Mullen, Roderic L. The New Testament Text of Cyril of Jerusalem. The New Testament in the Greek Fathers, no. 7. Atlanta: Scholars Press, 1997. ISBN: 0-7885-0339-1. Pp. xiv+431. US $39.95. 1. This expanded dissertation has a goal of analyzing the textual affinities of the New Testament quotations made by the fourth century Bishop Cyril of Jerusalem. An attempt is made to sketch the evolution of the NT text in Roman Palestine by comparing what we learn of Cyril with past results obtained for Origen and Eusebius of Caesarea and Epiphanius of Gaza and Salamis. In summary, Cyril's NT text is mixed and varies from book to book but is related primarily to the later Alexandrian witnesses and secondarily to the Byzantine witnesses. His text of Mark is related to the Koridethi Gospels (ms [Theta]) and the miniscules 13, 28, and 565, as also are the last seventeen chapters of Matthew. Overall, Cyril's text is less Alexandrian than that of Origen or Eusebius, and it is somewhat similar to that of Epiphanius in character. The Data 2. The quotations that Mullen uses for analysis are extracted from Cyril's Catechetical Lectures and the Procatechesis, his Epistle to Constantine, and the Sermon on the Paralytic. The Mystagogical Lectures are rejected as spurious, along with many other pseudonymous works that were easier to cull out. Those genuine works were all composed in the neighborhood of 350 C.E. The critical editions and the numerous manuscript sources for the texts are named by Mullen, but only one manuscript date is mentioned, that being of the tenth century. Cyril's citation habits are evaluated and found to be generally accurate and normally from a written source, not just from memory. 3. Mullen lists all the New Testament quotations found in the documents just named, giving the full text and reference, but without context. In addition, Mullen cites 250 ambiguous quotations, for which the biblical passage being referred to could not be uniquely identified. These are listed separately and not analyzed further. The identifiable quotations are classified as being either citations, adaptations, or allusions, and the adaptations are evaluated to see whether they supply any clear text-critical data. The citations and usable adaptations are collated against the texts of sixteen to twenty-six Greek and Latin manuscripts representing the familiar major text-types in the New Testament. Unfortunately, UBS3 and the TR are included among these control witnesses, even though they are modern productions of mixed origin. The collations identified about 1100 points of textual variation, of which 615 are used in the classification profiles. The points of variation discarded seem to have been singular and sub-singular variants, but Mullen fails to document just which points were used and which discarded, and I was unable to find or deduce the precise rule upon which that decision was based. This oversight makes it rather difficult to reproduce the statistics Mullen reports. The Analysis 4. The textual affinities of Cyril's NT quotations are gauged by two kinds of statistics, Colwell and Tune's quantitative method (Colwell 1969) and Ehrman's Comprehensive Profile Method (Ehrman 1986; Ehrman 1987). Mullen's conclusions are based only on the statistics, and not on any further analysis of the readings (he does comment, however, on several examples of Cyril's stylistic preferences, such as his fondness for the "vivid details" of Mark's gospel or his choice of readings that bolstered his own theological convictions). Colwell's Quantitative Method measures the agreement of Cyril's text with the text of several control witnesses chosen to represent the main landmarks of the textual tradition, expressed in terms of a percentage of agreement at the points of variation. In addition, Mullen averages together the percentages of agreement of Cyril with the various members of a text-type grouping to reveal overall affinities to the text-types. Ehrman's Comprehensive Profile Method measures several kinds of affinity of mss to text-type groupings by classifying readings with respect to a text-type as being a "Uniform" reading of the text-type, or likewise a "Predominant," "Distinctive," "Exclusive," or "Primary" reading of the text-type, and by counting percentages of agreement between ms and text-type within each of these classes of readings. Mullen is not able to draw much meaningful information out of Colwell's statistics that is not more clearly observable in Ehrman's statistics. His description of the two methods as being complementary is overly gracious to the older method; Ehrman has given us a substantially more powerful tool for analyzing mss. 5. Mullen consults a statistics textbook and a social scientist on the issue of how to choose a sample size large enough for his results to be significant but small enough to detect shifts in quality due to block mixture. Whatever he may have read or been told, the advice given seems to have been fundamentally muddled and misunderstood. His model of a binomial probability density for agreement counts is reasonable, but the numerical calculation on page 305 is incomprehensibly wrong (see the following paragraph for an example of such a calculation done correctly), and his concern about normal approximations to the binomial distribution is not to the point. The requirement he consequently adopts of a minimum of twenty points of variation to analyze in each block of text is much too small. This leads him, for instance, to divide the Pauline Corpus into blocks that yield too few points of variation to support reliable results from each block. 6. Here is a numerical example, a review for the Gentle Reader who may not be on intimate terms with binomial densities. If among 25 points of variation Cyril agrees with Sinaiticus, say, 15 times and disagrees 10 times, then the agreement rate from that sample is 15/25 or 60%. Our estimate for the true overall agreement rate of the two witnesses (both quoted and unquoted parts) would also be 60%, except that this 60% is a very fuzzy, uncertain measurement, because of the small sample size. If we require a confidence level of 95%, because we have compared the witnesses at only 25 points of variation, we can only claim that the true agreement rate is in the range 40-78%, a range that includes most of the agreement rates seen in Mullen's book. Even relaxing the level of confidence to 80% only lets us narrow the uncertainty range to 46-72%. Because Mullen treats percentage differences of 5-10% as significant, even when lacking the 100 or more samples that might justify it, numerous comments about these differences being "significant" or "clear" (e.g., that Cyril's agreement with witness x is significantly greater than with witness y) remain unconvincing. 7. The whole issue of accuracy and significance needs analysis, not guesswork. It might be argued that differences in percentages of agreement with different control witnesses are not really differences of independent binomial samples, but Mullen does not raise this issue or suggest any other model. Also symptomatic of his confusion about how much accuracy the figures support is the fact that percentages are always cited with a decimal point (e.g. 5/7 = 71.4%), even though most sample sizes are less than 100 and no statistic in the whole book has a standard deviation as small as 1%. 8. Despite weaknesses in his statistical understanding, the general tenor of Mullen's classification results is convincing, even if some specifics are less than solid. Cyril is shown to join Eusebius and Origen in citing a text of Mark that relates to the Caesarean manuscripts. Cyril's text shows no signs of affinity to Greek and Latin "Western" witnesses. In several books of the NT, Cyril's quotations prove the existence of an Alexandrian text outside of Egypt. In some other books, Cyril provides one of the earliest witnesses to a Byzantine form of text. Mullen's assumption that there was a single Palestinian text of the NT, the evolution of which can be traced through time from Origen through Eusebius and Cyril to Epiphanius, seems tenuous, particularly since these authors lived in five different cities, and Epiphanius does not even quote a Caesarean form of Mark. Improving Methods 9. One could regard both the averaged levels of agreement with text-type groups (from Colwell's statistics) and the percentage agreements with the Uniform and Predominant readings of the text-types (from Ehrman's statistics) as two heuristic attempts to measure something that could be more clearly modeled and measured another way. Imagine that behind each text-type lies a single ideal or archetypical text, which is only imperfectly represented in the extant witnesses belonging to the type. If so, then we might model our uncertain knowledge of that ideal text by means of a probability distribution, related to the voting consensus of the manuscripts. Based on that distribution, the statistically expected level of agreement between the testee and the ideal text can be computed. It is the average over all the points of variation of the probability that the testee and the ideal text agree at that point. This expected level of agreement would improve on Ehrman's Intra-Group profiles by lumping together the Uniform and Predominant readings with a weight factor that reflects the degree of predominance of each reading. It would also improve on the averaged levels of agreement from Colwell's statistics by deemphasizing the sporadic group defections that are not practically important and are overweighted by this kind of averaging. 10. Ehrman's profiles of agreements with Distinctive, Exclusive, and Primary readings, even with Mullen's improvements to their definitions, suffer from a practical problem: they are not easy to compare. For example, if Cyril agrees with 50% of the Distinctive Alexandrian readings and 25% of the Distinctive Western readings in some book, is the 50% actually greater than the 25%? Not necessarily, seeing that text-types vary greatly in how many distinctive readings they exhibit, how widely they are found among even strong members of the type, and how much borrowing blurs the lines between types. The ad hoc rules defining these categories make the different percentages of agreement incommensurable. There is an important concept, which is needed for weighing the significance of textual differences, trying to break loose here among these Inter-Group Relationship statistics, but its clear and consistent formulation still needs to be worked out. © TC: A Journal of Biblical Textual Criticism, 1997. Bibliography Colwell, Ernest Cadman 1969. Studies in Methodology in Textual Criticism of the New Testament. NTTS 9. Leiden: Brill. Ehrman, Bart 1986. Didymus the Blind and the Text of the Gospels. SBLNTGF 1. Atlanta: Scholars Press. Ehrman, Bart 1987. "The Use of Group Profiles for the Classification of New Testament Documentary Evidence." JBL 106: 465-486. Ehrman, Bart; Fee, Gordon; and Holmes, Michael 1992. The Text of the Fourth Gospel in the Writings of Origen. Vol. 1. SBLNTGF 3. Atlanta: Scholars Press. Eldridge, Lawrence 1969. The Gospel Text of Epiphanius of Salamis. SD 41. Salt Lake City: University of Utah Press. Fee, Gordon 1981-1982. "Origen's Text of the New Testament and the Text of Egypt." NTS 28: 348-364. Greenlee, J. Harold 1955. The Gospel Text of Cyril of Jerusalem. SD 17. Copenhagen: Ejnar Munksgaard. Jarrett, Paul 1992. "Eusebius of Caesarea and the Text of the Gospels." Ph.D. diss., Southwestern Baptist Theological Seminary. Kim, K. W. 1949. "The Matthean Text of Origen in his Commentary on Matthew." JBL 68: 125-139. Kim, K. W. 1950. "Origen's Text of John in his On Prayer, Commentary on Matthew, and Against Celsus." JTS, n.s., 1: 74-84. Suggs, M. Jack 1954. "The New Testament Text of Eusebius of Caesarea." Ph.D. diss., Duke University. Vincent Broman San Diego, California, USA