Environmental Engineering Reference
In-Depth Information
leads to increased training sessions for data collection and recording. Identification
of potential nuisance variables may be a result of a pilot study. Finally, data from a
pilot study should be subjected to the proposed statistical analyses to identify
potential issues such as deviation from statistical assumptions, probability distribu-
tion of independent variables, and identification of correlated independent
variables.
A primary approach to testing data quality is the resampling of a subset of each
data set and comparing results. Other proposed actions to ensure data quality
include redundant measurements by two or more observers or analysis of duplicate
laboratory samples. Recorded data should be proofed shortly after being measured
(e.g., completion of sampling period) by an independent observer to eliminate
recording and transcription errors. Researchers should assign unique study
responsibilities to each observer so that mistakes or errors can be linked to unique
individuals and thus, provides an opportunity to correct any incorrect data.
During data collection and at the conclusion of the study (for long-term studies or
monitoring efforts at frequent intervals), a final proofing of collected data is essential.
In addition, numerous copies of the data forms and electronic data bases need to be
made and stored in secured areas for future reference. Data collection points, plots, or
units need to be uniquely identified and locations recorded to assist in interpretation of
results, recollection of lost or erroneous data, and for ease of relocation for future
investigators or return for a comparison study by current investigators. All equipment
must be removed from the field, maintained/serviced, and again tested for accuracy.
During the data analysis and hypothesis testing stage, a number of additional steps are
necessary. Investigators should use statistics to describe the data (e.g., means,
measures of variation, missing data, distribution form, and range of values) prior to
primary statistical analyses. Graphical representation of the data prior to analyses is
appropriate as long as the subsequent planned analyses are not altered due to perceived
patterns observed in the data. Researchers should also realistically assess the sample
size relative to planned analyses. For example, some multivariate and modeling
approaches require relatively large sample sizes that may not be present. All statistical
tests have underlying assumptions (e.g., normality, constant variance) and, although
most approaches are relatively robust to at least minor violations of assumptions, it is
desirable to test the assumptions for each analysis. It is recommended to use a
statistical test that is appropriate for the data rather than alter the data through
transformation or some other technique just to use a particular statistical test.
Interpretation and eventual publication of the study results represent the conclu-
sion of a study design. Interpretation of results includes not only describing the data
and subsequent analyses but also discussing the relevance and context of the results
relative to previously published information (i.e., literature). Care must be taken not
to inappropriately extend the inference of the results beyond the study population.
Here, one must guard against letting the data and analyses determine the study
conclusions rather than using the data and results from analyses to support a
conclusion based on the accumulation of evidence. Fixation on statistical results
is a poor substitute for critical thinking of the results in an ecological context.
Reliable conclusions must be supported by data and be capable of withstanding
