46 
Fishery Bulletin 109(1 ) 
an empirical approximation for X c at a standardized 
temperature (0°C) is identical to that in Equation 1, 
except that a-- 2.8. 
Stomach fullness had no effect on response mea- 
surements within the half- or whole-body groups. The 
stomach and alimentary canal are encased by less con- 
ductive layers of muscle than the surrounding organs 
located in the peritoneal cavity (Pethig, 1979). Much 
like an insulated wire, these less conductive muscle 
layers will insulate the stomach contents even if the 
stomach contents are more conductive than the sur- 
rounding tissue. The insulation provided by the muscle 
layer reduces the chance that the current pathway will 
include stomach contents. Decreased R values (D/cm) 
are seen in various animals in the peritoneal spleen, 
liver, and kidney, and relatively higher R values in 
nearby muscle tissue (Pethig, 1979). These insulating 
muscles, coupled with less resistant alternative path- 
ways (i.e., organs), indicate that stomach fullness does 
not need to be accounted for in BIA measurements. 
Sensitivity analyses show that significant deviations 
from the procedures found in Cox and Hartman (2005) 
can lead to unacceptable errors in predictive estimates 
of R and X c , but nonsignificant deviations are more ac- 
ceptable. The average of all significant errors in this 
study is 26% and would cause parameter estimates 
to be off by about 25% to 30%, which is too large for 
most biological studies. The nonsignificant error aver- 
ages of <3% will cause parameter estimation errors to 
be around 2% to 4% (or about a 1:1 ratio), which may 
be acceptable in some studies. It should be noted that 
if several nonsignificant errors are encountered at the 
same time, they can be cumulative and result in an 
estimation error that is significant. 
In all electrical volume equations, length between 
detectors (L d ) is a squared term in the numerator, 
making predictive estimates extremely sensitive to 
changes in L d while also diluting the error effects on 
the denominator. Likewise, in parallel equations, the 
term R is either in the numerator (as in X c in parallel, 
see Table 1) or in the denominator (as in R in paral- 
lel, see Table 1) and is typically a much larger num- 
ber than X c , and therefore increases the influence of 
errors on parallel equations, especially when R is in 
the numerator as in reactance in parallel ( X c ). When 
the subsequent volume equations are used, predictive 
estimates are more sensitive to L d changes. The non- 
significant errors seen and described in this study are 
still deviations from the standard protocol found in Cox 
and Hartman (2005); therefore with a standard proto- 
col, these “nonsignificant” errors will not be reflected 
and any errors that are, would be from other factors 
not measured here (e.g., anatomy, thickness of skin and 
scales, condition, or biochemical composition). 
In summary, sources of error have been identified 
and found to significantly affect parameter estimates, 
but small errors that are not significant may be accept- 
able. In particular, electrode locations with respect to 
anatomy can significantly affect parameter estimates, 
and if electrodes needles are placed in the same ana- 
tomical location on each fish, impedance measurements 
will reflect the same relative volumetric areas within 
and between fish samples. Measurements need to be 
taken on a nonconductive surface that is clear of salt 
water, on blot-dried fish, and standardized with specific 
needle gauges and depths. New users need to be trained 
and taught stable body and hand positions and positions 
that allow a view of the needle to ensure accurate and 
precise measurements. Because temperature affects R 
and X c measurements, internal temperature needs to 
be measured to allow adjustments of R and X c . values to 
0°C or fish need to be stored on ice. Time is critical in 
taking impedance measurements, but icing fish can add 
9 h between fish death and the time of BIA measure- 
ments. Stomach fullness of fish does not affect half- or 
whole-body impedance measurements, and therefore 
does not have to be accounted for. Sensitivity analysis 
in our study showed that significant deviations from 
the procedures of Cox and Hartman (2005) can lead 
to unacceptable errors in predictive estimates of BIA 
measurements but nonsignificant deviations are more 
acceptable. Although adherence to these protocols can 
provide consistent measurements of impedance, compa- 
rability between researchers will depend on the develop- 
ment of training procedures, improved understanding 
of temperature effects, development of improved elec- 
trodes, continuous calibration with actual laboratory 
measurements, and unified standard protocols. It should 
also be noted that multifrequency impedance analyz- 
ers are available and currents at different frequencies 
could possibly have different measurements than those 
with a single frequency. The identification of sources 
of error illustrated here and subsequent adherence to 
a standardized protocol will offset the sources of error 
that may be present in bioelectrical impedance research 
and allow the technology to advance. 
Acknowledgments 
We acknowledge the financial support of the Alaska 
Fisheries Science Center, National Oceanic Atmospheric 
Administration, West Virginia University, U.S. Depart- 
ment of Agriculture, U.S. Forest Service, MeadWestvaco 
Corporation, West Virginia Department of Wildlife, 
United States Forest Service, University of Alaska, 
Fairbanks, and the Sitka Sound Science Center, Sitka, 
Alaska. We thank the many researchers and volunteers 
that helped collect data. 
Literature cited 
Blanca, L., P. Miquel, M. Soledad, P. Marina, A. P. Lucy, and 
H. A. Ildefonso. 
2009. Sources of error and its control in studies on the 
diagnostic accuracy of “-omics” technologies. Proteomics 
Clin. Appl. 3:173-184. 
Caton, J. R., P. A. Mole, W. C. Adams, and D. S. Heustis. 
1988. Body composition analysis by bioelectrical imped- 
