Cox et al.: Measurements of resistance and reactance in fish with the use of bioelectrical impedance analysis 
41 
Control 1 
Control 1 
User 
Figure 3 
Boxplots of resistance and reactance measurements from five 
different people in a comparison of a trained user (control) with 
users without any training (1—4). Five coho salmon (Oncorhyn- 
clius kisutch) were measured by each person. Open circles (O) 
represent outliers determined by a Grubbs test. Closed circles 
(•) represent mean values. Different symbols indicate differ- 
ences as determined by the statistical tests applied. 
values of R for combinations of fish with full or empty 
stomachs with both half- and full-body measurements 
were 321, 312, 830, and 805 12, respectively (Fig. 6A). 
Although there was not a significant difference between 
mean values off?, nonsignificant differences in R means 
between full and empty stomachs for both half- and 
full-body measurements were -9.4 12 (-3%) and -25.2 12 
(—3%) (Table 2). Mean X c values were not significantly 
different between fish with full and empty stomachs 
for mid-body measurements (Tukey HSD, P=0.99) or 
full-body measurements (Tukey HSD, P=0.14) (Table 2, 
Fig. 6B). Mean values for R for fish with full or empty 
stomachs for both half-, and full-body measurements 
were 89, 89, 204, and 185 12, respectively (Table 2, Fig. 
6B). Although there was not a significant difference 
between mean values of X c , nonsignificant differences 
in means between fish with full or empty stomachs for 
both half- and full-body measurements were 0.6 12 (<1%) 
and -19.4 12 (9%). Variation of the estimations was also 
greater in the half-body measures in both R and X c 
measures (Fig. 6, A and B). 
Sensitivity analysis 
Predictive models for estimating total body water were 
highly inaccurate when significant errors were inserted 
into the models and considerably more accurate when 
nonsignificant errors were inserted into the models 
(Fig. 7, A and B). Inserted significant errors (-58% to 
10%) were inversely correlated with parameter estima- 
tion errors. The maximum significant negative error 
(-58%, conductive board) resulted in an overestimation 
1200 - 
1100 - 
1000 - 
900 - 
800 - 
700 - 
600 - 
no difference 
o.. 
O 
20 
40 
60 
80 
3 260 
ro 
(D 
cn 
240 
220 
200 - 
180 
160 
5 10 15 20 
Time (h) 
Figure 4 
Resistance (A) and reactance (B) measurements 
taken over time on six groups of six coho salmon 
( Oncorhynchus kisutch) (n = 36) that were killed 
and placed on ice. Open circles (O) represent 
outliers determined by a Grubbs test. The heavy 
dashed line indicates mean values and the light 
dashed line indicates 95% confidence intervals. 
The vertical dashed line and the arrow from it 
represent the times in which there was no differ- 
ence in resistance or reactance measurements. 
>120% and a 10% error (decreased needle depth) resulted 
in an underestimation >-10% (Fig. 7A). The addition of 
length errors (0% to 5%) produced results that were posi- 
tively correlated with parameter estimation errors and 
compounded the overall parameter estimation error (Fig. 
7A). Inserted nonsignificant errors (-3% to 3%) were 
also inversely correlated to parameter estimates (Fig. 
7B). The maximum nonsignificant negative error (-3%, 
full-stomach) resulted in an underestimation of 2.7%, 
and the maximum nonsignificant positive error (3%, R 
at 3 h) resulted in an overestimation of 2.6% (Fig. 7B). 
Length error alone caused overestimations to range from 
