Geoscience Reference
In-Depth Information
One of the goals of the study was to determine if there was a significant
trend in the contamination levels over the 12 years of surveying. Another
goal was to assess the variability in contamination across the stations and to
see if this variability was more, or less, than the variability in mercury con-
centration for each station. The first goal mentioned directly relates to deter-
mining changes in lake pollution, whereas the other goal relates to decisions
about future allocation of sampling effort if the study is to continue. For
example, if there were more sampling effort available, the decision would
need to be made about whether to increase the frequency of observations per
station or to increase the number of stations.
When repeated observations are made on each sampling unit, there are
two ways to approach the data: through a unit analysis or through a pooled
analysis. The unit analysis involves a separate analysis on each sampling
unit (e.g., transect, quadrat, radio-collared animal, etc.). Results from the
individual units that are summarized across the units can be used to make
inferences about the population of all units. In our example, this would
involve first looking at the data from each station separately. Pooled analyses
generally involve mixed models (Myers et al
.,
2010). These models exploit the
fact that the sampling units are a random sample from a larger population of
units. The sampling units will vary from each other because of unmeasured
factors, such as lake depth, water currents, and aquatic vegetation. For terres-
trial studies, these factors may be forest composition, elevation, aspect, and
other intrinsic factors related to population dynamics. These unmeasured
factors are essentially rolled together into sampling unit variance compo-
nents. In our example, it is easy to imagine that there will be variation in
mercury measurements among stations because of small-scale variation in
the lake bed structure and fish diversity.
Mixed models allow for inferences to be made on individual units as well
as to the population of units as a whole. Although more complicated than
the unit analyses, there are many advantages. For example, these models can
easily accommodate missing data, such as when there are unequal numbers
of observations on each sampling unit. Mixed models can also be applied to
the analysis of data organized with multiple layers of nesting (e.g., monthly
data recordings nested within seasons).
11.2 Basic Methods for Trend Analysis
The use of simple but effective graphics is an essential part of initial data
exploration. Graphical displays of the data are useful for reporting on popu-
lation changes because they can be an effective way to convey to others the
main trends. An example is the index plot shown in Figure 11.1 for contami-
nation. The concentration of mercury measured at a particular station in a
