared distance is one test: on the printed p. 422 means and standard deviations, Oxford’s modal distance sits 8.5 standard deviations above Shakespeare’s mean (the claimants’ own prose says “almost eight”, p. 374), and his grade level −4.4 (they write “five standard deviations below”, p. 347). Leave modal distance out on both sides, Oxford and the Venus block alike, and the continuous odds fall from 8.8 × 10²¹ to 2.3 × 10⁸.

Three details of the printing, reported rather than fixed. Table 3.1 counts 7 rejections in fifteen tests, but its 7.75E-07 is the value for 6 of 14: the seventh rejection is relative clauses, the one test Appendix Six leaves out because it is counted by hand (p. 370, note 99). The table’s own note 100 says its odds differ from the appendix’s “because we use one more test here” (p. 371), yet its printed composites are the fourteen-test values; that relative clauses stayed out of them is the page’s inference from that arithmetic. Counted in, as the table’s own words invite, the claim grows again: 7 of 15 gives 4.300E-08 against 3.264E-01, 7.59 million to 1. The table’s caption says the odds are “between 400,000 to 1.5 quadrillion times worse”; nothing in the table yields 1.5 quadrillion. And under the continuous method Shakespeare’s most discrepant poem block is not Venus 1 but Venus 2 (printed 8.9960E-02, the p. 422 threshold, which the p. 426 key misprints as 8.9660E-02); against its recomputed 0.088 the page’s continuous odds are 5.1 × 10²¹.

The claim deserves its record, too. The same tests reject verse by everyone else the Clinic measured: of the 87 3,000-word poem blocks by other poets, Oxford’s among them, only 3 have fewer than two rejections, and 34% of their individual tests fail, against 2 of 196 for Shakespeare’s own poem blocks and 35 of 1,150 for his play verse. A Funeral Elegy, which Donald Foster had ascribed to Shakespeare and which the Clinic scored at 6 of 14, the same count as Oxford, is the call the paper itself singles out; it records that “In 2002 Donald Foster conceded that our ascription of A Funeral Elegy was right and his was wrong” (p. 377, note 127).

II · the deciding control

Shakespeare against himself

The odds rest on one comparison: a sample whose iambic-pentameter poems the claimants date 1576 to 1593 (pp. 374 to 375), and whose whole they label 1572 to 1594 (pp. 384 to 386), against Shakespeare’s poems, “written between 1593 and 1609 by conventional dating” (p. 374). The critics’ objection is that styles change with time and genre, so a mismatch may measure a young writer, or a song, rather than a different writer. The claimants grant part of it. “Three of Oxford’s rejections could be time-sensitive,” they write (p. 376), and of their continuous method: “Continuous does not account for time periods for traits like line endings, where Shakespeare’s style changed over the years. Discrete analysis distinguishes between early and late profiles; Continuous does not.” (p. 350)

In 2009, in The Oxfordian, the journal of an Oxfordian society, John M. Shahan and Richard F. Whalen proposed a test of exactly that:

Suppose Elliott and Valenza split Shakespeare’s plays into two groups — the latest three-fourths, and the earliest one-fourth — and computed expected ranges for stylistic measures from the later plays. Would the earlier plays fall within the expected ranges derived from the later ones? It is unlikely that all of them would.

Shahan and Whalen, “Auditing the Stylometricians”, The Oxfordian XI (2009), p. 259, point 7

Neither side ran it, and our searches found no one who has (the searches are listed below). The claimants’ 2010 reply in the next volume does not take it up (a search of its text for “split”, “one-fourth” and “three-fourths” finds nothing). Their standing answer to the time objection is older: “Many of Shakespeare’s measurable patterns did not change at all during his known productive lifetime. Why should we suppose that these constants of his maturity must have changed drastically in his youth?” (“Of Grubs and Butterflies”, 2000). The appendix makes the test possible, because it prints the scores of 82 dated 3,000-word blocks of Shakespeare’s play verse, from Richard III (1593) to The Tempest and The Winter’s Tale (1611). Below, you run it.

How the time arm builds its ranges: each test’s expected range is the later blocks’ minimum to maximum, and the means and standard deviations are theirs. The range rule is the page’s choice, not the claimants’ profile construction: each Clinic profile is set so that at least 95 per cent of Shakespeare blocks pass (p. 348), which makes the printed play-verse profile narrower than its blocks’ extremes on several tests (grade level 3 to 8, where the 82 blocks run from 2 to 10, p. 428), so a minimum-to-maximum range is the most lenient one the later blocks allow. For 59 later blocks, a minimum-to-maximum range leaves out a new block drawn from the same distribution about 2 times in 60, some 3 per cent per test, close to the Clinic’s own 2.6. Every earlier block then goes through the unmodified scorer.

At 1596, which puts Richard III, Richard II and Romeo and Juliet in the earlier group (three of fourteen plays, 23 of 82 blocks, the nearest the dated blocks come to an earliest quarter), the earlier blocks collect 6, 8, 5, 2 and 2 blocks with 0 to 4 rejections. The rejections fall mostly on feminine endings (9) and open lines (9), the two tests the claimants themselves give separate early and late ranges.

On the proposers’ own measure, their prediction held. 17 of 23 earlier blocks fall outside at least one range built from the later plays, in every one of the three plays, and 9 carry two or more rejections, which the Clinic’s own play-verse profile puts outside Shakespeare (“0 to 1” rejections, p. 421; its own baseline has 4 of 82 such blocks, which it counts as composite rejections, p. 348, note 57). For a yardstick, score each later play the same way, against ranges from the other later plays only: 21 of 59 of those blocks fall outside at least one range (36%), and 9 carry two or more (15%). The earliest plays drift, as the critics said they would, and by more than the later plays drift among themselves. None reaches Oxford’s count. On the continuous composite the earliest plays travel further: Romeo 4 scores 3.9 × 10⁻¹², which against the Venus block is 3.9 × 10¹⁰ to 1 for verse Shakespeare certainly wrote, though still short of Oxford’s. On the discrete composite the worst early block (Romeo 4, 4 of 15) is 622 to 1.

The second arm: across the genre line

The critics’ first objection was genre: eight of Oxford’s sixteen poems first appeared in The Paradyse of Daynty Deuises (1576), which the critics call an anthology of songs, and the claimants concede that “Any or all of them could be song lyrics, not poems proper, and, hence, not suitable for comparison with poems” (p. 392). Shahan and Whalen asked for Oxford’s songs to be compared with “the fairly large collection of known Shakespeare songs” (2006). That cannot be run from the published tables, and in 2007 the claimants recorded that they had declined to do it themselves: “When Nina Green asked us to do it, we declined, considering it too narrow, too Oxford-centric, too subjective and inconclusive, and too much work for us to replicate for Shakespeare’s songs the process that took our students six years to work out for his poems and plays. But we reminded her of our still-standing offer to let her do it, if she wished, using our software.” What the tables do allow is one genre boundary the Clinic itself flags: Shakespeare’s own play verse, scored against his own poem profile, on the 13 tests both tables share. (The modal test was not measured for play verse, and the genre arm refuses to impute it.)

Against his poems, 81 of his 82 play-verse blocks fail grade level, which the claimants flag as genre-sensitive; 60 fail nothing else, and none collects more than 4. Oxford on the same 13 tests scores 5 of 13; against the Venus block on its fourteen tests (the arm’s common scale throughout) that is 24,000 to 1 on the discrete composite, where the worst genuine play block is 1,140 to 1. The continuous composite behaves differently: 46 of 82 genuine blocks score at or beyond Oxford’s 5.7 × 10⁻¹⁰ on the same tests, and the median block scores 6.2 × 10⁻¹¹. Tempest 1 reaches 2.4 × 10⁻²⁴: 6.4 × 10²² to 1 against the Venus block, where Oxford on the same thirteen tests is 2.6 × 10⁸ to 1 and on all fourteen of his tests 8.8 × 10²¹ (its open lines alone sit 8.2 standard deviations above his poems’ mean; open lines are a test the claimants flag as time-sensitive, and the continuous composite does not adjust for date, so this arm measures genre and date together). 2 genuine blocks score beyond Oxford’s full fourteen-test figure with only thirteen tests.

What this control can and cannot do. It measures how far the Clinic’s own machinery moves when text Shakespeare certainly wrote is shifted by a few years or across the play and poem boundary. It reaches back only to 1593, because the Clinic’s play baseline begins there, while Oxford’s poems were published from 1573 and, on the critics’ reading, some were written in the 1560s. Play verse is not song. A genuine block scored far out says the formula is not calibrated across that shift; it says nothing about who wrote anything.

III · the control on the control

Could the split have caught an Oxford?

A control that finds nothing is worth something only if it could have found something. So the page plants the claimed effect, at the claimed size, into copies of the control’s own blocks: on every test a block shares with Oxford’s row, it adds Oxford’s departure from Shakespeare’s poem mean (grade level −4.79, hyphenated compounds −80.09, feminine endings −12.00, enclitics −41.07, proclitics −214.29, BoB7 +245.07, BoB8 +211.14, and so on for each shared test), read from the frozen Oxford row and the frozen printed means. Values are clamped only at bounds the tables themselves show (no negative counts, no percentage above 100, BoB7 no higher than 1,000), and every clamp is logged. Then the unmodified scorer runs on the doctored copies, with the ranges and moments built from the real record, which the plant cannot touch. The decision rule, set in the page’s specification (which had already seen a pilot of these fractions) and not changed since: an arm could have confirmed the claim on a composite if at least 80% of planted blocks score beyond every unplanted block of that arm, 80 per cent being the conventional power target in study design.

The negative control is a departure Shakespeare himself shows: each of his 14 poem blocks’ own distance from his poem mean, planted the same way through the same scorer. None meets the line: at most 6 of 23 and 4 of 23 in the time arm, 18 of 82 and 25 of 82 in the genre arm (the largest share 30%). The rule does not fire on differences inside Shakespeare’s own poems.

Grade A, by injection, and the answer is split. On the discrete composite, the 2009 test had the power to see an Oxford-sized departure in Shakespeare’s earliest play verse and did not see one. It found the drift its proposers predicted, and a drift of Oxford’s size is not what it found: on the rejection count, that is as far as the claim survives the control its critics proposed, at the split nearest their “earliest one-fourth”. Scaled down, the planted departure still meets the line at 0.75 of Oxford’s size (20 of 23) and falls under it at 0.7 (16 of 23). The result holds at 1595 and 1596 only: at later splits, which take in more plays and leave fewer to build the ranges from, the planted departure is found less often (down to 67%), so for 1597, 1601, 1602 and 1603 the page calls the discrete result inconclusive too (the sweep below). On the continuous composite the same test could not have confirmed the claim (74%, under the line), because genuine early blocks already reach 3.9 × 10⁻¹²; its real result is inconclusive. Across the genre line neither composite could have confirmed it. The planted departure is added to play-verse scores although it was measured against poem means, and a proportional plant moves the fractions; switch it above. Read the proportional genre cell with care: scaling pulls play-verse open lines, which run far above the poems’, toward the poem mean, so 12 of 82 planted copies end less extreme than their own unplanted copy on the continuous composite, a limit of that plant form rather than a verdict on the arm.

IV · the claimants’ method on nothing

What 96 blocks can certify

Both composites turn a count or a distance into odds by treating the fourteen tests as independent draws. The claimants assume it for the calculation (“Assuming that rejections occur randomly at a fixed rate, one can calculate the precise rejection odds”, p. 349) without asserting that it is true; of the scores behind their earlier standard errors they wrote in 1991: “We make no claim that our distributions of tested scores are normal (they are not), or that the individual observations are statistically independent.” Their own baseline can test what the independence model extrapolates. Shakespeare’s 96 baseline blocks, 14 of poems and 82 of play verse, are text known to contain no authorship difference, scored by the claimants themselves.

The observed counts, 66, 26, 2, 1 and 1 blocks with 0 to 4 rejections, fit the model where the model is checkable: 30 of 96 blocks have at least one rejection, against 29.6 expected. In the tail they run heavier than it: more blocks with three or four rejections than the model expects, which is small numbers, so the page says only that. The claimants saw it too: “Although we rejected slightly fewer blocks than expected, two blocks had an unexpectedly high number of rejections, three and four” (p. 348, note 57).

Where the claim lives, the model cannot be checked at all. No Shakespeare block has six rejections, so the only honest statement about how often Shakespeare produces six is a bound: none in 96 means at most 3.07% at 95 per cent confidence (about 1 in 33). The boundary block’s one rejection is shared by 31.3% of the baseline. So the most the Clinic’s own baseline can certify for a six-rejection sample is odds of about 10.2 to 1, where the model asserts 398,000. That figure takes the boundary block’s share as observed; bound it too (at least 23.5% at 95 per cent) and it is 7.6. For the continuous composite, none of the 96 is as extreme as Oxford and 16 of 96 are at least as extreme as the Venus block: about 5.4 to 1, or 3.5 with both shares bounded. This bounds what the data can show. It does not show the odds are small: 96 blocks cannot certify a chance smaller than about 1 in 33, whatever the truth is. Everything past that comes from the independence assumption.

V · the second layer

Shakespeare’s exchange rate

Put the arms together and the Clinic’s machinery has an exchange rate: what a few years and a genre boundary cost verse Shakespeare certainly wrote, in rejections and in powers of ten of continuous odds, beside what Oxford’s sample costs. Each dot below is a genuine block under the chosen arm; the marked point is Oxford.

Every split the record allows

The 2009 proposal names a quarter; the page runs every split the dated blocks allow. From 1595 to 1603 the earlier blocks never collect more than 5 rejections, while the most extreme continuous composite deepens to 7.9 × 10⁻²⁰ at 1603. The planted discrete departure is found in 67% to 100% of blocks across the splits, the continuous one in 23% to 100%. 7 dated years are refused, with their reasons, and the refused splits are disclosed rather than hidden: unrefused, the earlier blocks would reach 5 rejections at 1604 (ranges from 33 later blocks), 6 at 1605 (28), 6 at 1606 (24), 8 at 1607 (17) and 10 at 1610 (10), so from 1605 on genuine early blocks would reach Oxford’s count. Ranges from so few later blocks reject genuine Shakespeare more often than the Clinic’s own 5 per cent even with no drift, which is why the page refuses them; what they would show mixes narrow ranges with a decade of drift, and the page cannot separate the two.

The discount ladder

Table 3.1 marks each test with what it may be sensitive to: t for time of composition, p for prosody, g for genre, e for editing. Remove the tests the claimants themselves flagged, and watch Oxford’s count and both odds recompute against Shakespeare’s most discrepant poem block on the same tests.

Remove the time-flagged tests and Oxford keeps 4 of 11 rejections, 7,680 to 1 on the discrete composite. Remove time, prosody and genre together and he keeps 0 of 7: discrete odds of 1, and continuous odds of 5.2 to 1 against the Venus block. The two sources disagree on two tests (feminine endings: Table 3.1 “t, p”, the key “P”; BoB7: Table 3.1 “t, s/m”, the key nothing), so the ladder lets you choose.

We searched the web, Google Scholar’s list of the works citing Oxford by the Numbers (by title), a site search of the Shakespeare Oxford Fellowship’s pages, the listing of Ward Elliott’s Claremont conference papers, Early Modern Literary Studies and GitHub (repositories and code) on 2026-09-23 and did not find a recomputation of the Claremont Clinic’s Oxford composite odds from its published Appendix Six block values, or a run of Shahan and Whalen’s 2009 early-against-later split on those values. The nearest precedents we found are MacDonald P. Jackson (2007), a Bayesian treatment of the Clinic’s tests for Hand D of Sir Thomas More; Gray Scott (2006), chance models run over the Clinic’s whole-play rejection counts; and Thomas Merriam (2009), who noted that the discrete composite tacitly treats the Clinic’s tests as independent and that seventeen of the forty-eight whole-play tests in Oxford by the Numbers correlate significantly with date.

VI · the verdict, dated

What the record says, as of now

Run on the Clinic’s own published block values, its formula reproduces the 400,000-to-1 discrete odds (398,000) and puts the continuous odds at 8.8 × 10²¹, far past the printed 150 trillion; scored against ranges from his own later plays, Shakespeare’s earliest play-verse blocks fall outside at least one in 17 of 23 cases, as the 2009 proposers predicted, but none reaches Oxford’s rejection count, a departure the test is shown able to catch in 21 of 23 planted copies at the proposal’s split (though not reliably at later ones); the continuous formula gives his own play verse, scored against his poems, odds up to 6.4 × 10²², and his 96 baseline blocks can certify only about 10.2 to 1 of the model’s 398,000.

Sources that establish it

1. John M. Shahan and Richard F. Whalen, “Apples to Oranges in Bard Stylometrics: Elliott & Valenza fail to eliminate Oxford”, The Oxfordian IX (2006): 113-125, which argues that genre, time of composition and scope of sample make the elimination “unwarranted”.
2. Ward Elliott and Robert J. Valenza, “My Other Car is a Shakespeare: A Response to Shahan and Whalen’s ‘Apples to Oranges in Bard Styometrics’” (sic), dated 29 October 2007, published in The Oxfordian X (2007), which concedes the songs point and declines the songs comparison.
3. John M. Shahan and Richard F. Whalen, “Auditing the Stylometricians: Elliott, Valenza and the Claremont Shakespeare Authorship Clinic”, The Oxfordian XI (2009): 235-267, which proposes the early-against-later split (p. 259, point 7) and separates style odds from authorship odds (point 6).
4. Ward E. Y. Elliott and Robert J. Valenza, “The Shakespeare Clinic and the Oxfordians”, The Oxfordian XII (2010), with Shahan and Whalen’s reply in the same volume: the exchange ends with the split and the songs comparison unrun.
5. Ward E. Y. Elliott and Robert J. Valenza, “Oxford by the Numbers”, Tennessee Law Review 72 (2004): 323-453, p. 351, where the claimants themselves write that composite probability scores “are not indicators of the absolute, actual probability that Shakespeare wrote the block in question”.
6. Thomas Merriam, “Untangling the derivatives: points for clarification in the findings of the Shakespeare Clinic”, Literary and Linguistic Computing 24 (4) (2009): 403-416, doi:10.1093/llc/fqp026, which finds the Clinic’s tests “never 100% statistically independent as Elliott and Valenza tacitly assume them to be with discrete analysis”, and seventeen of the forty-eight whole-play tests correlated significantly with date.

The dispute, in order

1. 1991Elliott and Valenza, “Was the Earl of Oxford the True Shakespeare? A Computer-Aided Analysis”, Notes and Queries: “None of the poets tested matched Shakespeare.”
2. 2000“Of Grubs and Butterflies”, their reply to W. Ron Hess’s Oxfordian re-dating of the plays.
3. 2004“Oxford by the Numbers”, Tennessee Law Review 72: the 400,000 and the 150 trillion, with every block’s scores in Appendix Six.
4. 2006Shahan and Whalen, “Apples to Oranges in Bard Stylometrics”: genre, time of composition and scope of sample.
5. 2007Elliott and Valenza, “My Other Car is a Shakespeare”: the songs concession, and the songs comparison declined.
6. 2009Shahan and Whalen, “Auditing the Stylometricians”: the early-against-later split, proposed and not run.
7. 2010Elliott and Valenza, “The Shakespeare Clinic and the Oxfordians”, with Shahan and Whalen’s reply: no side runs the split.
8. 2026This page runs it on the published blocks.

The claimants’ replies, in their words

To the genre objection: “If we stripped Oxford’s verse of every non-iambic pentameter line, and every line from the Paradyse of Daynty Deuises, we would still have a gross mismatch with Shakespeare” (2007). To the time objection, the 2000 answer above, and its last sentences: “Maybe some future writers can show that Oxford’s style was not a bit like Vic Damone’s, or that Shakespeare’s was, just as Louis Benezet liked to think. We won’t know till they try.” To the split, no reply is on record. The critics’ central objection, in theirs: “The odds that two sets of works are in the same style are not the same as the odds that one person wrote them” (2009, p. 259, point 6).

VII · the check

The check

Every free choice, and what it moves

• Which composite (discrete or continuous): moves every odds readout.
• The four verse tests’ ranges (1,500-word, as the claimants printed, or 3,000-word early): moves Oxford’s count from 6 of 14 to 7 of 14 (13.1 million to 1).
• Relative clauses: in or out of the Oxford-against-boundary comparison only, because the baseline’s per-block counts are not printed.
• The discrete rate: 2.6 per cent as printed, or the rate each baseline actually shows.
• The reference: the independence model, or the 96 blocks with a 95 per cent Clopper-Pearson bound where no block is as extreme.
• The arm and the split year: which shift genuine Shakespeare is put through.
• The plant form: additive with clamps, or proportional where Shakespeare’s mean is positive.
• The discount flags, and whose flags.
• Fixed by the page, not offered: the 80 per cent decision rule; population standard deviations (the printed ones are population, verified below); the minimum-to-maximum rule for the time arm’s ranges; the refusal of splits with fewer than 39 later blocks (what each would print is shown in the sweep); the 13 shared tests for the genre arm.

What remains uncertain

• The two composites are reconstructions from the claimants’ words; they reproduce the printed values, but the claimants print no formula.
• Oxford’s continuous error is recomputed from rounded published scores; the printed means and standard deviations were computed from unrounded ones, so the page’s value differs from print by 0.03%.
• For the 42 play-verse blocks that carry enclitic and proclitic counts, the printed continuous errors cannot be reproduced from the printed scores and profile (see the check); the 40 without them reproduce closely. The page’s arms use its own recomputation throughout.
• The plant measures Oxford’s departure against poem means and adds it to play verse; that is a choice, and the proportional form is offered beside it.
• The time arm reaches back to 1593 only. Nothing in the published tables reaches Shakespeare before then, or songs.
• The play tables date each play by its latest supposed date (the column is headed “Date Late”, pp. 427 to 428); the split inherits those dates.

Sources and data

1. Ward E. Y. Elliott and Robert J. Valenza, “Oxford by the Numbers: What Are the Odds That the Earl of Oxford Could Have Written Shakespeare’s Poems and Plays?”, Tennessee Law Review 72 (2004): 323 to 453. Author-hosted PDF, www1.cmc.edu, retrieved 23 September 2026. Shahan and Whalen report the issue as appearing in spring 2005 (2006) and cite it as 72:1, Fall 2004 (2009). The page’s two data files are transcriptions of its Table 3.1 and Appendix Six (pp. 370, 421 to 429), made from the PDF’s own text layer and checked against its rendered pages.
2. Ward E. Y. Elliott and Robert J. Valenza, “Was the Earl of Oxford the True Shakespeare? A Computer-Aided Analysis”, Notes and Queries 38 (4) (1991): 501 to 506, doi:10.1093/nq/38.4.501; online copy dated 7 April 1991 at shakespeareauthorship.com, note 5.
3. Ward E. Y. Elliott and Robert J. Valenza, “Of Grubs and Butterflies: Computers and the Oxford Claimancy Revisited”, dated 16 August 2000, www1.cmc.edu.
4. John M. Shahan and Richard F. Whalen, “Apples to Oranges in Bard Stylometrics: Elliott & Valenza fail to eliminate Oxford”, The Oxfordian IX (2006): 113 to 125, shakespeareoxfordfellowship.org.
5. Ward Elliott and Robert J. Valenza, “My Other Car is a Shakespeare: A Response to Shahan and Whalen’s ‘Apples to Oranges in Bard Styometrics’” (sic), dated 29 October 2007, www1.cmc.edu; published in The Oxfordian X (2007).
6. John M. Shahan and Richard F. Whalen, “Auditing the Stylometricians: Elliott, Valenza and the Claremont Shakespeare Authorship Clinic”, The Oxfordian XI (2009): 235 to 267, shakespeareoxfordfellowship.org.
7. Ward E. Y. Elliott and Robert J. Valenza, “The Shakespeare Clinic and the Oxfordians”, The Oxfordian XII (2010), shakespeareoxfordfellowship.org; and John M. Shahan and Richard F. Whalen, reply, same volume, shakespeareoxfordfellowship.org.
8. MacDonald P. Jackson, “Is ‘Hand D’ of Sir Thomas More Shakespeare’s? Thomas Bayes and the Elliott-Valenza Authorship Tests”, Early Modern Literary Studies 12.3 (2007), extra.shu.ac.uk.
9. Gray Scott, “Signifying Nothing? A Secondary Analysis of the Claremont Authorship Debates”, Early Modern Literary Studies 12.2 (2006), extra.shu.ac.uk.
10. Thomas Merriam, “Untangling the derivatives: points for clarification in the findings of the Shakespeare Clinic”, Literary and Linguistic Computing 24 (4) (2009): 403 to 416, doi:10.1093/llc/fqp026.

The claimants’ PDFs and the Oxfordian PDFs carry no licence statement; the page ships only transcribed numeric test values, with attribution and page references, and short quotations. The full list of sources and terms is in the page’s NOTICE.txt.