66 
Fishery Bulletin 99(1 ) 
was performed on the individual errors, the factors be- 
ing year, silhouette, and transect (nested within each 
combination of year and silhouette). 
With the stereo-video system, the mean of ten mea- 
surements should result in greater accuracy and pre- 
cision than use of a single measurement. In order to 
assess the extent of this improvement, the above anal- 
ysis was performed a further five times by using the 
first 1, 2, 4, 6, and 8 measurements. 
One of the most commonly cited papers on the train- 
ing of divers to estimate fish lengths with accuracy 
and precision is that of Bell et al. (1985). The data for 
this paper originated from a report published by the 
Great Barrier Reef Marine Park Authority (GBRMPA, 
1979). Tables 7, 10, 12, and 15 in that report contain 
underwater length estimates of pieces of orange PVC 
conduit (cut into 50 lengths ranging from 6 to 94 cm) 
determined by experienced scientific divers. Three ex- 
perienced divers undertook four transects estimating 
the lengths of the conduit seen. The estimates of a 
fourth diver are also included in Tables 10, 12, and 
15 of the report. We have chosen to disregard this div- 
er’s data because he was considered inexperienced at 
the time. We have summarized these results by first 
calculating the mean length estimate over all four 
transects for each silhouette and diver. Each of these 
means was then expressed by using the four summa- 
ry measures of error described: error, absolute error, 
relative error, and relative absolute error. For each 
measure of error, a one-factor analysis of variance was 
performed, with silhouette being the factor. The indi- 
vidual divers provided the replication needed for this 
analysis. 
Results 
Comparison of length estimates made by novice 
and experienced scientific divers with estimates 
generated by a stereo-video system 
Figure 1 shows the range of length estimate means over 
five transects for the novice and experienced scientific 
divers and for the stereo-video system. The variability 
of the estimates was greatest for the novice divers and 
slightly less for the experienced scientific divers. By com- 
parison, the length estimates made by the stereo-video 
system showed little variability around the true lengths. 
The coefficient of variation (CV=standard deviation/ 
mean) was significantly lower for the stereo-video than it 
was for either the experienced scientific or the novice div- 
ers (Fig. 2). 
For all four measures of error (E, RE, AE, and RAE), 
there was no significant interaction between type of diver 
and silhouette size, suggesting that any differences be- 
tween experienced scientific and novice divers were con- 
sistent across the silhouettes. For this reason, the results 
are presented as means across all silhouettes (Fig. 3). For 
both E and RE, the difference between the experienced sci- 
entific and novice divers was highly significant (P<0.0001); 
for AE and RAE, the difference was close to significant at 
the 5% level (P=0.08 and P= 0.05 respectively; F x gi =5.96 
[Ej, =1.75 [AE], =6.28 [RE], =1.94 [RAE]). 
For both the diver and stereo-video data, there were sig- 
nificant differences (at the 5% level) between silhouettes, 
for all measures of error except RAE on the diver data. In- 
spection of the silhouette means showed no clear pattern for 
these differences. Any pattern would be difficult to interpret 
because the silhouettes were placed at different distances 
from the transect line in order to provide a range of sizes at 
a range of distances from the transect line and thus make 
the comparison of means in Figure 3 widely applicable. 
The GBRMPA divers had a mean measurement error of 
-2.4 cm (SE=0.2 cm) which is similar to the -2.1 cm (SE=0.6 
cm) mean error recorded by the experienced scientific divers 
used in our study ( Fig. 3 ). Because 26 of the silhouettes used 
in the GBRMPA study were larger than any of those used 
in our study, it might be argued that the relative errors are 
more directly comparable. The mean RE for the GBRMPA 
divers was -4.6% (SE=0.5%), compared with a mean of 
-8.6% (SE=1.9%) for the experienced scientific divers used 
in our study. These results suggest that the experienced 
divers in the two studies had comparable skills. 
