Kendall W (coefficient of concordance) was computed for those studies containing the same morphemes in the minimum number of obligatory occasions. For studies with nine morphemes in common (Marta, Uguisu, Dulay and Burt, 1974-combined, Dolores, Andersen, 1976, and Larsen-Freeman's two administrations of the BSM), W = 0.619, p < 0.001. For studies with the same morphemes in common (Jorge 7, 11, 18, 20, Krashen et al., 1977), W = 0.64, p < 0.01. For studies with the same seven morphemes in common (Birnbaum, Butler, and Krashen, 1977-Free I, Edit. I, Free II, Edit. II, Cheo, Alberto), W = 0.618, p < 0.01. de Villiers (1974) computed a Kendall W of 0.60 (p = 0.001) for her individual agrammatic subjects.

The relationships proposed in Fig. 1 are also supported in Andersen (1977), who reanalyzed his 1975 data in several interesting ways. Andersen also presents data indicating significant agreement among individual subjects. Additional evidence against excessive individual variation is Bailey et al. (1974), who found "a high level of agreement" among different classes of ESL students for grammatical morpheme difficulty order. Each subgroup contained about ten students.

While all correlations with the "natural order" for Monitor-free studies are positive, a few miss statistical significance at the 0.05 level. This is occasionally due to unusual performance in one morpheme: in Juan, for example, there was very high performance in the III singular morpheme (16/16). In my judgment, this failure to reach significance in every case is not serious, as several studies that "miss" come quite close (e.g. Cheo) and the effect is reliable. See Ferguson (1971), among others. for a discussion of the prevalence of type II errors when such near misses are analyzed as non-rejection of the null hypothesis when n's are small, as they are here.

Wode, Bahns, Bedey, and Frank (1978) discuss several "shortcomings of the morpheme order approach" which deserve repeating. First, they correctly point out that any approach that focuses exclusively on "the relative chronology of target-like mastery of several items... necessarily misses all developments leading toward and preceding the final state of achievement" (p. 181). Wode et al.'s data from child second language acquisition, along with earlier studies in L1 acquisition, illustrate quite clearly that the study of transitional competence, the intermediate structures performers use on their way to "target-like mastery", reveals an enormous amount about language acquisition that focusing on final forms misses. Second, Wode et al. claim that "morpheme order studies" miss avoidance phenomena. A good example is provided by Wode et al.'s subject, who produced no constructions of the sort

N + 's + N

where the first N is not a name. That is, they would produce utterances like

Johnny's dog

but not

the cat's ear

in English. Wode et al. suggest that the reason for this avoidance is the fact that such constructions are ungrammatical in the L1, German:

Heikos Angel (Heiko's fishing pole)