Environmental Engineering Reference
In-Depth Information
species/wetland combination as a blank entry. It is important to realize that missing
data and data containing zeros represent vastly different representations of the data.
Observer bias
is a mechanical error and constant factor to consider in studies
and represent variation among observers. Such bias can be represented in
differences in skill of ocular or aural estimate of a variable (e.g., number of birds
in flock, percent vertical cover of vegetation, soil moisture relative to field capacity,
species of calling amphibians), ability in using a technique (i.e., proficiency with an
instrument, ability to distinguish the appropriate scale of measurement) to measure
a variable, and human error in recording and transcribing data. If one can measure
the magnitude and direction of inter-observer variation, then the data can be
adjusted for the bias (Morrison et al.
2001
). However, it is quite rare to be able to
adjust for observer bias. Therefore, it is important that all observers are trained and
tested relative to the data being collected prior to sampling. In most instances, it
would be appropriate to consider minimizing the number of observers that record
noninstrumented data to reduce observer bias (e.g., same person should conduct
bird counts, listen for amphibian calls, estimate percent cover of vegetation types).
However, even with limited observers, one must be able to determine if a system-
atic bias resulting from observer bias where a variable is consistently under- or
overestimated due to the selection of sampling points or unit of data measurement
(Thompson et al.
1998
).
A procedural type of error is what Cochran (
1977
) termed
gross error
where
mistakes are made in transcribing, entering, typing, and editing data and results
from analyses. Therefore, all study designs must have a well-defined, unambiguous
observation/measurement methodology prior to collection of data. In addition, a
protocol for data management is necessary prior to initiating any study. For
example, it should be mandated that all paper data sheets be copied at the end of
each data collection period and copies placed in safe locations. All electronic data
should be backed up in at least two locations and paper copies of electronic data
should be printed and stored in a safe location. Loss of complete records of
historical data, while rare, can occur due to natural disaster (e.g., hurricane),
human error (e.g., inadvertently discarded), loss of electronic data (e.g., hard
drive failure), misfiling, or mislabeling. Finally, all data sets and preliminary data
analyses results should be reviewed and copy-edited prior to conducting final
analyses to ensure observations were accurately transcribed. This is critical if one
is using voice-recognition software to enter data. All data should be linked to
specific observers. In addition, it is preferable for data review to occur shortly
after collection or transcription so that (1) technicians collecting the data remain
available to answer any questions, (2) illegible handwriting can be deciphered using
unsullied memories, and (3) there is an increased likelihood for recollection of data
should issues be identified.
Procedural errors occur when design and sampling protocols are not correctly
followed for recording measurements, transcribing and storing data, and conducting
data analyses. Development of a structured Quality Assurance and Quality Control
(QA/QC) program prior to initiating a project will minimize this bias. A QA/QC
program is the foundation for risk management in a study. In addition, any ethical
