6
Sampling Design and Statistical Inference
for Ecological Assessment
Brian M. Steele
6.1 Introduction
The success of ecological assessment depends on
relevant and accurate information about the ecosystem or landscape under study. Data collected by
sampling are a primary source of information about
ecosystems and usually the only source of information that is specific to the ecosystem. The information value of sample data cannot be overestimated: these data directly reflect the processes
and organisms constituting the system and are independent of human-held assumptions and theories. There are three general approaches to using
sample data to describe ecosystems and ecosystem
processes: (1) often important ecosystem components can be assessed by analyzing a few summary
statistics; (2) ecosystem monitoring makes use of
sample data for trend estimation and uses hypothesis testing for change detection; and (3) occasionally, ecological assessment involves the modeling of ecosystem processes, and sample data are
used for model estimation and prediction.
Inference is the process of extrapolating from
sample data to popUlations, and statistical inference
is a methodology for minimizing the risk of drawing the wrong conclusions from sample data. Statistical inference can also quantify the risk incurred
by extrapolation. Despite the risks involved, proper
statistical inference based on good sample data is
a highly effective technique for scientific inquiry
and, more specifically, for ecosystem assessment.
The quality of the data and the inference drawn
from the sample are largely determined by the
method of data acquisition; in particular, statistical
inference is impossible without a statistically valid
sample. Thus sampling design plays a key role in
statistical inference and hence in ecological assessment.
In addition to providing statistically valid samples, sampling design promotes effective application of the scientific method to ecosystem assessment. Devising a sampling plan requires review of
ecosystem knowledge, identification of assumptions regarding the ecosystem, and statement of unambiguous research questions. Scientific rigor is
introduced into a sampling design by identifying a
relatively short list of clear and concise questions
regarding the ecosystem. For instance, if the objective is to investigate forest health, then the sample design defines a specific popUlation of interest,
such as forest stands located within a particular watershed. Formal inference requires unambiguous
questions, such as "what proportion of the population is infected with white pine blister rust?" Statistical inference provides the basis for constructing an interval, for example, [0.23, 0.35], from the
sample and ensures that the method succeeds in
bracketing the true unknown proportion at a specified rate, for example, 95% of the time. Without
a proper sampling design, a valid confidence interval cannot be constructed. Without statistical validity, the true unknown proportion will be bracketed at some other unknown rate.
Sampling designs are used to collect statistically
valid samples efficiently. Useful criteria for assessing and comparing sampling designs are accuracy, cost, and feasibility. An efficient sampling design is one that produces accurate estimates or
sensitive test statistics at low cost and effort. Efficiency is often difficult to achieve in ecological assessment because populations tend to consist of
complex entities that are often widely dispersed and
dissimilar. Furthermore, most ecosystems are dynamic and only partially, or poorly, understood. It
is challenging to design an efficient and statistically
valid sampling protocol for ecosystem assessment,
79
Sampling Design and Statistical Inference
for Ecological Assessment
Brian M. Steele
6.1 Introduction
The success of ecological assessment depends on
relevant and accurate information about the ecosystem or landscape under study. Data collected by
sampling are a primary source of information about
ecosystems and usually the only source of information that is specific to the ecosystem. The information value of sample data cannot be overestimated: these data directly reflect the processes
and organisms constituting the system and are independent of human-held assumptions and theories. There are three general approaches to using
sample data to describe ecosystems and ecosystem
processes: (1) often important ecosystem components can be assessed by analyzing a few summary
statistics; (2) ecosystem monitoring makes use of
sample data for trend estimation and uses hypothesis testing for change detection; and (3) occasionally, ecological assessment involves the modeling of ecosystem processes, and sample data are
used for model estimation and prediction.
Inference is the process of extrapolating from
sample data to popUlations, and statistical inference
is a methodology for minimizing the risk of drawing the wrong conclusions from sample data. Statistical inference can also quantify the risk incurred
by extrapolation. Despite the risks involved, proper
statistical inference based on good sample data is
a highly effective technique for scientific inquiry
and, more specifically, for ecosystem assessment.
The quality of the data and the inference drawn
from the sample are largely determined by the
method of data acquisition; in particular, statistical
inference is impossible without a statistically valid
sample. Thus sampling design plays a key role in
statistical inference and hence in ecological assessment.
In addition to providing statistically valid samples, sampling design promotes effective application of the scientific method to ecosystem assessment. Devising a sampling plan requires review of
ecosystem knowledge, identification of assumptions regarding the ecosystem, and statement of unambiguous research questions. Scientific rigor is
introduced into a sampling design by identifying a
relatively short list of clear and concise questions
regarding the ecosystem. For instance, if the objective is to investigate forest health, then the sample design defines a specific popUlation of interest,
such as forest stands located within a particular watershed. Formal inference requires unambiguous
questions, such as "what proportion of the population is infected with white pine blister rust?" Statistical inference provides the basis for constructing an interval, for example, [0.23, 0.35], from the
sample and ensures that the method succeeds in
bracketing the true unknown proportion at a specified rate, for example, 95% of the time. Without
a proper sampling design, a valid confidence interval cannot be constructed. Without statistical validity, the true unknown proportion will be bracketed at some other unknown rate.
Sampling designs are used to collect statistically
valid samples efficiently. Useful criteria for assessing and comparing sampling designs are accuracy, cost, and feasibility. An efficient sampling design is one that produces accurate estimates or
sensitive test statistics at low cost and effort. Efficiency is often difficult to achieve in ecological assessment because populations tend to consist of
complex entities that are often widely dispersed and
dissimilar. Furthermore, most ecosystems are dynamic and only partially, or poorly, understood. It
is challenging to design an efficient and statistically
valid sampling protocol for ecosystem assessment,
79