120 



Fishei-y Bulletin 100(1) 



pie retained for CF analysis and the sub- 

 sample e.xcludecl from analysis were not 

 significantly different ichi-square test of 

 homogeneity Z"=--80. 1 df). A nonparamet- 

 ric growth model was estimated by fitting 

 a scatterplot smooth to the carapace length 

 and estimated age data (Fig. 4A). Because 

 the CF method could not be applied to the 

 largest turtles in the sample, the model 

 provided no information about growth in 

 mature turtles. 



Age estimates for posthatchling turtles 

 based on the SI method range from 3.3 

 to 49.4 yr (n = 103). The nonparametric 

 growth model for these data fitted very 

 well (Fig. 4B ), a result iiot unexpected giv- 

 en the dependence of the SI age estimate 

 on humerus diameter and the tight linear 

 relationship between humerus diameter 

 and carapace length. Variation in carapace 

 length around the SI growth model reflect- 

 ed only variance in the linear relationship 

 between carapace length and humerus di- 

 ameter: it did not incorporate variance in 

 the age of turtles at a fi.xed humerus di- 

 ameter or estimates of their age. The boot- 

 strap distributions of age estimates for the 

 SI method suggested that age estimates 

 for a turtle can vary by up to 10 years 

 I Fig. 5 A I. Nevertheless, the nonparamet- 

 ric confidence inten'als for predicted mean 

 length at age were narrow (Fig. 5B). 



Although the CF and SI methods yield- 

 ed different estimates of age iFig. 4). over 

 a wide range of carapace lengths, the dif- 

 ferences were not striking. The magnitude 

 and direction of differences in age esti- 

 mates depend on SCL. Judging from the 

 fitted smooth curves, for turtles up to 

 about 81 cm SCL. the correction factor 

 method was expected to give estimates 

 of age up to 2 years higher than those 

 produced by the spline-integration meth- 

 od. For larger turtles, however, age esti- 

 mates derived from the correction-factor 

 method were predicted to be as many as 10 

 years (or more) lower than those generat- 

 ed by spline integration iFig. 6). The /-tests 

 (Table 1) indicated there were highly sig- 

 nificant differences in age estimates be- 

 tween methods within the 40-50 cm size 

 group (/=4.36, 9 df) and the 50-60 cm group 

 ( f =4.66, 12 df). Differences for the 30-40 cm. 

 60-70 cm, and 70-80 cm size groups were 

 nonsignificant; samples were too small for 

 comparisons in other size groups. 



The Molokai mark-recapture data al- 

 lowed a visual comparison of observed 

 growth and growth predicted by the CF 

 and SI models derived from skeletochro- 



