FISHERY BULLETIN: VOL, 83, NO. 4 



some extent by previous studies. The sex ratio and 

 the fraction of yearhng dolphins in a sample were 

 found to vary with kill size (No. 5 above). The frac- 

 tion of neonatal animals in tuna-vessel samples was 

 found to vary significantly with area, season, chase 

 time, dolphin kill, and tuna catch (Nos. 1, 2, 3, 5, and 

 6 above)^. Preliminary work by A. A. Hohn* and M. 

 D. Scott^ has indicated that immature spotted 

 dolphins may segregate into schools with different 

 characteristic school sizes (No. 4 above). 



The significance of each of the above factors was 

 tested separately using a Pearson chi-squara For 

 each test, the null hypothesis was that either per- 

 cent pregnant, percent lactating, or percent mature 

 is independent of the factor being tested. If this 

 hypothesis is shown to be false, then it is likely that 

 reproductive parameters may be affected by sam- 

 pling bias. Unfortunately, the above sampling fac- 

 tors are not independent; a factor may appear signifi- 

 cant when, in fact, that factor is merely correlated 

 with a causative factor. Because of this, probabilities 

 should be interpreted with caution. 



Multiway comparisons were used to discriminate 

 factors which are truly important from those which 

 are correlated to significant factors. A hierarchical 

 approach was used, based on the log-linear model 

 for discrete multivariate analysis (program 

 BMDP4F, Dixon 1981). The above 2-way tests were 



used to identify factors that may be significant. The 

 factors that were significant in the 2-way tests were 

 used in 3-way tests. The factors which proved signifi- 

 cant in the 3-way tests were then included in 4-way 

 tests (which proved to be a practical upper limit on 

 multiway tests using our data set). In this manner, 

 multiway comparisons of factors could be tested, 

 whereas a 7-way test of all factors would not have 

 been feasible. 



Analysis of variance (ANOVA) was tried and re- 

 jected as an alternative to the log-linear model for 

 multiway comparisons. The method used in this trial 

 was to calculate percent pregnant for each set and 

 to use sets as replicates in an ANOVA. Although 

 ANOVA is recognized to be robust to violations in 

 assumptions, the sample size for individual sets is 

 very small (mean number of mature females per set 

 is 1.6, mode is 1). As a result, the percent pregnant 

 in 72% of sets was either 0% or 100% of mature 

 females. No transformation was able to normalize 

 these data. Using an arc-sine transformation, 2-way 

 ANOVA was not even able to recognize the four 

 significant factors affecting percent pregnant that 

 were identified using a simple Pearson chi-squara 

 Because of these problems, the ANOVA model was 

 rejected for use in the multiway comparisons. 



RESULTS 



^Powers, J. E., and J. Barlow. 1979. Biases in the tuna-net sam- 

 pling of dolphins in the eastern tropical Pacific Doc SOPS/79/31, 

 7 p.; available from Southwest Fisheries Center La JoUa Laboratory, 

 National Marine Fisheries Service, NOAA, P.O. Box 271, La Jolla, 

 CA 92038. 



*A. A. Hohn, NOAA, National Marine Fisheries Service; P.O. Box 

 271, La Jolla, CA 92038, pers. commun. November 1983. 



^M. D. Scott, Inter-American TVopical Tlma Commission, P.O. Box 

 271, La Jolla, CA 92038, pers. commun. November 1983. 



Percent Pregnant 



The fractions of sexually mature females that were 

 pregnant are given in Thble 1 for the samples of the 

 northern and southern offshore stocks collected 

 from 1971 to 1983. Sample sizes decline in the later 

 years for the northern stock, but are typically MOO. 



Table 1. — Fractions a) of females that were sexually mature, b) of sexually mature females that were pregnant, and c) of preg- 

 nant females that were lactating. Samples include 1971-83 specimens from the northern and southern offshore stocks of spotted 

 dolphins. Fraction mature was not sampled in 1971 and 1972. 



660 



