FISHERY BULLETIN: VOL. 85, NO. 2 



might be expected to persist in the absence of treat- 

 ment effects. All ANOVA's were performed on un- 

 transformed data (seagrass biomass or differences 

 in seagrass biomass) because Cochran's test for 

 heteroscedacity was nonsignificant on 2 of the 6 data 

 sets and log and square root transformations failed 

 to reduce the significance level (P < 0.05 on 2 and 

 P < 0.01 on the other 2). 



There was a clear and large effect of intense kick- 

 ing. The ANOVA's on adjusted data were highly 

 significant for every posttreatment sampling date, 

 indicating that the initial differences among sea- 

 grass matrices in average seagrass biomass shifted 

 significantly after application of harvest treatment 

 and never returned to initial levels even by fall 1984. 

 The 2 intense-kicking treatments had consistently 

 low seagrass even after the first light treatment but 

 especially after both treatment applications. Light 

 kicking and raking never differed significantly from 

 one another in seagrass biomass. The shelly control 



1 matrix diverged from the other control (II) in 

 having low values in all posttreatment samplings, 

 often grouping with the 2 intense-kicking matrices 

 in the Duncan's test (Table 5). 



Average biomass of seagrass in each treatment 

 matrix is compared in Figure 4 to the changes that 

 would be predicted from the average biomass in the 



2 untreated control matrices. This approach 

 smoothes out the seasonality and other temporal 

 variability by normalizing all the treatment means 

 to the control values. It assumes that the differences 

 among matrices observed in spring 1980 in average 

 biomass would be expected to persist indefinitely 

 and then calculates what percent of the expected 

 seagrass biomass each treatment matrix actually ex- 

 hibited on each sampling date. This assumption is 



clearly violated by the divergent behavior of the 2 

 control matrices, but it provides a conservative 

 estimate of the effects of harvest because the aver- 

 age of the 2 controls includes control matrix I, which 

 exhibited low seagrass biomass, perhaps because of 

 enhanced illegal clamming. Clam harvest treatments 

 immediately reduced seagrass biomass below the ex- 

 pected amounts, with greater effects of the second, 

 more intense (see Table 2), harvest treatments. The 

 2 intense clam-kicking treatments exhibited a 

 decline of about 65% in expected biomass from 

 spring 1980 until spring 1981, while biomass de- 

 clined by about 25% below expected in the raking 

 and light-kicking matrices. Seagrass biomass re- 

 covered to equal and even exceed expected values 

 by the very next sampling period in fall 1981 in the 

 raking and light-kicking matrices, and remained 

 high for the next 3 years. However, recovery in 

 seagrass biomass in the 2 intense-kicking matrices 

 did not begin to occur until sometime in fall 1982- 

 fall 1983 (Fig. 4) and was not yet complete by fall 

 1984. In fall 1984, almost 4 years after the second 

 harvest treatment, average biomass of seagrass in 

 the 2 intense-kicking plots was only 65% of the ex- 

 pected levels. These estimates are conservative if 

 the shelly control (I) matrix is actually a poor con- 

 trol for this experiment because we used the mean 

 of both controls as an expected value for Figure 4. 

 Scheffe a priori contrasts of matrix means (in Table 

 5) show that, despite the divergence of the 2 con- 

 trols, the mean seagrass biomass was significantly 

 (P > 0.05) less in the 2 intense-kicking matrices than 

 expected from the 2 controls in all sampling periods 

 after application of both harvest treatments. This 

 test provides the statistical justification for our 

 presentation of differences in Figure 4. 



Table 5. — The impact of clam harvesting on the average seagrass dry mass (± SE) per % m^ within the seagrass 

 habitat. Data presented for each date and matrix are the mean ( + SE) dry mass of seagrass per sample minus the 

 mean dry mass in spring 1980 for that particular matrix (from Table 3). Sample sizes appear in Table 1 . Clam harvesting 

 treatments occurred between spring 1980 and fall 1980 and again between fall 1980 and spring 1981. Superscripts 

 A-D indicate significant differences among matrices in Duncan's test at o = 0.05, with those means sharing capital 

 letter superscripts not differing significantly. 



' * * ' - P < 0.001 In one-way ANOVA's on untransformed dry masses, comparing the matrix means on each separate date. ANOVA's 

 were performed on the differences from spring 1980 matrix means because of pre-existing significant differences among matrices in 

 spring 1980 before application of harvest treatments. 



290 



