scales of the required data, the populations from which data are needed,
and the feasibility of the alternatives. A statistician should be consulted
when making these decisions.
An experimental design, if done properly, can provide a defensible scientific approach with controls for extraneous factors, and can hence be used
to test for cause–effect relationships among variables of interest (Platt
1964). However, experimental designs have their drawbacks: properly
controlling for external factors may be complicated, expensive, or simply
impossible given the experimental situation; replication of the experiment
can be problematic in ecological situations, particularly when the spatial
scale is at the landscape level; and extending an experimental design to a
synoptic scope may not be feasible. On the other hand, sampling designs
are common in a wide spectrum of the sciences, and can provide information on the populations of interest, as well as the distributions of and correlations between key variables.
While data observed in sampling designs can provide insights into important patterns and relationships in the ecological systems, these data cannot
be used to test cause–effect relationships. Sampling designs provide observational information on target populations, not cause–effect structures with
complete control for all auxiliary variables. Typically, it is too costly to
analyze for all auxiliary variables at all sampling sites. Lack of information
on these auxiliary variables may lead to incorrect conclusions because the
missing information can lead the investigator to spurious correlations,
apparent relationships that are really artifacts caused by the lack of information on underlying factors. In general, consultation with an experienced
statistician is strongly recommended.
10.3.4.3 Sample Size and Sample Sites
Additional design choices must be made, whether the decision is for a sampling design or an experimental design. The investigator must determine
what is an appropriate size and shape for the sampling unit. The sampling
unit may be the size of a quadrat or an individual tree or a plot or a part
of a landscape. The choice of the size must be based on the design requirements. Usually, the size of the sampling unit is driven by such factors as the
variables to be measured, the desired resolution of the data, the scope of
the study, and the existence of standard protocols that meet the data needs.
By using established protocols for data collection, data are more likely to
be comparable across different studies, the variability and/or distribution of
the data is easier to assess from previous studies, and variability between
field crews or laboratories can be easier to control. The necessary number
of sampling units depends not only on the time and expense of data collection at each sample site and on the variability of the features of interest
but also on the type of design selected. Optimal sampling designs and
optimal experimental designs can reduce the number of samples needed to
192
David Hohler et al.
and the feasibility of the alternatives. A statistician should be consulted
when making these decisions.
An experimental design, if done properly, can provide a defensible scientific approach with controls for extraneous factors, and can hence be used
to test for cause–effect relationships among variables of interest (Platt
1964). However, experimental designs have their drawbacks: properly
controlling for external factors may be complicated, expensive, or simply
impossible given the experimental situation; replication of the experiment
can be problematic in ecological situations, particularly when the spatial
scale is at the landscape level; and extending an experimental design to a
synoptic scope may not be feasible. On the other hand, sampling designs
are common in a wide spectrum of the sciences, and can provide information on the populations of interest, as well as the distributions of and correlations between key variables.
While data observed in sampling designs can provide insights into important patterns and relationships in the ecological systems, these data cannot
be used to test cause–effect relationships. Sampling designs provide observational information on target populations, not cause–effect structures with
complete control for all auxiliary variables. Typically, it is too costly to
analyze for all auxiliary variables at all sampling sites. Lack of information
on these auxiliary variables may lead to incorrect conclusions because the
missing information can lead the investigator to spurious correlations,
apparent relationships that are really artifacts caused by the lack of information on underlying factors. In general, consultation with an experienced
statistician is strongly recommended.
10.3.4.3 Sample Size and Sample Sites
Additional design choices must be made, whether the decision is for a sampling design or an experimental design. The investigator must determine
what is an appropriate size and shape for the sampling unit. The sampling
unit may be the size of a quadrat or an individual tree or a plot or a part
of a landscape. The choice of the size must be based on the design requirements. Usually, the size of the sampling unit is driven by such factors as the
variables to be measured, the desired resolution of the data, the scope of
the study, and the existence of standard protocols that meet the data needs.
By using established protocols for data collection, data are more likely to
be comparable across different studies, the variability and/or distribution of
the data is easier to assess from previous studies, and variability between
field crews or laboratories can be easier to control. The necessary number
of sampling units depends not only on the time and expense of data collection at each sample site and on the variability of the features of interest
but also on the type of design selected. Optimal sampling designs and
optimal experimental designs can reduce the number of samples needed to
192
David Hohler et al.