A new chronology for Shakespeare’s plays

Douglas Bruster, Geneviève Smith
DOI: http://dx.doi.org/10.1093/llc/fqu068 First published online: 9 December 2014


It is widely recognized that Shakespeare’s verse lines grew progressively longer as his career unfolded. Scholars have traditionally used this fact, among others, to date the plays. Drawing on the existing and original data relating to their verbal arrangements, this essay constructs a new chronology for 42 dramatic texts, and parts of texts, by Shakespeare. This chronology is based on a constrained correspondence analysis of the plays’ internal pauses, qualified in relation to a principal component analysis of other verbal features and the recorded closings of the London playhouses owing to plague. The result is a more specific ordering of the Shakespeare canon than has previously been available.

1 Introduction

It is a commonplace that Shakespeare’s lines became longer throughout his career. Nearly as familiar to scholars is how this aspect of his verse, particularly his lines’ internal pauses as marked by punctuation, sheds light on his works’ chronology. Together with external evidence (such as publication, or records attesting to performance or the availability of a manuscript for printing), changes to Shakespeare’s habits in versification have helped establish our timeline of his works. Major chronologies of the plays and poems, including those of E.K. Chambers (1930), G. Blakemore Evans (Evans, 1974, rev. 1996), and Gary Taylor (1987), have drawn on what we know about the patterns of Shakespeare’s prosody to order his works. These three chronologies agree on the general shape of his literary output, placing Julius Caesar and Henry V at the midpoint of 38 plays so evaluated. They disagree, however, as to which year or years various works were written, as well as which came before or after others in the canon.

The present study offers a new chronology for Shakespeare’s plays based on an analysis of the most extensive data available concerning the structure of Shakespeare’s verse lines: the pause counts collected by Ants Oras (1960). We revise some of Oras’s numbers in light of new findings concerning attribution, narrowing Shakespeare’s portion of particular plays (Titus Andronicus, 1 Henry VI, Timon of Athens) and adding to our data set parts of four other texts (Arden of Faversham, Edward III, Sir Thomas More, and the Additional Passages to the 1602 Spanish Tragedy). This enhanced data set is then subjected to a constrained correspondence analysis (CCA), with various methodological modifications. The latter includes setting a range for Shakespeare’s literary output and fixing selected ‘anchor’ texts for the determination of dates for the remaining plays. These dates are then compared with the predictions from a principal component analysis (PCA) of new data concerning various linguistic features in Shakespeare’s verse (Tarlinskaja, 2014). We employ a bootstrapping procedure to establish a likely range for the composition of each work. Finally, in light of a theory advanced by J. Leeds Barroll (1991), we construct our timeline of Shakespeare’s plays by coordinating the CCA’s date predictions with periods when the playhouses of Shakespeare’s time were open for business.

2 Chronology: Background

Oras’s aim was to demonstrate historical and authorial patterns in iambic pentameter by tabulating where punctuated pauses fall within its first nine syllables (punctuation after the 10th syllable is not counted). Pauses can be counted in three different ways in Oras’s tabulation; he labels these A, B, and C pauses. A pauses are those signaled by punctuation of any kind within a pentameter line (Oras counts short lines, but not their terminal punctuation). B pauses, a subgroup of the A pause, are so-called ‘strong’ pauses within the line: those signaled by any punctuation mark other than a comma, including periods, question marks, colons, semi-colons, and dashes. C pauses are composed of punctuation marks dividing ‘split-’ or ‘shared’ lines. Oras counted A, B, and C pauses for 38 Shakespeare plays, adding the A and B pauses as well for Venus and Adonis, The Rape of Lucrece, and the Sonnets.

Oras presented his counts both in tables and in line graphs that display the percentages of pauses in the first nine syllabic positions of Shakespeare’s pentameter line. Because Shakespeare’s verse is iambic, typically with an unstressed syllable followed by a stressed one, pauses tend to come after the even syllables, making these graphs a virtual study in peaks and valleys. The changes across plays that they reveal have a clear significance for the study of chronology. For example, the changing pause patterns—the averages for each of the nine pause positions—in groups of plays traditionally identified with successive phases of Shakespeare’s career are strikingly different (Fig. 1). In works identified with the beginning of his activities as playwright in the early and mid-1590s, pauses cluster heavily after the fourth syllable (Fig. 1, left panel). As his career progresses, however, the distribution balances between the fourth and the sixth positions (Fig. 1, center panel). Toward the end of his time as a dramatist, the most significant proportion of pauses shifts toward to the sixth position, with a greater number in the second half of the line than the first (Fig. 1, right panel).

Fig. 1

Pauses in Shakespeare’s plays from three periods: early, middle, and late. Average percentage of pauses at each position is indicated by the black line; gray shading indicates 95% confidence intervals. Left panel: Titus, Shrew, 1 Henry VI, 2 Henry VI, 3 Henry VI, Richard III, and Two Gentlemen; Center panel: Much Ado, As You Like It, Julius Caesar, Hamlet, Henry V, and Twelfth Night; Right panel: Coriolanus, Tempest, Winter’s Tale, Cymbeline, Henry VIII, and Two Noble Kinsmen

The relevance of such data for chronologies of Shakespeare’s work has long been recognized (Bathurst, 1857). Because it is so comprehensive, Oras’s research was used for what we will call the ‘Oxford chronology’ (Taylor in Wells and Taylor 1987, pp. 69–144), which reproduces Oras’s A, B, and C pauses in separate columns opposite an ordered list of plays. In its discussion of various plays, the Oxford chronology refers to Oras’s data for confirmation of an estimated date or range. Yet, there is some divergence between the order implied by Oras’s counts and the order of the Oxford chronology. That is, 9 of the 38 plays ordered in the Oxford chronology share an exact position with the sequence that Oras’s A pauses suggest: Shrew, 3 Henry VI, Richard III, King John, Julius Caesar, Timon, Lear, Macbeth, and Kinsmen. Fifteen fall within two slots of each other in the Oras A order and Oxford chronology: 1 Henry VI, Errors, Loves Labors Lost, Richard II, Romeo, Dream, Merry Wives, 2 Henry IV, Much Ado, As You Like It, Measure, Othello, Pericles, Winters Tale, and Cymbeline.

Yet, almost as many plays, 14, are separated by three or more places in the two lists: Two Gentlemen of Verona, 2 Henry VI, Titus Andronicus, Merchant, 1 Henry IV, Henry V, Hamlet, Twelfth Night, Troilus, Alls Well, Antony, Coriolanus, Tempest, and Henry VIII. Of course, no test of any single linguistic feature—whether run on lines, feminine endings, or colloquialism in verse—should be expected to produce comprehensively satisfying results. This is particularly the case because so many factors, extrinsic and intrinsic alike, can affect the makeup of a literary text. At the same time, however, it seems significant that the Oxford chronology and the Oras data disagree to this extent. Some of the plays are quite divergent in their places: Oras’s counts for ‘first half’ pauses, for instance, would have us put Troilus seven places earlier than Oxford locates it, and Merchant seven places later; Titus and Antony are, by this measure, six places later in Oras, and both Two Gentlemen and Coriolanus four places later than in the Oxford chronology. Added to this puzzle is the extremely unlikely positioning, in the Oras data, of 2 Henry IV before 1 Henry IV, and of The Tempest before Pericles—chronological placements with which few if any scholars would be likely to agree.

The Oxford chronology’s use of the Oras data formed the basis of the most sustained examination to date of the relation between syntax and temporal ordering in Shakespeare: MacDonald P. Jackson’s ‘Pause Patterns in Shakespeare’s Verse: Canon and Chronology’ (Jackson, 2002). There Jackson describes Oras’s methodology and findings before submitting his A-pause counts to statistical analysis. Jackson compared A pauses among all plays, producing 1640 Pearson product moment correlation coefficients to indicate how close each text is, in terms of its pause patterning, to all the others. Listing each play separately, Jackson provides its five closest correlations in descending order, and notes that the results tend to confirm the accuracy of the Oxford chronology and support our traditional understanding of Shakespeare’s development.

Because his methodology emphasizes relation and proximity among plays, Jackson does not seek to establish new dates for them. Yet he acknowledges that his analysis produced correlations that diverge significantly from what the Oxford chronology would predict. Six plays come in for particular mention: The Merchant of Venice (which his results would place later than Oxford); Merry Wives (later than Oxford); 2 Henry IV (varied, but on the whole earlier than Oxford); Troilus (earlier than Oxford); Othello (earlier than Oxford); and Alls Well (later than Oxford). These differences seem important, not least because such divergence also characterizes Oras’s relation to the 1930 chronology of Chambers, the most authoritative chronology of the time. That is, Oras employed Chambers’s chronology but did not revise it, even though his own graphs and numbers challenged its order in numerous instances. The reluctance is understandable, for chronologies by definition have many working parts. Like received narratives generally, chronologies can be ‘sticky’ phenomena: something fixed through custom, and hard to dislodge (Kuhn, 1962).

3 Correspondence Analysis and PCA

To address the differing number of pauses in various texts, Oras quite understandably made them equal by converting pauses to percentages. But the plays (and parts of plays) vary greatly in the amount of data they offer. Thus, treating the shortest Shakespeare text in our sample (in this case, his contributions to the lightly punctuated Sir Thomas More, with a scant 32 pauses) as statistically equivalent to his most pause-heavy text (Cymbeline, with 2,735 pauses) emphasizes the former at the expense of the latter. Making their pause data equal 1, that is, imposes a statistical constraint on them both, and implies equal confidence in how representative their information is. Thus, the element of the Oxford chronology that looks to Oras for confirmation, as well as other studies based on percentages (Gray, 1931; Wentersdorf, 1951; Jackson, 1995), rely on artificially constrained evidence. How, then, to acknowledge the differential weight of the Oras-type data?

A valuable method for comparing compositional data—among other types of multivariate data—is correspondence analysis (or CA) (Hirschfeld, 1935; Benzécri, 1973; Hill, 1974; Greenacre, 2007). At its most basic, CA is a statistical methodology that takes categorical information and looks for associations, and strength of associations, in the relations of rows to columns in a contingency table (also known as a cross tabulation table, which contains frequencies of occurrence). Correspondence analysis first attempts to identify, and then rank, the most statistically significant variation in that data. Thus, the first CA axis will account for the largest amount of variation in the original data, the second axis will account for the next largest portion, and so on. By identifying the most crucial of these variables, researchers can visualize and interpret the bulk of the variability in any original data.

Germane to our purposes here is the extensive, and sophisticated, use of CA in seriation studies (Greenacre, 2007; van de Velden et al., 2009; Peeples and Schachner, 2012). Seriation, or simply ‘putting things in order’, is usually an exercise in relative dating, employed when an absolute dating method may be unavailable. In the field of archeology, for example, researchers are often confronted with artifacts that only occasionally have information regarding production or use attached to them. When carbon dating or information relating to, say, tree rings or chemical composition is unavailable, archaeologists have refined the statistical bases of CA to help them order, and thus date (however approximately), things of uncertain origin. By comparing the known composition of pottery remains, for instance, archeologists may place certain assemblages closer together in time based on how similar they are to one another.

Shakespeare’s plays may not strike one as artifacts, of course, but the procedures of CA as refined for seriation nevertheless provide a statistically rigorous methodology for examining their material components. On the basis of such an examination, in fact, we can offer a provisional chronology that responds to the differential distribution of features—in this case, pauses as recorded in the verse of early texts—throughout the canon and various plays with sections attributed to Shakespeare. This chronology should be understood as a provisional timeline of when the iambic pentameter in the play texts under examination was mainly composed. By emphasizing this last phrase, we mean to call immediate attention to two things. First, Oras’s prose data come from the pentameter verse in plays that are sometimes made up heavily of prose. (This is particularly the case in the late 1590s). Pause pattern analysis is therefore limited by the amount of verse in each play. Second, there is a strong likelihood that a number of Shakespeare’s plays were written at one time (even over various times) and revised at another, or others. A play title thus need not connote an event, but may sometimes have been multiple events, or even a process. The fixity of any ordering, then, needs to be qualified in the context of diachronic composition.

Like Taylor and Jackson, we focus on Oras’ A pauses, though after deliberation have subtracted values for shared lines (his C pauses). Our rationale for this comes from the uneven distribution of shared lines across the canon, and our sense that this different kind of writing—with a line-ending full stop built into its very structure—needs to be measured separately (Reinhold, 1942). We have also performed original tabulations of A (minus C) pauses for various texts, and parts of texts, not included or not disaggregated in Oras. These include parts of the collaborative plays Titus Andronicus, 1 Henry VI, Edward III, Timon of Athens, Arden of Faversham (scene 8), Sir Thomas More (the Hand D passages), and the Additional Passages to the 1602 Spanish Tragedy. These counts appear in Appendix 1, below, with information regarding the source texts and passages analyzed in Appendix 2.

We first performed two statistical procedures on our resulting data set. These are, respectively, a PCA and a CA. We would note that these procedures draw on distinct data: for the PCA, we used proportional values of pause abundance (so, for each play, the total proportion of the nine types sums to 1); for reasons already mentioned (and discussed further below), we used raw counts for the CA. We performed these and all subsequent analyses in R (R Core Team, 2014), using the ca package (Nenadic and Greenacre, 2007). (Note that all of the code used to produce the analyses and graphs in this manuscript is available from https://github.com/genevievekathleensmith/shakespeare-chronordination.)

The first Principal Component (PC 1) captured 62.3% of the total variation in pause composition among plays, mostly capturing variation in the 4th, 6th, 7th, and 8th position pauses (Fig. 2a, looking at how far each arrow travels left-right—that is, along PC 1—those arrows reach the farthest). For its part, PC 2 accounted for another 18.1% of the total variation, mainly reflecting variation in the 4th, 7th, 2nd, and 8th positions; again, one may notice far up or down each arrow goes to gauge its contribution (Fig. 2a). The patterns are broadly similar in the CA. The first CA axis comprised 77.4% of the total variation in pause counts among plays, and CA 2 accounted for a further 8.9%. Yet, now most of the arrows are more closely aligned with the horizontal plane (Fig 2b). This means that changes in pauses among plays are more well represented by a single axis in the CA, rather than in the PCA. We can additionally map the PCA and CA results onto the same dimensions to compare how they are distributed relative to one another (Fig 2c and d). The circles in Figures 2c and d have been scaled to reflect the relative weights of the data in each analysis. In PCA, each play contributes equally to determine the major axes of variation, while in CA, plays with low total pause counts (those that appear as slightly smaller circles) contribute less to the analysis than those with more pauses (which appear as larger circles). This is one advantage of CA: as opposed to PCA, it allows one to recognize differences in the amount of data contributed. One can interpret the positions of the play points by looking back at the axes of variation in Figures 2a and b. Texts are positioned in relation to where pauses occur in their pentameter lines, and in what proportion. For example, plays like Romeo and Juliet and Richard III have relatively more pauses early on in lines (i.e. 1st through 5th position), while plays like Macbeth and Tempest have more late-position pauses (6th through 9th).

Fig. 2

Results of PCA and CA as applied to pause variation in Shakespeare’s plays. (a) The contribution of each pause type to PC Axis 1 and PC Axis 2. (b) The contribution of each pause type to CA Axis 1 and CA Axis 2. Arrow direction indicates whether each pause type’s contribution is positive or negative, and arrow length indicates strength of each pause type’s contribution to that axis. (c) The projection of each play on PC Axis 1 and 2. (d) The projection of each play on CA Axis 1 and 2. The size of each point represents the weighting assigned to each text in the analysis. The same five titles have been highlighted in each plot. See Table 1 for the full list of titles included in these analyses

4 Bootstrap Methodology

On their own, our pause counts cannot give us any estimate of how certain (or uncertain) we are of the relative ordering obtained by our CA. To understand how small differences in our observed data may influence the outcome of our analysis, we employ a ‘bootstrap’ method. This is commonly used across a variety of disciplines, including in archeological studies (Ringrose, 1992). The bootstrapping procedure is a method of resampling. In our case, that means taking random samples of pauses from each play and rerunning our CA using the new values (Lockyear, 2012; Peeples and Schachner, 2012). Because we sample with replacement (meaning some pauses from the original data will be sampled more than once, and others not at all) the new counts of each pause type will vary slightly from the original counts. We repeat this resampling 1,000 times, repeating the CA with each new set of counts. This affords us some measure of uncertainty for our CA scores, and allows us to estimate 95% confidence intervals for the results in two dimensions (Fig. 3). The resulting confidence intervals produce a polygon for each play and trace a gradual arc up and to the right. We have called out five canonical texts–Richard III, Romeo and Juliet, Hamlet, Macbeth, and The Tempest—to show how these data reveal the chronological progression of Shakespeare’s works.

Fig. 3

Bootstrapped Correspondence Analysis of Shakespeare’s plays. Each polygon indicates the 95% confidence bounds calculated from 1,000 randomized bootstraps. Note that we have reversed the sign of the CA Axis 1 scores, for the purpose of maintaining a left-to-right chronological sequence. Since the sign of the values is arbitrary, this does not affect our timeline predictions. The same five titles have been highlighted as in Figure 2

This ‘arc’ of Shakespeare’s verse pauses also provides a basic template for understanding the syntactical development of his contemporaries’ iambic pentameter. Plotting the pause profiles of six contemporary playwrights over and against those of Shakespeare’s works (reproduced from Fig. 3, shaded light gray) to compare their pause content, we can see their plays tracing the general movement plotted by his verse, from Kyd and Marlowe through Fletcher. Jonson’s less iambic practice (Fig. 4, bottom row, center) identifies him as an exception. We should note that Marston’s polygons fall almost entirely within Shakespeare’s, a result that is not surprising given the fact that Marston began and ended his career as a playwright while Shakespeare was still working, and appears to have fashioned his plays (including the Antonio plays, and The Malcontent) strongly in response to the senior playwright’s (Cathcart, 1997).

Fig. 4

Bootstrapped Correspondence Analysis for Shakespeare and six contemporaries. Each polygon indicates the 95% confidence bounds calculated from 1,000 randomized bootstraps. In each panel, Shakespeare’s plays are indicated in light gray, while the corresponding author is illustrated in dark gray. See Appendix 1 for titles and data

5 CCA and Shakespeare’s Chronology

Correspondence analysis, we should point out, provides relative ordering: object X most likely comes before, or follows, object Y, at Z distance. Not being content with a relative chronological order, we were interested in using external evidence to suggest more specific determinations for the plays. An extension of CA called ‘constrained correspondence analysis’ (CCA), allows us to do just that (Groenen and Poblome, 2003; van de Velden et al., 2009). By incorporating such information as interval constraints for Shakespeare’s career, as well as hypothesized dates for some plays, and upper and lower limits for others, we can constrain the calculation of the CA scores (van de Velden et al., 2009). As we saw earlier, CA produces only a single score for each play. The same is true of CCA. We opted, therefore, to employ a bootstrapping procedure again, which allowed us not only to estimate exact dates for each play, but also to generate confidence intervals around those estimates (Peeples and Schachner, 2012). In this manner, we were able to produce a revised chronology of Shakespeare’s plays, using only interval constraints, a few dates, and the pause-position data itself. For this procedure, we modified MATLAB code (MATLAB, 2011) from van de Velden (2008) and wrote additional procedures to implement the bootstrapping (again, available from https://github.com/genevievekathleensmith/shakespeare-chronordination).

To constrain our correspondence estimates, we assigned numerical values to three plays for which plausible dates could be advanced: 3 Henry VI, last quarter of 1591 (= 1591.75); Henry V, middle of 1599 (= 1599.5); Pericles, first quarter of 1607 (= 1607.25); and fixed Tempest as after 1611 (>1611.0). Scholars could argue over these designations, of course, and other plays and dates could have been employed; these seemed among the more reasonable of our options. In addition, we set upper and lower bounds on the extent of Shakespeare’s writing career, demarcating it from 1589.5 to 1614.0. While these boundaries are also open to debate, it seemed to us that they are defensible so long as they are understood to be judgments rather than facts.

In CCA, the relative positioning of an object (in this case, a text) is made concrete through the addition of specific information regarding other objects in the timeline (Fig. 5). Thus, Macbeth is assigned a date of 1606.2, or March of 1606, as a manifestation of its statistical distance from all the other plays, using 3 Henry VI, Henry V, and Pericles to orient the chronology in time. As mentioned, we constrained the composition of The Tempest, placing it no earlier than 1611. Obviously, the accuracy of such a ‘forcing’ method depends in part on the soundness of these anchors (van de Velden et al., 2009), yet the procedure has the advantage of providing a specific date, rather than relative position, for each play.

Fig. 5

Illustration of CCA approach, redrawn (with modifications) from van de Velden et al., 2009. Macbeth is dated using three ‘anchor’ plays (indicated along the diagonal) as constraints. We also have illustrated the lower limit placed on the possible dates for Tempest. The relationship between CCA score and time is used to predict dates for the remaining plays

By bootstrapping this CCA, we were able to estimate confidence intervals for each play’s predicted date of composition using the 0.025th and 0.975th quantile of the bootstraps (Fig. 6). For some plays, our prediction falls earlier than the Oxford date (e.g. Troilus and Cressida, As You Like It); in other cases it is later (e.g. Coriolanus, The Winter’s Tale).

Fig. 6

Predicted dates of composition resulting from CCA of pause counts in Shakespeare’s plays. Filled circles represent the average predicted date over 1,000 bootstraps, horizontal black bars indicate 95% confidence intervals. Vertical gray shading indicates periods when London’s playhouses are thought to have been closed. Shifts made to our predictions in light of these closings are indicated with open circles

6 Adjusted CCA Chronology

To use our CCA results to produce a hypothesized timeline, we further wanted to consider historical data available on the disposition of the Elizabethan playhouses (Chambers, 1930; Harrison, 1941, 1955; Barroll, 1991). J. Leeds Barroll has posited that Shakespeare would have reduced his dramatic writing or ceased writing new plays altogether when the playhouses were closed, largely owing to the plague (Barroll, 1991). Such occasioned the years 1592–94, when, it is commonly thought, Shakespeare penned Venus and Adonis and The Rape of Lucrece while the public amphitheaters were closed for business. Because a noticeable spread in the data appears after 3 Henry VI (as well as prior to The Merry Wives), we have elected to adjust our estimates for the dates of composition on the assumption that Shakespeare turned his energies to his two narrative poems during the closure of the playhouses (Fig. 6).

During the drafting of this essay, we gained access to new data: the metrical and linguistic tables of Marina Tarlinskaja’s recent monograph (Tarlinskaja, 2014). There Tarlinskaja provides a massive tabulation of original prosodic data for numerous playwrights, including Shakespeare. The Shakespeare data, however, do not include various prose-heavy plays (such as Merry Wives, Much Ado, As You Like It, and Twelfth Night), or two collaborative works (Sir Thomas More, Timon). Further, Tarlinskaja offers only partial data for Edward III, so we chose to omit it, leaving a total of 34 texts, with over 1,500 pieces of data regarding 45 stylistic and structural categories. These categories include data on strong and weak syllabic positions, respectively; word boundaries (total and after positions); strong syntactic breaks; run on lines; proclitic and enclitic stresses; syllabic -ed and -eth; disyllabic -ion; grammatical inversion; meter-sense; syntactical run ons; feminine endings (total); feminine endings built by simple and compound constructions, respectively; and alliterative lines. Because Tarlinskaja’s data are rendered largely in percentages, we could not apply a CA or CCA. Thus, with Tarlinskaja’s permission, we ran a PCA on these new data and used PC 1 to estimate approximate dates of composition (Table 1).

View this table:
Table 1

Dates from received (Taylor, 1987; Evans, 1996; Brainerd, 1980) and original chronologies (PCA of Tarlinskaja data; Bruster–Smith bootstrap mean, and Bruster–Smith final), compared. Key to abbreviations following

PlayOxford dateRiverside dateBrainerd dateDate predicted by PCA of Tarlinskaja dataBruster–Smith bootstrap mean date predictionBruster–Smith final prediction
  • Notes: 1H4 1 Henry IV; 1H6 1 Henry VI; 2H4 2 Henry IV; 2H6 2 Henry VI; 3H6 3 Henry VI; ADO Much Ado About Nothing; ANT Antony and Cleopatra; ARD Arden of Faversham; AWW All’s Well That Ends Well; AYL As You Like It; COR Coriolanus; CYM Cymbeline; E3 Edward III; HAM Hamlet; H5 Henry V; H8 Henry VIII; JC Julius Caesar; JN King John; LLL Love’s Labor’s Lost; LR King Lear; MAC Macbeth; MM Measure for Measure; MND Midsummer Night’s Dream; MV Merchant of Venice; OTH Othello; PER Pericles; R2 Richard II; R3 Richard III; ROM Romeo and Juliet; SHR Taming of the Shrew; SPT Spanish Tragedy (Additional Passages); STM Sir Thomas More (Hand D); TA Titus Andronicus; TGV Two Gentlemen of Verona; TIM Timon of Athens; TMP Tempest; TN Twelfth Night; TNK Two Noble Kinsmen; TRO Troilus and Cressida; WIV Merry Wives of Windsor; WT Winter’s Tale.

We wish to emphasize that the results are purely our own, rather than Tarlinskaja’s, who should not be assumed to endorse any of the datings we adduce from analyzing her data. Further, our analysis of her data assumes the same intervals for Shakespeare’s career (1589.5 to 1614.0) as in the CCA. These estimated dates are provided in Table 1 below, in addition to dates from the Oxford chronology, the Riverside chronology, Brainerd’s multivariate chronology (Brainerd, 1980), the dates predicted by our CCA (averaged across 1,000 bootstraps), and our final predictions, which adjust the CCA dates in relation to playhouse closures and exigencies of composition. By the latter, we refer to adjustments that assume a working distance among Shakespeare’s texts—even though it is quite possible he may have composed various works simultaneously. The asterisks in the final two columns flag the anchor dates we provided (for 3 Henry VI, Julius Caesar, and Pericles, respectively), to distinguish these from predictions generated from analysis.

7 Dates and Discussion

The following discussion of our results begins with title, hypothesized date, and bootstrap range. In addition to the chronologies and tests outlined in Table 1, our discussion alludes to Langworthy’s tabulations of verse-sentence data (Langworthy, 1931), Reinhold’s research on shared lines (Reinhold, 1942), Waller’s doth/does and hath/has ratios (Waller, 1966), Slater’s tables regarding link words between Shakespeare’s plays (Slater, 1988), and Ilsemann’s data regarding speech word-length (Ilsemann, 2008). We have also consulted McManaway, 1950; Uhlig, 1968; and Schabert, 2000.

  • Titus Andronicus–1590.7 (1589.7–1591.7). As with other works, our adjusted CCA treats only the portion currently ascribed to Shakespeare (see Appendix 2 for breakdowns). Early dating concurs with the PCA of Tarlinskaja’s data, as well as with Slater’s order, and initiates a group that concludes with Edward III.

  • Arden of Faversham (Sc. 8)—1590.9 (1589.5–1593.9). Paucity of data makes dating difficult. PCA of Tarlinskaja places this before Titus, as Shakespeare’s first surviving writing.

  • The Taming of the Shrew–1591 (1590–1592.2). Our adjusted CCA concurs with Oxford and Brainerd. We believe this play was written before the closing of the theaters, and preceded A Shrew. Waller’s doth/does and hath/has ratios place Shrew, alone among the early plays, in the second half of the canon.

  • 1 Henry VI–1591.2 (1589.5–1593.3). Issues of collaboration complicate dating. Our CCA treats 2.4, 4.2–4.5; CCA of the complete text places it only marginally later, at 1591.9.

  • 3 Henry VI–1591.75* (1591.75*). Our adjusted CCA’s first anchor, fixed at a position supported by Oxford, Riverside, Brainerd, the Tarlinskaja PCA, and Ilsemann.

  • Edward III–1592 (1590.5–1593.2). Our adjusted CCA treats 1.2, 2.1, 2.2, 4.5–4.9 (Elliott and Valenza, 2010). Probably written before the closing of the theaters, but Slater’s rare-word linkage with Lucrece could suggest the hiatus or immediately after the reopening. CCA places 1.2, 2.1, 2.2, and 4.4 at 1591.8.

  • 2 Henry VI–1593.8 (1591.8–1593.8). Stylistically, 2 Henry VI postdates 3 Henry VI, most likely owing to revision at a much later date. (We see such signs most obviously in 2.1.1-4.1.147 and 5.1.1-end, though there are indications of revision throughout). Its position here likely averages writing from earlier and later in Shakespeare’s career. Placement after 3 Henry VI is replicated in Brainerd, the Tarlinskaja PCA, Langworthy, Reinhold, and Ilsemann. First in a group running through Richard II.

  • The Two Gentlemen of Verona–1594.2 (1591.7–1594.3). Our adjusted CCA (adjusted to recognize the availability of the theaters) is close to Riverside. Slater’s rare-word list ranks it 7th of the plays, Reinhold’s data 9th–just prior to Loves Labors Lost—and Ilsemann’s just after Love’s Labor’s Lost. Langworthy supports Oxford in seeing it as a very early play.

  • The Comedy of Errors–1594.4 (1592.1–1594.7). Our adjusted CCA matches well with Oxford, Riverside, and Brainerd, and anticipates the recorded performance at Gray’s Inn 28 December 1594. Cross-genre rare-word links (Slater) suggest proximity to Richard III, as does Ilsemann’s data.

  • The Additional Passages to The Spanish Tragedy (1602)–1594.5 (1589.5–1598). Paucity of data here, reflected in the bootstrap range, lessens our confidence in this dating. PCA of the Tarlinskaja data puts them much later, as would Ilsemann’s data (after Henry V).

  • Romeo and Juliet–1594.6 (1592.7–1594.8). This first of the lyric plays perhaps finishes earlier owing to its formal, amatory verse (our unadjusted CCA places it in mid-1593). Our adjusted CCA squares with the PCA of Tarlinskaja’s data.

  • Loves Labors Lost–1594.8 (1592.6–1595) Our adjusted CCA squares with Oxford, Riverside, and the PCA of Tarlinskaja’s data.

  • Richard III–1595 (1593.2–1595.1). Our adjusted CCA agrees with the PCA of Tarlinskaja’s data and with Brainerd in positing a later date for this text than Oxford or Riverside.

  • A Midsummer Nights Dream–1595.3 (1593–1595.6). Our adjusted CCA agrees with Oxford, Riverside, Brainerd, and the PCA of Tarlinskaja’s data.

  • Richard II–1595.5 (1593.7–1595.8). Dated 1595 in four of the chronologies, with the PCA of Tarlinskaja’s data only a year later.

  • King John–1596.1 (1594.3–1596.6). Crucial since Honigmann’s argument for an extremely early composition (Honigmann, 2000). Our positioning agrees with Oxford and Riverside. Brainerd has a later date; PCA of the Tarlinskaja data an earlier one, though after Romeo and Dream; Langworthy locates immediately after Richard II.

  • As You Like It–1597 (1594.8–1597.5). This first of the prose-heavy plays (with fewer pauses for analysis) suggests a much earlier date than is conventionally accepted. Brainerd (1600.3) better accords with Oxford and Riverside; Tarlinskaja has no data for it. Shares unexpected extra-generic rare-word links with Richard III and Romeo, in addition to Henry V. For arguments concerning a date before 1599, see Knowles, 1977, pp. 340-49, 365-67. Stylistically first in a group running through Troilus.

  • The Merchant of Venice–1597.2 (1595.5–1597.7). Both Brainerd and PCA of the Tarlinskaja data locate it in 1598. Barely distinguishable, statistically, from Much Ado, it finishes slightly earlier than that text on the bootstrap procedure.

  • Much Ado About Nothing–1597.35 (1595–1598.5). Our positioning is slightly earlier than Oxford and Riverside.

  • 1 Henry IV–1597.50 (1595.5–1598). Our dating accords closely with Oxford, Riverside, and Brainerd.

  • 2 Henry IV–1597.8 (1595.8–1598.5). We believe this play immediately followed 1 Henry IV. Our dating accords closely with Oxford and Riverside.

  • Julius Caesar–1598.2 (1596.2–1598.3). Seen by Platter in September of 1599. Brainerd has it at the end of 1598, PCA of the Tarlinskaja data in late 1600. Shares rare-word links with 2 Henry IV, Troilus, and Hamlet.

  • Troilus and Cressida–1598.4 (1596.2–1598.5). Typically dated some 2- to 4 years later. Composition in 1598 would locate it near the completion and publication of Chapman’s Homer, but prior to the War of the Theaters, Gilbert’s De Magnete, and Nashe’s Summers Last Will and Testament, which are sometimes seen as implicated in its language. Brainerd and the Tarlinskaja PCA would place it near Oxford and Riverside; Langworthy with Henry V and before Hamlet; the Reinhold data after Twelfth Night but before Othello and Hamlet.

  • The Merry Wives of Windsor–1599 (1595.4–1602.7). Our dating comes later than Riverside, Brainerd, and Langworthy. As is evident in the wide bootstrap range, scant pause data makes placement less certain. Groups stylistically with Hamlet and Henry V.

  • Hamlet–1599.25 (1597.9–1600). Statistically close to Henry V, this play’s references to Julius Caesar suggest that it followed the Roman tragedy. Our dating is slightly earlier than Oxford and Riverside; both Brainerd and the Tarlinskaja PCA have it later than those authorities. Several passages in the Folio text hint at revision during the early 1600s.

  • Henry V–1599.5* (1599.5*). Second of our anchor texts. Choruses almost certainly composed prior to Essex’s disastrous return from Ireland on September 28. A date in late summer would just enable it to inspire certain passages in 1 Sir John Oldcastle, completed by 16 October. Shifting this anchor a year later would retain the order of the middle plays but bring them closer to conventional dating.

  • Twelfth Night–1600.8 (1598.9–1602.3). Close match with Oxford, Riverside, and Brainerd. Attaches to Othello stylistically.

  • Othello–1601 (1599.5–1602). Our CCA locates Othello earlier than Oxford, Riverside, Brainerd, and the Tarlinskaja PCA. Echoes in Q1 Hamlet indicate that it was already extent, even familiar, by 1602 (Honigmann, 1993). Langworthy has it before Henry V.

  • Measure for Measure–1602.2 (1600.2–1602.9). Like As You Like It and Troilus, Measure finished earlier in the CCA than in all other chronologies. Commonly thought of as a Jacobean play owing to various of its themes, Measure features very little that dates it certainly. Pushing it into 1603–04 might imply correspondingly later dates for texts immediately prior and following. Revisions by Middleton may skew the results here.

  • Sir Thomas More Additions–1602.8 (1596–1610.3). This dating squares with Oxford; a seventeenth-century date has been argued as well by Jackson (Jackson, 1978, 2006, 2007).

  • Alls Well That Ends Well–1604.3 (1602.5–1605.4) Our CCA squares with Oxford but is later than Riverside. Brainerd and the Tarlinskaja PCA suggest a later date (1607.2 and 1607.37, respectively), as does Reinhold and Langworthy’s data, both of which put this play after Lear and before Macbeth. Jackson argues for 1606.5 or later (Jackson, 2001).

  • King Lear–1605 (1603.4–1606). Oxford suggests 1605–06, Riverside 1605, Brainerd 1606.2, and the Tarlinskaja PCA 1607.91. Our location has it preceding the second edition of King Lear (May of 1605).

  • Timon of Athens–1606.1 (1604.3–1607). This collaboration with Thomas Middleton is obviously connected to King Lear, with which it has the highest number of rare-word links (Slater). Our CCA of the parts attributed to Shakespeare puts it in the year after Lear, attaching stylistically to Macbeth.

  • Macbeth–1606.3 (1604.5–1607.4). Our CCA agrees closely with Oxford, Riverside, and Brainerd. The Tarlinskaja PCA places it two years later.

  • Pericles–1607.25* (1607.25*). The third of our three anchor texts. The date provided here comes prior to its entry in the Stationers’ Register (20 May 1608). The PCA of the Tarlinskaja data puts it just before 1607, in a cluster with All’s Well, Lear, and Macbeth.

  • Antony and Cleopatra–1610.5 (1609.2–1611.9). Entered in the Stationers’ Register 20 May 1608. This later CCA date may indicate revision. Brainerd and the Tarlinskaja PCA place it in 1608 and 1609, respectively. Slater’s rare-word catalog links it not only with Macbeth and Coriolanus, but also Cymbeline, Winters Tale, and Tempest. Groups stylistically with Tempest, Coriolanus, and Kinsmen.

  • The Tempest–1611 (>1611). Last in Brainerd’s ordering; later, and also last in the Tarlinskaja PCA (1614) and Slater’s rare-word ranking (Slater, p. 99). Our dating coincides with Oxford and Riverside.

  • Coriolanus–1611.6 (1610–1612.9). As with the other late plays, our CCA produces a later-than-conventional date. The Tarlinskaja PCA suggests 1610.6. Slater’s rare-word index links it most tightly with Cymbeline and Winters Tale. Langworthy places it after Antony.

  • The Two Noble Kinsmen–1611.9 (1609.9–1613.9). This dating is at least a year earlier than Oxford (1613–14) and Riverside (1613). Brainerd (who does not separate according to collaboration) has it much earlier (1605.5).

  • Henry VIII–1613.1 (1611.2–1614). According to our CCA, this play follows Kinsmen but precedes two of the romances. Certainly composed prior to late June 1613, when it was performed at the Globe. The PCA of the Tarlinskaja data puts it at 1613.22. Of the late plays, it has statistically significant rare-word links with only Winters Tale. Attaches stylistically to Winter’s Tale and Cymbeline.

  • Winters Tale–1613.3 (1611.7–1614). Seen by Forman, May 1611. Brainerd’s 1609.4 accords with Oxford, and just precedes Riverside. The PCA of the Tarlinskaja data places this play at the end of 1612. Langworthy has Winters Tale and Cymbeline as two of his last three texts (with Henry VIII).

  • Cymbeline–1613.5 (1612.1–1614). Seen by Forman in 1611. The Tarlinskaja PCA has this text around the middle of 1611. Slater’s rare-word links group it with Winters Tale, Tempest, and Coriolanus. Our CCA suggests it is Shakespeare’s final surviving writing.

8 Conclusion

The preceding chronology suggests a timeline for the composition of the pentameter verse in Shakespeare’s plays as represented in the First Folio of 1623 and various early printed editions (Fig. 7). In doing so, it makes at least four assumptions: first, that Oras’s counts accurately record the pauses in the Folio texts it takes up (this was spot-checked, and confirmed in multiple instances); second, that Shakespeare’s verse line developed in one direction, and regularly, without significant deviation; third, that, on the basis of pause data, a CCA may accurately order the plays; and fourth, that the dates for the three textual ‘anchors’ have validity (van de Velden et al., 2009). The soundness as well as particulars of our chronology could be called into question by refuting or qualifying any of these assumptions, and we welcome research that does so in the service of expanding information about Shakespeare’s compositional practice. Prior to such interventions, however, we would point out the general agreement between our ordering and the established timeline of the plays, as well as its confirmation of various independent challenges to the conventional chronology. Further, although we advance much earlier dates for several texts (including As You Like It and Troilus and Cressida), and later dates for many of the plays following Pericles, in no instance does our adjusted CCA locate a title chronologically after its appearance in print.

Fig. 7

Comparison of proposed Bruster–Smith chronology with standard dates (e.g. Oxford et al.). Open circles indicate standard, previously-held date, arrows the shifts proposed to the timeline. Arrows pointing right indicate plays for which later dates are proposed; arrows to the left indicate ones for which an earlier date of composition is suggested. As in Figure 6, vertical gray shading indicates periods when London’s playhouses are thought to have been closed

Our ordering is closer to the Oxford/Riverside chronologies than are Brainerd or the PCA of the Tarlinskaja data, yet we differ from Oxford/Riverside on a number of plays in addition to those mentioned above. For example, our adjusted CCA places Two Gentlemen significantly later than Oxford, Richard III later than both chronologies, the Sir Thomas More passages much later than Riverside, and Julius Caesar, Hamlet, and Othello earlier than their conventional dating—Othello significantly so. We have seen, with 2 Henry VI, that a text may date differently from a performance owing to revision. It is important to keep in mind that the chronology offered here concerns the time when the pentameter in these texts was largely composed. If we are correct in situating various of the plays that follow Pericles later than the conventional dating, it would suggest that Shakespeare was actively elaborating their scripts immediately prior to or during his retirement from writing for the theater.

Appendix 1 Pause Data

Pause data employed in this study, A-C pauses from Oras (1960) and hand counts. We have emended what appears to be a typographical error in All’s Well C-count, position 8, from 23 to 3 (p. 80).

1 TamburlaineMarlowe2659401941198026194
2 TamburlaineMarlowe28746126914210543191
Jew of MaltaMarlowe13211893339212157973714
Dr. FaustusMarlowe1626296457371370
Edward IIMarlowe632118448618225962192
Spanish TragedyKyd65144613611459950215
Soliman and PersidaKyd2761362761289519131
Edward IIIKyd9222015279383261
Edward IPeele7442516261511444
Titus AndronicusPeele175123134737222101
Antonio & Mellida 1–2Marston15623510258241446416213653
Jack Drum’s EntertainmentMarston29583523015514451135
Dutch CourtesanMarston928892497641243
What You WillMarston206538183138136563819
Insatiate CountessMarston276839253215238953112
Tale of TubJonson14620813960246954338726194
Case Is AlteredJonson4186292372142441415313
Every Man InJonson365517111807964176
Every Man OutJonson43592210080110923813
Cynthia’s RevelsJonson656937154139129844711
Henry 8Fletcher23925723321327534717835
Two Noble KinsmenFletcher29526024121630630315943
Titus AndronicusaShakespeare45885228114816545113
Comedy of ErrorsShakespeare35973422914015663252
1 Henry 6aShakespeare162076141301831
1 Henry 6bShakespeare8018371474304203127493
3 Henry 6Shakespeare138205143706309347148713
Taming of the ShrewShakespeare631175140022119655317
Richard 3Shakespeare13830512068043044518610820
Two Gentlemen of VeronaShakespeare9713876308172222763912
Love’s Labor’s LostShakespeare42693027315416654308
Midsummer Night’s DreamShakespeare358843300229172833312
Romeo and JulietShakespeare4415765567323319965919
Richard 2Shakespeare61182905683233751307914
King JohnShakespeare79146574413152871336524
Merchant of VeniceShakespeare396642272206212132249
1 Henry 4Shakespeare5190422481872041223113
2 Henry 4Shakespeare60149823402162591277918
Merry WivesShakespeare14201961345118158
Much AdoShakespeare17322590697836127
Henry 5Shakespeare2883502841982421465114
Julius CaesarShakespeare771141023763113661656121
As You Like ItShakespeare18373516511114948213
Twelfth NightShakespeare8372613610013860395
Troilus and CressidaShakespeare61122664392133361347819
Measure for MeasureShakespeare4188342431563171207614
All’s WellShakespeare936232051352791235415
King LearShakespeare401035535414344818312540
Antony and CleopatraShakespeare3910568360293554299255107
Winter’s TaleShakespeare4413878380261492307378133
Henry 8aShakespeare24342515613526120012261
Two Noble KinsmenaShakespeare1034121111091861299232
Edward 3aShakespeare8241313363582543
Arden of FavershamaShakespeare11130513020
Spanish Tragedy Add. PassagesaShakespeare2028123332301482
Sir Thomas More PassagesaShakespeare000968900
  • aCollaborative works for which we provide counts for only Shakespeare’s portion(s), as listed in Appendix 2.

  • bIn the case of 1 Henry VI, we have also included the total pause counts for the entire play.

  • Appendix 2 Source Texts and Play Breakdowns

    Because Oras’s counts are based on a facsimile of the 1623 First Folio, our hand counts of Titus and Timon used the Oxford Text Archive’s versions of those plays in Folio form. We used EEBO texts for Faversham (1592; STC 793) and Edward III (1596; STC 7501); LION for the Additional Passages to the Spanish Tragedy (1602; STC 15089); and the Riverside Shakespeare for the Hand D material in Sir Thomas More.

    Breakdowns for Shakespearean portions of the hand-counted collaborative plays, determined in part by consulting Vickers, 2002, 2007, are as follows: Titus Andronicus: 2.3–3.2, 4.2–5.3; Arden of Faversham: scene 8; Edward III, conventional Shakespeare attribution: 1.2, 2.1, 2.2, 4.4; Edward III, per Elliott and Valenza, 2010: 1.2, 2.1, 2.2, 4.5–4.9; 1 Henry VI: 2.4, 4.2–4.5; The Spanish Tragedy (1602): Additional Passages 1–5;Timon of Athens: 1.1., 2.1–2.2, 4.1, 4.2.1–29, 4.3.1–457, 5.1–5.4.


    View Abstract