in the model-building process, such as assumptions regarding measurement
error (Stoms et al. 1992). Like the previously discussed data collection
methods, simulating data introduces assumptions into the model. Choices
of parameters and data ranges to simulate, as well as the algorithms used
for simulation, are important assumptions and considerations. In situations
where there is not time for data collection and the ecological model is going
to be used as a decision tool, simulated data for a reasonable range of model
parameters can be used in a modeling process to gain insight about the
system. Although quantitative predictions of realistic system behavior may
not be possible, the relative importance of model parameters can be determined. This type of modeling process can then lead to improved data collection of the key model parameters (Starfield 1997).
Sensitivity analysis (SA) is the study of how the variation in the output
of a model can be apportioned, qualitatively or quantitatively, to different
sources of variation. A large number of sensitivity analysis methods are
available in the literature (Beck 1987; Bedford 1998; Fürbinger and Roulet
1994). Each method has its advantages and disadvantages. The choice of the
method to adopt to perform an SA experiment on a model is, therefore, a
very delicate step that depends on a number of factors: the properties of
the model under study (linearity, additivity, monotonicity, etc.); the number
of input factors involved in the analysis; the computational time needed to
evaluate the model; and, last but not least, the objective of the analysis
(Saltelli et al. 2000).
10.4 Data Quality Concerns
Because data fundamentally influence the model through all steps of the
modeling process, it is critical to identify and quantify sources of error in
the data. Identifying the source of the error allows a modeler or manager
to correct or manage the data appropriately. In some cases, the target population for extrapolation can be adjusted to reflect problems with the data.
10.4.1 Sources of Error in the Data
Errors in the data can be of two kinds: sampling and nonsampling errors.
Sampling error is the topic of many sampling-theory texts [e.g., Cochran
(1977); Kish (1995)]. These errors are based on sampling only a portion of
the population rather than the entire population and on the fact that we
are not certain of the relationship between the sample data and the population of interest. In addition, modeling errors, including incorrect specification of the probability distribution of the population and incorrect
assumption of homogeneity of variance, can affect the model results.
Sampling errors include such issues as field measurement error, analytical errors, recording errors, coding errors, field-crew variability or drift,
194
David Hohler et al.
error (Stoms et al. 1992). Like the previously discussed data collection
methods, simulating data introduces assumptions into the model. Choices
of parameters and data ranges to simulate, as well as the algorithms used
for simulation, are important assumptions and considerations. In situations
where there is not time for data collection and the ecological model is going
to be used as a decision tool, simulated data for a reasonable range of model
parameters can be used in a modeling process to gain insight about the
system. Although quantitative predictions of realistic system behavior may
not be possible, the relative importance of model parameters can be determined. This type of modeling process can then lead to improved data collection of the key model parameters (Starfield 1997).
Sensitivity analysis (SA) is the study of how the variation in the output
of a model can be apportioned, qualitatively or quantitatively, to different
sources of variation. A large number of sensitivity analysis methods are
available in the literature (Beck 1987; Bedford 1998; Fürbinger and Roulet
1994). Each method has its advantages and disadvantages. The choice of the
method to adopt to perform an SA experiment on a model is, therefore, a
very delicate step that depends on a number of factors: the properties of
the model under study (linearity, additivity, monotonicity, etc.); the number
of input factors involved in the analysis; the computational time needed to
evaluate the model; and, last but not least, the objective of the analysis
(Saltelli et al. 2000).
10.4 Data Quality Concerns
Because data fundamentally influence the model through all steps of the
modeling process, it is critical to identify and quantify sources of error in
the data. Identifying the source of the error allows a modeler or manager
to correct or manage the data appropriately. In some cases, the target population for extrapolation can be adjusted to reflect problems with the data.
10.4.1 Sources of Error in the Data
Errors in the data can be of two kinds: sampling and nonsampling errors.
Sampling error is the topic of many sampling-theory texts [e.g., Cochran
(1977); Kish (1995)]. These errors are based on sampling only a portion of
the population rather than the entire population and on the fact that we
are not certain of the relationship between the sample data and the population of interest. In addition, modeling errors, including incorrect specification of the probability distribution of the population and incorrect
assumption of homogeneity of variance, can affect the model results.
Sampling errors include such issues as field measurement error, analytical errors, recording errors, coding errors, field-crew variability or drift,
194
David Hohler et al.