Environmental Engineering Reference
In-Depth Information
should not be used following a study to evaluate confidence when failing to reject a
null hypothesis (i.e., retrospective power; Gerard et al.
1998
).
Statistically, although referred to as error, variation within the target population
is important to correctly estimate as it is the foundation of many statistical
techniques used for testing differences among levels of dependent variables. Esti-
mation of
experimental error
is the inherent variation among experimental units
treated alike or variation not explained by treatments or other variables. Accurate
estimation of experimental error is critical for testing treatment effects on response
variables. Experimental error differs from
sampling error
, which is the variation
among samples (or observations) of a given experimental unit. Sampling error can
be due to natural variability among units under study and can result from chance or
sampling bias in selecting subjects for sampling (Cochran
1977
). Any time that
more than one sample or observation is recorded per experimental unit
(e.g., multiple plots or water samples/wetland), accounting for sampling error
needs to be considered as the study design is developed. An example of experimen-
tal error would be variation of above-ground biomass among wetlands; this could
be the result of a single sample collected in each wetland or the variation among
wetlands of the average of multiple samples taken within a wetland. Sampling error
would be the variation among samples within a single experimental unit; that is, the
variation of multiple samples of biomass collected within a wetland designated as
an experimental unit.
Mechanically, during the course of data collection for a study, a number of errors
are possible. Cochran (
1977
) outlined these and other sources of error in ecological
studies for which investigators must be prepared and vigilant. Proper methodology
is the primary protection from a study suffering from investigator bias, personal
values, and preconceived results. However, if the observations or measurements are
made incorrectly or with the inappropriate equipment, then
measurement error
is
a likely outcome. For example, species can be misidentified, counts incomplete,
flow meters improperly calibrated, and measurements taken at the improper scale
(e.g., meters recorded instead of millimeters) are among the countless potential
other sources of measurement error. Each observer tasked with data collection must
be trained, occasionally assessed, and dedicated to consistent effort to reduce
effects of measurement error on final results.
Another source of mechanical error is
missing data
due either to the failure to
record the proper measurements or loss of recorded data (e.g., nonfunctional
equipment, weather, electronic storage failure, loss of paper copies). Missing data
can cause serious issues with subsequent data analysis unless accounted for by an
appropriate analysis. Investigators should take steps to avoid missing data by
securing data, checking equipment functionality, and ensuring that procedures are
understood by all. At times, individuals fail to record an appropriately measured
zero in the data, choosing to leave the data cell blank or empty creates the
impression of missing data when, in reality, the results may be biased due to lack
of a zero. For example, when inventorying plant species in multiple wetlands, one
must be careful to ensure that when a species is not detected in a wetland that a zero
or absent is recorded and properly transcribed rather than leaving the results for the
