SIZE-FRACTIONATED PRIMARY PRODUCTIVITY 369 



Statistics 



Univariate analysis of variance (ANOVA) computations were 

 made with the program NWAYl of the STAT JOB series (Academic 

 Computing Center, University of Wisconsin, Madison). The 

 M ANOVA computations were made with program BMD12V of the 

 BMD series (Dixon, 1968). 



RESULTS 



Primary Productivity 



The unfractionated productivity measurements ranged from < 1 

 to about 30 mg C m~^ hr~ ' , with the highest values occurring in late 

 October. Interestingly, the < 10- and the 10- to 64-iUm fractions 

 remained relatively low and constant throughout the year, usually 

 contributing less than 5 mg C m~^ hr~' . The > 64-^m fraction, on 

 the other hand, was responsible for the large seasonal variations 

 observed in the unfractionated productivity measurements, reaching 

 peak values of about 20 mg C m~^ hr~ ^ in the fall. 



The results of the productivity measurements were tested by the 

 MANOVA technique. The model used as a first step was a three-way 

 partial hierarchical design. The three factors were date, station, and 

 sample nested within station. The dependent variables were the three 

 size fractions (the unfractionated totals and the sums of the fractions 

 were tested separately). With the experimental design used, we hoped 

 to estimate adequately the within-station variance and factor it out 

 from the true between-station variance. Each sample was replicated 

 (the two light bottles) to account for technique errors. The three 

 samples taken at each station should estimate within-station variance. 



The MANOVA results are shown in Table 1. The first analysis, 

 which was for the six dates when all four stations were sampled, 

 indicates that all main effects and their interactions were significant. 

 When the analysis was repeated for the 18 dates with only samples 

 from stations FRB (precondenser) and KD (postcondenser), again all 

 effects were significant (Table 1). 



The MANOVA was also applied to the results for the four dates 

 when there was a AT of zero. The only significant factor at a = 0.05 

 was date (Table 1). A test on the remaining dates with a AT again 

 showed all effects to be significant. This indicates that productivity 

 rates at KD were different from FRB because of plant operations. To 

 make the tests more comparable, since only four dates had no AT, 

 we selected four dates randomly (June 25, Sept. 20, July 12, and 

 Jan. 15). The MANOVA test on these four dates again showed all 

 effects to be significant. Mean productivity rates were slightly lower 



