83Footnote: a question of authorship

In his article, "King John Divided" (Literary and Linguistic Computing 19.2 [2004]: 181-95), Thomas Merriam argues that King John was written by two authors, one of whom displayed use of "stylistic markers" typical of Shakespeare, while the other's passages are clearly different. Merriam uses a number of statistical tests, employing "function words" (common words like"and," "I," "is," and so on) and other ways used to differentiate Shakespeare's usage from other writers. He refers to work by Gary Taylor in the Textual Companion to the Oxford Works (1986), and Elliott and Valenza (1996, 1988-89, 2001).

84 The core of Merriam's conclusions is summed up in a table that lists passages by Shakespeare and "Non-Shakespeare" (184). An unexpected feature of this table is that it omits almost a third of the play: according to Merriam, 751 lines are by Shakespeare (28%), and 1156 lines by "non-Shakespeare" (42%). This leaves a total of 820 lines (30%) in a logical paradox, neither by Shakespeare nor not by Shakespeare. The table reads thus in simplified form--Merriam adds columns to provide act and scene lines from the Oxford and Arden editions, and lists potential sources for the passages; I have added a list of the passages his table omits, plus columns that add up the various totals and percentages.

TLNs start	TLN finish	# of lines	Authorship	Total lines	Percent
146	290	145	Shakespeare
407	503	97	Shakespeare
1569	1850	282	Shakespeare
1995	2164	170	Shakespeare
2192	2248	57	Shakespeare	751	27.54%
1	145	145	Non-Shakespeare
291	406	116	Non-Shakespeare
506	737	232	Non-Shakespeare
997	1193	197	Non-Shakespeare
1851	1994	144	Non-Shakespeare
2249	2570	322	Non-Shakespeare	1156	42.39%
738	996	259	Nobody
1194	1568	375	Nobody
2165	2191	27	Nobody
2571	2729	159	Nobody	820	30.07%
				2727	100%

85 Statistical tests of the kind Merriam is employing must rely on sufficiently large chunks of data if they are to be meaningful. The table indicates that two of the passages attributed to Shakespeare are less than 100 lines in length (one, attributed to the paradoxical "Nobody" is only 27 lines long). The number of words in these passages is well under a thousand; Hugh Craig has established a block of 2,000 words as a statistically minimal sample of text for analysis of common words (Ellegård argues for a safer 4,000; qtd Craig and Kinney 27). Merriam's selection of smaller chunks, together with the omission of almost a third of the text, suggests that his statistics are vulnerable to precisely the kinds of normal variation that makes the samples invalid in statistical tests. Craig has run his own battery of tests on the two samples of "Shakespeare" and "Non-Shakespeare" from King John (but not on the missing "Nobody" samples, since they were not provided). His conclusion was that his tests do not give "any decided direction on the problem. They did not confirm the hypothesis, but they do not disconfirm it either" (private response to Merriam, 2009).

86 In a more recent piece, "Feminine Endings in King John" (2009), Merriam returns to the fray with evidence from the frequency of feminine endings in the two strands of the play he has identified as "Shakespeare" and "not Shakespeare." In this piece the whole of the play is included, with a new table. Once again the play is divided into sections, some of which are clearly too short to be amenable to reliable statistical analysis:

87 Merriam's contention that many plays in the period were collaborative is unquestionably accurate--the earlier plays on Henry VI almost certainly involve some degree of collaborative authorship. However, his division of King John into multiple, relatively short components, attributing these in part to two different hands, does not provide clear evidence of multiple authorship.