The Synoptic Problem

Summary

Matthew, Mark and Luke tell many of the same stories in many of the same words.  John does not.  That single fact is called the synoptic problem, and it is not a puzzle invented by skeptics.  It is a measurement anyone can repeat.

What the measurement shows.  Every figure below was computed for this article from the Greek text and can be re-run in under a minute.¹

• Mark and Matthew share twelve separate runs of sixteen or more identical consecutive Greek words. The longest is thirty-three words and the next is twenty-nine. Three of the twelve are Old Testament quotations, which prove only that both writers owned the same Greek Bible. Nine are not. These counts are from one edition of the Greek text. On the Nestle-Aland edition the twenty-nine becomes twenty-three and one of the twelve falls to twelve words, for reasons given under "Method and reproducibility" below.
• Matthew and Luke share twelve such runs. Eleven are in material Mark does not have at all. The twelfth, Matthew 16:25 with Luke 9:24, is a saying Mark also has. None of the twelve is an Old Testament quotation. The Nestle-Aland edition has fourteen, two of them in passages Mark also has.
• John shares none — zero — with any of the three.
• The controls that show the method works: the one long run between Luke and Acts, two books by the same author, is a quotation of Psalm 110:1. The one long run between Acts and Matthew, two books nobody says copied each other, is a quotation of Isaiah 6:9–10, and at forty-seven words it is the longest run in the whole comparison.² Long runs between unrelated books turn out to be shared Scripture and nothing else.

What that forces.  Somebody had somebody else's text open in front of him.  Robert McIver and Marie Carroll established the threshold by experiment:  volunteers reproducing a passage from memory never once produced a run longer than fifteen identical words, while a copying group routinely did.³  Their rule leaves out poetry, song and short memorable sayings, and it matters which runs survive that sieve.  The thirty-three-word run sits in one of the nine passages they call almost certainly copied.  The twenty-nine-word run is a saying they set aside as distinctive enough to have been memorised.³  So the case for copying rests on the runs that are not short sayings, and those remain.  The rule itself is disputed, and from one direction only.  John Poirier in 2004 and Mark Goodacre in 2014 both argued that it finds too little copying, not too much.^{4 5}  This article therefore treats its long runs as a floor.  It rests the claim of copying on its own control:  John and Acts share no run of even ten words with a Synoptic Gospel except where both quote Scripture.

Who copied whom.  Mark almost certainly came first — but the reason usually given for that is not the reason that works, and the article says so.

• The famous argument fails. Students are taught that Matthew and Luke never agree against Mark in the order of events, so Mark must be the source they both followed. The reasoning does not hold: if Mark came second, copying Matthew, and Luke came third, copying Mark, you would also never find Matthew and Luke agreeing against Mark. Being the middle term proves you are in the middle. It does not prove you were there first.^{6 7}
• What does work is a pattern anyone can check in a synopsis in five minutes. A writer reworking someone else's story makes a change at the start and forgets to keep it up. Matthew corrects Mark's inaccurate "king Herod" to the historically correct "tetrarch" — then eight verses later writes "the king" anyway. Luke moves the feeding of the five thousand into "a city called Bethsaida" — and then has the disciples say they are "in a desert place" and ask for the crowd to be sent to villages for food. Six such cases run from Mark to Matthew or Luke. The changes cluster at the beginnings of passages, where each writer is most himself, and the lapses fall later, where he is following his source.^{8 8 8}
• Against the main rival theory, four of the six count, not six. The Two-Gospel theory has Luke copying Matthew. Matthew also has the "no root" and the "desert place" that Luke slips back into, so those two cases fit either order, and the author of the argument says so in his own footnote.⁸ The three cases where Matthew slips, and Luke's story of the paralytic, do not fit Matthew first. The Two-Gospel school has offered one case running the other way, a single plural word at Mark 4:8.⁹ It leaves the story coherent, and two of the standard Greek editions do not print it.¹⁰
• So Markan priority rests on a narrower and more recent base than textbooks admit — not on order, not on style, but on that pattern, published in 1998, plus judgements about which version reads as secondary. The scholar who assembled the pattern is candid about what it replaced: "Many believe in the priority of Mark but few are able to give a good reason for it."⁸

What it costs.  This is where the article stops being a technical exercise.

• The four-independent-witnesses argument is dead for the three Synoptics. Where Matthew reproduces thirty-three consecutive Greek words of Mark 13, Matthew is not a second witness to those sayings. He is a copyist of them. On every proposed solution the triple tradition is one witness copied, never three.
• Markan priority is the strongest single argument against the apostle Matthew having written Matthew. An eyewitness apostle reproducing a non-eyewitness's Greek account of a scene he personally attended is hard to explain on any reading.
• Richard Bauckham, the most prominent living defender of the Gospels as eyewitness testimony, concludes that the apostle Matthew did not write it — and his reason is one verse, not a skeptical prior. Mark 2:14 calls the tax collector Levi; Matthew 9:9 calls him Matthew; the rest of the sentence is nearly identical. Bauckham: "Matthew himself could have described his own call without having to take over the way Mark described Levi's call."¹¹

What survives.  Three things, and they are worth more than the thing that was lost.

1. John is literarily independent, and that is now a measurement rather than a claim. John's overlap with each Synoptic sits inside the noise band set by comparing books nobody thinks are related.
2. The special material of Matthew and of Luke — the parts each has that the others do not — is not copied from anything surviving.
3. Paul is independent of all four and earlier than all four.

A smaller set of witnesses, honestly counted, is worth more than a larger set counted wrongly.

What remains genuinely open.  Whether the material Matthew and Luke share came from a lost document (called Q) or from Luke reading Matthew.  This article does not settle that, and says why nobody else has either.

Implications

The synoptic problem is usually presented as a specialist's puzzle.  It is not.  It decides how many independent witnesses the Gospels supply, and it constrains both who wrote them and when.

It also carries a warning that cuts in every direction.  A whole discipline held Markan priority for decades partly on an argument that does not work.  Willi Marxsen's introduction to the New Testament, in its English edition of 1968, says the two-source theory "has been so widely accepted by scholars that one feels inclined to abandon the term 'theory' (in the sense of 'hypothesis')."  His next sentence is the one usually quoted, and it is usually cut short:  "We can in fact regard it as an assured finding—but we must bear in mind that there are inevitable uncertainties as far as the extent and form of Q and the special material are concerned."¹²  The hedge covers the contents of Q.  It does not cover the theory.  Any project that prefers data to consensus has to notice how that happened — and then ask the same question about its own settled points.

The data, before any theory

Three of the four Gospels can be set in parallel columns and read across.  That is what "synoptic" means:  seen together.  John cannot.

The material divides into four kinds:

The books themselves, counted in Greek words:¹⁵

Why the famous number is not used here

The figure everybody quotes is B. H. Streeter's, from 1924:  Matthew "employs 51% of the actual words used by Mark."¹³

That number is not used in this article, and the reason is worth showing rather than asserting.  A whole-book version of it was computed here, together with controls.  It asks what share of Mark's words can be paired off, one for one, with the same word form somewhere in the other book:

The measure fails its own control.  Mark shares two-thirds of its vocabulary with Acts, a book nobody thinks copied it, because Greek prose is mostly Greek function words.  Eighty-one per cent against a floor of sixty-six is a signal of fifteen points buried in noise, and no conclusion should be built on it.

Streeter's own figure is better than that, because he counted only inside passages he had already judged parallel.  But that makes it depend on his pericope list, and different scholars using different lists get different numbers — which is why the published statistics disagree with one another.

So this article does not inherit anybody's count.  It measures runs of consecutive identical words instead, where the noise floor collapses to under one per cent and the signal survives.

The measurement

Two methods, both threshold-free, both reproducible.

Shared runs of words, with a noise floor

For each pair of books, what share of the first book's words sit inside a run of n identical consecutive words occurring somewhere in the second?

The controls are what make the numbers mean anything.  Nobody claims John copied a Synoptic.  Acts is by Luke but on a different subject.  So the John and Acts columns measure how much agreement arises from Greek simply being Greek.

At six-word runs, Mark-to-Matthew is 14.0% against a noise floor of 0.0% to 0.7%.  Against the highest control figure that is a factor of twenty, and against the others it is more.  No account of independent oral tradition produces that.

And John sits inside the noise band.  Whatever John knew of the Synoptic tradition — and he may have known a great deal — he did not have a Synoptic scroll open beside him.

Every long identical run, listed

The second method needs no percentages at all.  McIver and Carroll ran the experiment:  people reproducing a text from memory never exceeded fifteen identical words in a row; people copying routinely did.  Their rule is sixteen or more, outside poetry and formula.³

Every such run between Mark and Matthew:

Discount the three scriptural quotations and nine remain.

McIver and Carroll's own sieve is stricter than a bare count, and it should be applied.  Their rule exempts sayings short or striking enough to be carried by memory, and they say their experiments show it "is accurate only for narrative material," while poems and short aphorisms "might be remembered exactly."  They found twenty-three Synoptic parallels with sixteen or more words in exact sequence, set seven aside as short aphorisms, set "a further seven" aside as longer distinctive sayings, and kept nine as almost certainly copied.³  Four of the nine runs above fall in parallels they set aside:  Mark 8:2–3, the twenty-nine-word saying at 8:34–35, 10:45 and 14:21.  Three fall in parallels they kept, all in Mark 13:  the thirty-three-word run at 13:12–14, and the runs at 13:16–18 and 13:29–30.  They kept the Psalm 110 passage at Mark 12:35–37 as well.  The other two, Mark 1:22 and 2:9–10, are not in their table at all.  Their counts differ from the counts here by a few words in either direction:  thirty-one where this scan finds thirty-three, twenty-three where it finds twenty-nine.  They counted inside the numbered sections of Huck's synopsis, so a run cannot cross from one section into the next, and the edition of the Greek changes some runs, as the Method section shows.  So on the stricter reading the certain copying between Mark and Matthew is concentrated in one discourse, Mark 13.  That is enough to prove a written link.  It is much less than twelve.

An independent hand count agrees with this one.  Poirier recounted the long agreements on the Nestle-Aland text.  He found that McIver and Carroll's table tended "to record shorter sequential agreements and to omit sequential agreements that should have been included," and printed a corrected chart of twenty-seven agreements of sixteen words or more.⁴  Twenty-five of the twenty-six runs listed in this article are in it.  Twenty have the same length there, four differ by a word or two, and the twenty-fifth is Mark 8:34–35, which is twenty-three words on his text.  The one run here that his chart lacks, Mark 2:9–10, and the two runs of his that this scan lacks, both between Matthew and Luke, all fall where the SBL edition and Nestle-Aland print different Greek.  The Method section gives the readings.

Between Matthew and Luke there are twelve more, and not one is an Old Testament quotation:  Matthew 3:9–10 // Luke 3:8–9 (24 words), 6:24 // 16:13 (26), 7:7–8 // 11:9–10 (24), 8:9–10 // 7:8–9 (25), 8:20 // 9:58 (24), 11:7–8 // 7:24–25 (19), 11:25–27 // 10:21–22 (27), 12:41 // 11:32 (24), 12:42 // 11:31 (16), 16:25 // 9:24 (16), 24:50–51 // 12:46 (26), and 3:10 // 3:9 (20).  Two of the twelve, 3:9–10 and 3:10, are one passage broken by a single differing word.  One of the twelve is not double tradition:  Matthew 16:25 // Luke 9:24 is the saying about saving and losing one's life, which Mark 8:35 also has, and ten of its sixteen words stand in Mark in the same order.  That leaves eleven runs in material Mark lacks.  Goodacre noticed the same thing in McIver and Carroll's list:  "Two of the eleven Matthew // Luke pairings are actually in triple tradition," and long agreements of Matthew and Luke inside a Markan passage suggest "a direct link of some kind between Matthew and Luke."¹⁶  The run found here is weak evidence of that by itself, because Mark has nearly all of its words too.

Notice how far this material travels.  Matthew 6:24 sits inside the Sermon on the Mount.  Luke 16:13 sits in a block about a dishonest steward.  Twenty-six identical Greek words, in two completely different settings.

Between Mark and Luke there are only two:  Mark 1:24–25 // Luke 4:34–35 (26 words) and Mark 10:14–16 // Luke 18:16–18 (30).  Luke rewrites.  McIver and Carroll set both aside as distinctive sayings, so their nine near-certain cases contain no Mark and Luke pair at all.³

Between John and any Synoptic, none.

And the controls.  Between Acts and Luke there is exactly one, and it is Psalm 110:1 — the same author, quoting the same Greek Bible twice.  Between Acts and Matthew there is also exactly one, Acts 28:26–27 // Matthew 13:14–15, forty-seven words of Isaiah 6:9–10.  That is the longest run in the whole comparison, and it joins two books nobody thinks copied each other.²  The method detects shared text, not shared authorship and not shared style, and a quotation of the Greek Old Testament is shared text.  That is the reason the three scriptural runs between Mark and Matthew are discounted above.

Who copied whom

The argument that does not work

Generations of students have been taught that Mark came first because Matthew and Luke never agree against Mark in the order of events — so Mark must be the common source they both followed.

The reasoning is faulty, and the fault is easy to see once stated.  Suppose instead that Mark came second, copying Matthew, and Luke came third, copying Mark.  Then Luke would follow Mark's order and Matthew would be Mark's source, and you would still never find Matthew and Luke agreeing against Mark.  The pattern is equally consistent with Mark being in the middle for a different reason.  Being the middle term proves you are in the middle.  It does not prove you were there first.

B. C. Butler set this out in 1951, in a chapter he called "The Lachmann Fallacy."  His conclusion from the order data:  "Mark is necessarily the connecting-link between Matthew and Luke in these passages, but not necessarily the source of more than one of them."⁶  What makes the episode worth telling is William Farmer's response.  Farmer, who went on to lead the revival of the Griesbach theory, had taught the order argument for a decade before he understood the objection.  His own account:  "I used it in my class lectures for ten years.  I estimate that there are at least three hundred intelligent human beings loose in the world who firmly believe in the priority of Mark because of the convincing use I made of that particular non sequitur."⁷

Both sides now grant the point.  Goodacre, who defends Markan priority, concedes it in his own words:  "Arguments that were once thought to be decisive, like appeals to Mark's rough Greek or the ordering of triple tradition material, are now seen to be unconvincing and reversible."⁸  The Two-Gospel research team, which puts Mark last, says the same of the argument and then claims no more than a balance of probability for its own reading of the order:  "The argument from the order of pericopae, as formulated by Streeter et al., is inconclusive because it is reversible."⁹

And Goodacre on the older arguments generally:  "All of these features are merely suggestive.  Not one of them appears decisive."  A paragraph later:  "What we would like is something that does not merely point the finger, but actually secures the conviction.  We need something decisive.  We need fingerprints on the gun."¹⁷

The argument that does work

Editorial fatigue.  The idea is simple.  A writer reworking someone else's account makes changes at the beginning that he cannot keep up.  Goodacre's definition:  "Like continuity errors in film and television, examples of fatigue will be unconscious mistakes, small errors of detail which naturally arise in the course of constructing a narrative."⁸

The single cases are older than the argument.  Goodacre credits the first to G. M. Styler and others to Michael Goulder.⁸  What he added in 1998 was the pattern, and a test for it.

Six cases, all checkable in a synopsis in five minutes:

Why this one is not reversible, in Goodacre's words:  "Rather, in most cases, Matthew and Luke differ from Mark at the beginning of the pericope, at the point where they are writing most characteristically, and they agree with Mark later in the pericope, where they are writing less characteristically.  It is not possible to find the same phenomenon in Mark."⁸

And he states in advance what would refute him — which is what makes it an argument rather than a debating point:  those who reject Markan priority "would need to find strong illustrations of the phenomenon as a control."⁸

What the other side says about it

The main rival is the Two-Gospel theory:  Matthew first, Luke using Matthew, Mark last and combining the two.  Its fullest statement is a commentary on all of Mark by a research team, One Gospel from Two (2002).  The team answers editorial fatigue in one sentence.  It accepts that the thing happens and claims a case of it running their way, in the parable of the sower:

"The final appearance of ἄλλα in the plural in Mk 4:8, in verbatim agreement with Mt 13:8, may be an example of what Mark Goodacre and others have come to identify as evidence of 'fatigue' on the part of one evangelist and, thereby, a clear indication of Mark's dependence on Matthew in this context.  That is, Mark failed to make his usage of ἄλλο consistent because he inadvertently followed Matthew in this case."⁹

The Greek is as they describe.  Matthew writes the plural "others" for every sowing, at 13:5, 13:7 and 13:8.  Mark writes the singular "other" at 4:5 and 4:7 and the plural at 4:8.  Three things weaken it as a counter-example.

• It causes no contradiction. Goodacre's cases are mistakes in the story: a tetrarch who becomes a king, a city that becomes a desert. Mark's parable reads smoothly with the plural.
• The plural has a reason inside Mark's own sentence. The good seed yields three separate crops, "thirty and sixty and a hundred," and in the editions that print the plural "others" the participle that follows, "growing," is plural as well.
• The reading is not secure. Westcott and Hort, the Nestle-Aland text and the SBL edition print the plural at Mark 4:8. Tregelles and the Byzantine text print the singular, and with the singular the case disappears.¹⁰

The same book also explains the clash behind one of Goodacre's six, without naming him, and here the reply works.  It grants that Luke's Bethsaida clashes with his "desert place," and explains the clash as Luke slipping back into Matthew:  "Luke's urban setting for the feeding, in Bethsaida (Lk 9:10), produced internal inconsistencies with Matthew's statement that this feeding took place in a desert place, which Luke adopted without change (cf. Lk 9:12//Mt 14:15)."⁹  On their theory Luke had Matthew open, and Matthew 14:15 has both the desert place and the villages.  So a Lucan lapse counts against the Two-Gospel theory only where Matthew lacks what Luke slips back into.

Checked case by case against the Greek, that leaves four of the six:

Goodacre saw this in 1998 and put it in a footnote:  "two of the examples from Luke above would make sense also on the theory of Matthean priority," the sower and the five thousand, while with the paralytic "this is not the case since Matthew (9.1-8) does not have any reference to the house that is mentioned in Mark and assumed in Luke."⁸

The three Matthean cases carry the weight, and the Two-Gospel side has answered half of one of them.  The death of the Baptist was used against Matthew first long before Goodacre.  G. M. Styler made both halves of it in 1962.  Matthew says Herod wanted John dead, and then "at xiv. 9 Matt. betrays the fact that he really knows the full version by slipping in the statement that 'the king' was sorry."  Styler's footnote adds the title:  Matthew "correctly calls him 'tetrarch' in xiv. 1, but lapses into calling him 'king' at xiv. 9."¹⁸  Lamar Cope, later a member of the Two-Gospel team, replied in 1976 and again in 1983.  His reading is that Matthew's Herod wanted John dead all along and held back only for fear of the crowd, which is what Matthew 14:5 says.  The girl's request forces him to do in public what he had not dared to do:  "Thus Herod regretted, not the execution, but the foolish promise that had boxed him into a trap."  Cope adds that Josephus also makes Herod John's enemy, so that "Matthew and Josephus, surely independently, agree against Mark about the character of Herod Antipas," and he concludes that "here Mark has edited Matthew."¹⁹

That reply is a fair one, and it removes the second half of the first case.  Matthew 14:9 has Herod grieved and then, "because of the oaths and the guests," giving the order.  A man annoyed at being cornered by his own oath fits those words.

It does not touch the first half.  Matthew calls Herod "the tetrarch" at 14:1 and "the king" at 14:9, and Cope does not discuss the title.  One Gospel from Two discusses it only from Mark's side:  Mark's "king" is a "loose description" that "may have been determined by the later narrative account at Mk 6:23 where he offers half of his kingdom."⁹  That explains why Mark would write "king."  It does not explain why Matthew, on their theory the first writer and working from no source, would correct himself in the wrong direction eight verses on.

On the other two Matthean cases, the leper healed before "many crowds" and told to tell no one, and the family standing "outside" a house never mentioned, nothing was found.  The paragraphs of One Gospel from Two on those two passages do not raise either problem.  The team's earlier volume, Beyond the Q Impasse (1996), came out two years before Goodacre's article and returns nothing for "fatigue."  David Alan Black's Why Four Gospels?, the short popular case for Matthew first, lists Goodacre's article in the bibliography of both its editions and never mentions it in the text.²⁰  A search that finds nothing is weak evidence, and an answer may exist in a journal this article did not reach.  One search here already failed that way:  an earlier pass reported that One Gospel from Two never mentions the tetrarch, because the scan garbles the Greek word and the English one is not used.

Its weaknesses, stated plainly.  Every example is individually answerable.  "The king" could be a stock phrase.  "A desert place" could be a set expression for open country outside a town.  A critic can always plead carelessness instead of dependence.  The Bethsaida case also depends on which Greek text is read:  the Byzantine text of Luke 9:10 has Jesus withdraw "to a desert place of a city called Bethsaida," which removes the clash, and Goodacre treats that reading as copyists mending the difficulty.^{10 8}  What is hard to answer is the pattern:  six cases running from Mark outward, four of them unexplained if Matthew came first, against one disputed word running the other way.  Goodacre also owns the weak link himself — in five of his six, not all six, does the evangelist begin characteristically, and the exception is the feeding of the five thousand.⁸

So Markan priority rests on a narrower and more recent base than most textbooks admit.  Not on order, not on style — essentially on editorial fatigue, published in 1998, plus judgements about which version looks secondary.  Against the Two-Gospel theory that base is narrower still:  four cases, three of them in Matthew, and half of one of those has a reasonable answer.  That is worth saying out loud.

Q, and the case against it

If Mark came first, where did Matthew and Luke get the roughly 200 verses they share and Mark lacks?

Two answers.  Q — a lost written collection of sayings, used independently by both.  Or Farrer's answer — there was no such document, and Luke simply read Matthew.

What the measurement contributes

Two figures from the table above bear directly on this, and they pull in opposite directions.

Against a purely oral explanation:  eleven runs of sixteen or more identical Greek words in the double tradition, none of them scriptural quotation.  Six of the eleven are sayings that McIver and Carroll set aside as memorable enough to travel by mouth.  The other five sit in four passages they keep as almost certainly copied:  the Baptist's preaching, Jesus on John, the thanksgiving to the Father, and the wise and wicked servants.³  Whatever the double tradition is, it involves a Greek text.  It cannot be two independent translations of Aramaic preaching.

A problem for Q that is rarely stated this baldly:  at ten-word runs, Luke's overlap with Matthew (4.2%) is higher than Luke's overlap with Mark (2.0%).  If Luke had both Mark and Q in front of him, he rewrote his narrative source heavily while copying his sayings source closely.  There are decent answers — most traditions treat sayings as more fixed than narrative — but the asymmetry is real, and it is equally consistent with Luke copying Matthew.

The minor agreements

The sharpest problem for Q is small and specific:  places where Matthew and Luke agree against Mark inside a passage all three are telling.  On the two-source theory those two writers never saw each other's work, so such agreements should not exist.

The clearest example was checked from the Greek for this article.  After Jesus is blindfolded and struck, Mark 14:65 has only:

Προφήτευσον — "Prophesy!"

Matthew 26:68 and Luke 22:64 both continue with the identical five words:

τίς ἐστιν ὁ παίσας σε; — "Who is it that struck you?"

The verb παίω (paiō, "strike") occurs exactly once in Matthew and once in Luke — this verse in each.²¹  It is not a habit of either writer.  Two authors who had never seen one another's work both added the same five-word question, using a verb neither uses anywhere else, to the same moment in Mark.

That is hard to call coincidence.  The standard replies are that the agreement entered by scribal harmonisation, or that Matthew and Luke used a slightly different edition of Mark.  Both are possible.  Both are also unfalsifiable rescues, and a reader is entitled to notice that.

A second case is longer, and it depends on the edition.  On the Nestle-Aland text Matthew 8:2–3 and Luke 5:12–13 share eighteen consecutive words in the healing of the leper, a story Mark also tells.⁴  Across those words Mark 1:40–42 differs five times.  It has no "Lord."  It names Jesus' emotion, which they leave out.  It has "stretched out his hand and touched" where they have "stretched out the hand and touched him."  It has "and says to him" where they have "saying."  It uses a different form of the word "immediately."  McIver and Carroll kept this passage among their nine near-certain cases of copying.  On the two-source theory they assume, Matthew and Luke never saw each other's work, so a copied run between them inside a Markan story should not exist.  Goodacre made that point against them.¹⁶  The SBL edition reads a different word at Luke 5:13 and shows a run of thirteen.

Where this article lands

Undecided, and deliberately so.  Editorial fatigue leans toward Farrer:  Goodacre finds Luke lapsing in double-tradition passages and reports of Matthew, "I have looked for examples and cannot find any."  He grants that such cases show Luke following a text and cannot show whether the text was Matthew or Q.⁸  Luke's ordering of the double tradition, and cases where Luke's version of a saying looks more primitive than Matthew's, lean toward Q.  Neither is decisive.

Note the incentives, because a skeptical reader will.  Conservatives have a reason to want Q — it would be an early source, plausibly from the 40s or 50s, close to the events.  They also have a reason to want no Q — it is an invented document nobody has ever seen.  Farrer-school scholars are mostly not conservatives.  The question should be decided on the minor agreements and on editorial fatigue, and whichever way it goes, something is lost.

What is lost under Farrer is worth naming precisely.  If Luke simply copied Matthew, then "Matthew and Luke both record this saying" collapses into "Matthew records this saying."  The Lord's Prayer stops being two independent versions of an earlier prayer.  The Beatitudes stop being a window onto a form behind both.  Much of the ethical teaching loses its second voice.

But Q would not supply what people think either.  Q is one hypothetical document.  "Matthew and Luke independently used Q" gives two copies of one source, exactly as "Matthew and Luke both used Mark" gives two copies of Mark.  Q moves the single source earlier.  It does not multiply witnesses.

The alternatives, and how much of the field holds each

There is no census of scholarly opinion, so the fairest witness is a specialist who is himself a dissenter and has no motive to inflate the majority.  Goodacre:  "the Two-Source Theory is accepted without question by the vast majority of scholars in the discipline."¹⁷

His next distinction is the more useful one:

"Among those who might be called experts on the Synoptic Problem, there is a variety of opinion — a good proportion believe in the Two-Source Theory but an equally high proportion question at least some aspect of it.  On the other hand, among those who write books on the Gospels not dealing directly with the Synoptic Problem, there tends to be a kind of blithe confidence, almost a complacency over the correctness of the Two-Source Theory."¹⁷

That is a general pattern worth carrying to every field:  confidence is often highest among those furthest from the primary evidence.

The two solutions that match the church fathers are the two that put Matthew first.  Every patristic witness that speaks to the order does.  Every internal argument that carries weight puts Mark first.  That conflict is real and this article does not hide it.

Goodacre's framing of how the majority weighs the two is the fairest available:

"It is the same with the Synoptic Gospels.  We listen to the external evidence, but if it does not square with the overwhelming internal evidence, we have no choice but to place a question mark against it."¹⁷

Two things that framing does not settle.  First, it assumes the internal evidence is "overwhelming," and Goodacre has written six pages earlier that none of the standard arguments "appears decisive."  The whole weight rests on one argument from 1998.  Second, he offers a deflationary account of the fathers — that they were making inferences from the same texts we have, not preserving memories — and labels it as an assumption.  Whether that assumption is true is a question for the authorship article, not this one.

Memory, not only copying

It would be a mistake to leave the impression that the Synoptics are nothing but scissors and paste.

McIver and Carroll's own verdict on the parallels is the half of their work that both sides tend to skip:  "overall, the majority of the parallels in the Synoptic Gospels have characteristics that are more typical of a process of transmission that involves memory rather than straight copying."³

That verdict has been challenged twice in the journal that printed it.  Both replies say the same thing from different sides:  the sixteen-word test finds too little copying.

John Poirier replied in 2004.  His first point is about what was compared.  McIver and Carroll set agreements between Matthew and Luke, two books they assume never saw each other, beside agreements between a student and the text on his desk:  "they compare these patterns of agreement with those of a known case of direct borrowing."  His second point is about the gaps.  A long agreement in the Gospels is often broken by one small editorial change, a different particle or a changed word order, and the wording is the same on both sides of the break.  "The absence of a 16-word sequence, therefore, does not mean that the patterns of verbal agreement between two Gospels at a given point can be characterized as atomistic or that they represent 'the mechanisms of memory.'"  His conclusion:  "The evidence of the Gospels does not suggest that we should gather the patterns of agreement into two baskets, the one marked 'written source' and the other 'memory.'  Rather, the numbers plot a continuous curve."⁴

Mark Goodacre replied in 2014.  "The mistake is simple but important:  their experimental data are in English whereas the Synoptic Gospels are in Greek."  His reason:  "it takes many more words to say something in contemporary English than it takes to say something similar in Koine Greek.  Therefore, sixteen words in English correlates to fewer words in Greek, and so the sixteen-word criterion is much too high for Greek."  He set the long Greek agreements beside the same passages in an English translation.  The thirty-one Greek words shared by Matthew 10:21–22 and Mark 13:12–13 are forty-five in English.  With a bar set for Greek, he concludes, there would be "a vastly greater number of passages that show evidence of copying in the Synoptic Gospels," and "their sixteen-word criterion should be abandoned."⁵

What this does to the argument here.  Three things, and two of them are concessions.

• Sixteen words is not an established threshold for Greek, and this article cannot say it is. The figure came from English prose written by Australian undergraduates. The claim that the Synoptics are copied does not need it. That claim rests on the controls in the tables above, which are Greek and come from books of the same kind. John shares no run of even ten identical words with any Synoptic Gospel or with Acts. Acts shares four runs of ten or more with the Gospels, and every one is Psalm 110:1 or Isaiah 6:9–10.
• The long runs are a floor. Both critics hold that copying reaches well below sixteen words, and this article's own coverage table agrees with them. At six-word runs fourteen per cent of Mark's words lie inside wording shared with Matthew, against under one per cent for the controls. That is far more of Mark than twelve long runs cover. Poirier's "continuous curve" is what that table shows.
• "Most parallels are memory" cannot be claimed as a finding. McIver and Carroll said it. Two published replies deny that their test can show it. McIver's later book does not name Poirier, by a search of its text, and no answer to either reply was looked for beyond that. What can be said is narrower. Where two Gospels share only scattered words and short phrases, the measures in this article cannot tell an edited copy from an independent memory of the same teaching. They do not show that the remainder is memory, and they do not show that it is copied.

So the picture is still layered, but not into the two baskets McIver and Carroll named.  In some passages a text was plainly open on the desk.  This scan finds twenty-six long runs across the three Synoptic pairs, twenty-three of them not scriptural quotation.  Around them is a wide band of shorter shared wording that no unrelated pair of books shows.  Beyond that band the evidence gives out.

What this costs, and what survives

The four-witnesses argument has to be retired

Popular apologetics counts the Gospels as four independent witnesses.  After this measurement, that count cannot stand.

• Where Matthew reproduces thirty-three consecutive Greek words of Mark 13, there is one witness to those sayings and a copy of it.
• On the two-source theory, Matthew and Luke are independent of each other but both depend on Mark — so the triple tradition is one witness with two copies.
• On Farrer and on Griesbach, it is one witness copied twice in series.
• No solution yields three.

The criterion of multiple attestation, used properly, applies to sources and not to books.  A saying found in Mark, Matthew and Luke is singly attested.  A saying found in Mark and in John, or in Mark and in Paul, is doubly attested and is worth far more.

What genuinely remains independent

1. John. Zero long runs with any Synoptic; six-word overlap inside the noise band. This is the cleanest number in the whole table.
2. The special material of Matthew and of Luke.
3. Paul, who is independent of all four and writes earlier than all four.
4. Possibly, some part of the remaining Synoptic parallels. McIver and Carroll assign most of them to memory. Poirier and Goodacre deny that the test can show it, so this article does not count them.

That is a smaller set than four Gospels.  It is also a set that survives scrutiny, which the larger claim does not.

Apostolic Matthew

This is the heaviest cost and it should be paid openly.

The tradition, reaching back to Papias around AD 100–110, is that the apostle Matthew — one of the Twelve, a tax collector called by Jesus — wrote the Gospel bearing his name.  Our Greek Matthew reproduces thirty-three consecutive identical Greek words of Mark at Mark 13:12–14, twenty-nine at Mark 8:34–35, and twelve runs of sixteen or more in all.  Mark, on every ancient account, was not an eyewitness; he wrote down Peter's preaching.

So the traditional attribution asks an eyewitness apostle to have copied, verbatim and at length, a second-hand report of scenes he was present at.

One honest qualification, measured rather than assumed.  Sorting the nine non-scriptural runs by genre, eight of the nine are sayings of Jesus — and the single narrative case, Mark 1:22, is exactly sixteen words, sitting on the threshold rather than above it.²²  Two narrative scenes were checked as controls.  At the Last Supper, where Matthew was present, the longest identical run is eleven words.  At the Transfiguration, where only Peter, James and John were present, it is ten.  Both fall below sixteen words.  McIver and Carroll's own sieve points the same way:  the Mark and Matthew passages they keep as copied are a discourse and a Psalm quotation, and none is narrative.  That is the pattern one expects if a writer treated dominical sayings as fixed wording and narrated events in his own words — not a strange thing for a disciple to do.

The qualification has a limit, and it is a real one.  Goodacre's objection applies here.  If sixteen words is too high a bar for Greek, then a run of ten or eleven is not evidence of memory.  The controls say the same:  between John or Acts and a Synoptic Gospel no run of ten words occurs outside quotations of Scripture.  So the narrative scenes are reworded more than the sayings.  They are not shown to be independent of Mark.

But the residue is the problem, and it is exactly the passage Bauckham points to.  Sixteen of the twenty words of Mark 2:14 reappear in the same order in the twenty-two words of Matthew 9:9, eighty per cent of Mark's sentence, and it is the author's own call story.  The longest unbroken run is nine words, so this is not a case the sixteen-word rule catches.  It is a case of a different kind:  one sentence, one name changed.

Bauckham's version of the problem is sharper than the word counts, and it turns on a single verse.  Mark 2:14 tells of the call of Levi son of Alphaeus.  Matthew 9:9 tells the same story about Matthew.  Apart from the name, the two accounts are virtually identical in wording.  Bauckham's conclusion:

"If this explanation of the name Matthew in Matt 9:9 is correct, it has one significant implication:  that the author of Matthew's Gospel intended to associate the Gospel with the apostle Matthew but was not himself the apostle Matthew.  Matthew himself could have described his own call without having to take over the way Mark described Levi's call."¹¹

There is a supporting detail.  In Mark the call is followed by dinner "in his house" (Mark 2:15).  Matthew keeps the dinner but changes it to "the house" (Matthew 9:10).  The writer took over exactly as much of Mark's Levi story as he needed to attach the call to Matthew, and no more.  That looks like an editor working on a text, not a man remembering his own dinner party.

The full weighing of the answers on offer belongs to the authorship article.  The synoptic verdict that article inherits is this:  on Markan priority, apostolic authorship of Matthew in its strong form is very hard to defend, and the field's leading defender of eyewitness testimony does not defend it.

What it does to dating

The standard dates are a chain, not four independent estimates:

1. Mark ≈ 65–75.
2. Matthew and Luke used Mark, so both are later.
3. Allow ten to twenty years for Mark to circulate.
4. Therefore Matthew ≈ 80–90, Luke ≈ 80–90.
5. Acts follows Luke; John is dated last on other grounds.

Steps 2 and 3 are pure synoptic theory.  Remove Markan priority and steps 2 through 5 do not follow.  A reader told that "Matthew is dated to the 80s" is being told the output of a source theory plus an assumed interval — not an observation.

There is also a circularity worth naming.  Goodacre uses the AD 70 evidence to support Markan priority.¹⁷  Elsewhere the same relative dating is derived from Markan priority.  Each move is legitimate alone.  Used together they form a loop.  The dating article takes this up.

Objections and responses

"The agreements just show they were describing the same events." No.  Two honest witnesses to one event produce agreement in substance, not thirty-three consecutive identical words in a language neither of them spoke at home.  The noise floor settles it:  John describes many of the same events and produces zero such runs.

"Copying means the Gospels are unreliable." It does not follow, and conflating the two is the commonest error in this area.  Using a good source is what a careful writer does.  Luke's first sentence says that many had already drawn up accounts, and that he had followed everything closely before writing his own (Luke 1:1–4).  What copying costs is independence, not accuracy — those are different currencies and this article is careful to spend only the one that is actually owed.

"Markan priority is just the liberal consensus." The strongest argument for it was published by a scholar who uses it to eliminate Q, against the same consensus.  And this article states the discipline's own embarrassment:  the argument most students were taught contains a logical error, and the man who revived the main rival theory had taught it for ten years without seeing the flaw.

"If Matthew used Mark, the tradition is simply wrong." Not simply.  The tradition could be right that the apostle Matthew stands behind the Gospel and wrong that he wrote the Greek text we have.  That is Bauckham's position, and it is not a dodge — it is what the evidence of Matthew 9:9 actually supports.  But it does concede that the man who composed our Greek Matthew was not the apostle.

"Q is an invented document, so the two-source theory is unscientific." Q is a hypothesis, and it should carry the burden that hypotheses carry.  But hypothetical sources are ordinary in the study of ancient texts, and the alternative is not "no hypothesis" — it is the different hypothesis that Luke read Matthew.  The honest complaint is not that Q was proposed; it is the inclination Marxsen put into print, to stop calling it a hypothesis at all.

"John's independence proves John is accurate." It does not.  Independence of text is not accuracy, and this article claims only what it measured.  What John's independence does establish is that where John and a Synoptic agree, there really are two witnesses.

What is not known

• The direction of dependence is not settled by any word count. Butler was right about that, and no measurement in this article changes it. Markan priority rests on editorial fatigue and on judgements about which version reads as secondary — not on arithmetic.
• Q or Farrer remains open. This article does not pretend otherwise.
• Whether Luke used Matthew as well as Mark and Q — a third position, available and rarely considered.
• No synoptic solution supplies an absolute date for anything.
• Whether the Two-Gospel school has answered the leper and the family cases. Nothing was found in the three books searched. Lamar Cope's 1976 article on the Baptist passage was read only through his own 1983 restatement.
• Where copying stops. The sixteen-word rule is contested, and the two published replies both say it sets the bar too high for Greek. No one has run the memory experiment in Greek. Below the long runs this article can show that shared wording far exceeds the controls. It cannot say, passage by passage, which agreements are edited copying and which are shared memory.
• How many minor agreements there are. No total is given in this article, because none could be checked against Neirynck's list or Morgenthaler's tables.

Method and reproducibility

Every figure computed for this article comes from the MorphGNT morphological tagging of the SBL Greek New Testament.  The tools are in the repository:  bin/synoptic-agreement.py for the runs and coverage, bin/nt-stylometry.py for the authorship measures used in the companion articles.  Every table and count was run again on 1 October 2026 and came out the same.

Three limits, stated because a reader is entitled to them:

• The n-gram measure counts a run occurring anywhere in the target book, not at the parallel place. For runs of five words or more this is harmless — the John and Acts controls put the accidental rate under 1% — but the three-word column should never be quoted alone.
• Comparison uses MorphGNT's normalised word form, which treats ἐστίν and ἐστί as the same word. A movable nu is a scribal habit, not an evangelist's choice.
• The run lengths depend on the edition in four places. The SBL edition was checked against its own apparatus, which records every place it differs from the Nestle-Aland 28th edition. Inside the twenty-six Synoptic runs there are two such places that matter. At Mark 8:34 the SBL text reads ἐλθεῖν with Westcott and Hort, as Matthew does; Nestle-Aland, Tregelles and the Byzantine text read ἀκολουθεῖν, which breaks the run after five words and leaves twenty-three, not twenty-nine. At Mark 2:10 Nestle-Aland puts "on earth" after "to forgive sins," which cuts that run from seventeen words to twelve, below the threshold. On Nestle-Aland, then, Mark and Matthew share eleven runs, eight of them not scriptural, and the longest is still thirty-three.¹⁰ The apparatus can only show an existing run shortening. Poirier's chart supplies the other direction: Nestle-Aland has two runs between Matthew and Luke that the SBL text breaks. One is the healing of the leper, Matthew 8:2–3 // Luke 5:12–13, eighteen words on Nestle-Aland and thirteen here, because the SBL edition reads εἰπών for λέγων at Luke 5:13. The other is Matthew 11:8–10 // Luke 7:25–27, eighteen words on Nestle-Aland, broken here by the word order of Matthew 11:9.^{4 10} So on Nestle-Aland the three pairs have eleven, fourteen and two long runs, twenty-seven in all, against twelve, twelve and two here. No conclusion in this article turns on the difference.
• The book totals were cross-checked against a second critical edition, and they hold. Peter Williams gives Matthew 18,347, Mark 11,103, Luke 19,463 and John 15,445 from the Tyndale House Greek New Testament. Ours differ most on Mark — 11,286 against 11,103 — and the reason is the longer ending: Mark 16:9–20 is 167 words in the SBLGNT, and removing it gives 11,119, leaving a residue of sixteen words across the whole book. Williams's own table note says his count leaves those verses out.²³ Two independently edited texts agreeing to within a fraction of a percent is the right kind of check on a word count, and it is why the book totals above are not edition-dependent.

An automatic verse-by-verse alignment was also built and then abandoned, because calibration against known controls — Luke's omission of Mark 6:45–8:26, the two infancy narratives, the special material — showed it reaching only 54% recall at the threshold where false positives fell to 5%.  Parallel verses do not align one to one.  That failure is reported rather than hidden.

Sources