projections (Church et al. 1989). So this technique merely introduces a
manageable complication to what was already going to be a probabilitybased statement.
Correlated independent variables in statistical models may also present
a difficulty to extrapolation. During model-fitting procedures, such as stepwise regression, correlated independent variables are typically not retained
because little additional variation is explained by adding either correlated
variable when the other is present. However, in instances where the two
variables are likely to be targeted for management actions, there may be
political, economic, or social reasons for including confounded independent variables. In cases where large numbers of interrelated variables are
involved, more complex analytical techniques, such as partial least squares
or principal component regression, may be used to build models without
creating a list of the key variables involved.
A case study from the Pacific Northwest demonstrates that problems
associated with including confounding variables and attempting to make
management recommendations from the statistical model. In the arid
regions of the Pacific Northwest (i.e., east of the Cascade Range), water
diversions for irrigation are common. Water diversions affect salmon
(Oncorhynchus spp.) by reducing (or removing) flows, and unscreened
water diversions may intercept juvenile salmon and divert them to agricultural fields. Diversions are located primarily on the east side of the
Cascades because the climate there dictates this type of irrigation. In
response to growing fears over salmon declines, estimating the impact of
these structures and the number of salmon that could be saved by removing the diversions was a restoration goal. Because the spatial distribution
of diversions was confounded with (i.e., not homogenous or randomly distributed with respect to) various other climate, habitat, and land-use variables, the data could not support a calculation of the number of salmon that
would be saved by removing the diversions (Feist et al. submitted) despite
all attempts by managers to do that.
10.5 Data Storage and Management
10.5.1 Database Design
Database design is a significant topic in and of itself. The design of the database and normalization of the tables in the database need to be decided
before data collection begins. Data sheets and form design are separate
from database design. Most database management systems allow the database structure to be changed at any time. The problem is that, for data elements that are added, the cells of records already entered will be null. As
a result, the structure of the database should not be substantially changed
after data have been entered, and this structure influences how data can be
10. Effective Ecological Modeling: Data Issues
199
manageable complication to what was already going to be a probabilitybased statement.
Correlated independent variables in statistical models may also present
a difficulty to extrapolation. During model-fitting procedures, such as stepwise regression, correlated independent variables are typically not retained
because little additional variation is explained by adding either correlated
variable when the other is present. However, in instances where the two
variables are likely to be targeted for management actions, there may be
political, economic, or social reasons for including confounded independent variables. In cases where large numbers of interrelated variables are
involved, more complex analytical techniques, such as partial least squares
or principal component regression, may be used to build models without
creating a list of the key variables involved.
A case study from the Pacific Northwest demonstrates that problems
associated with including confounding variables and attempting to make
management recommendations from the statistical model. In the arid
regions of the Pacific Northwest (i.e., east of the Cascade Range), water
diversions for irrigation are common. Water diversions affect salmon
(Oncorhynchus spp.) by reducing (or removing) flows, and unscreened
water diversions may intercept juvenile salmon and divert them to agricultural fields. Diversions are located primarily on the east side of the
Cascades because the climate there dictates this type of irrigation. In
response to growing fears over salmon declines, estimating the impact of
these structures and the number of salmon that could be saved by removing the diversions was a restoration goal. Because the spatial distribution
of diversions was confounded with (i.e., not homogenous or randomly distributed with respect to) various other climate, habitat, and land-use variables, the data could not support a calculation of the number of salmon that
would be saved by removing the diversions (Feist et al. submitted) despite
all attempts by managers to do that.
10.5 Data Storage and Management
10.5.1 Database Design
Database design is a significant topic in and of itself. The design of the database and normalization of the tables in the database need to be decided
before data collection begins. Data sheets and form design are separate
from database design. Most database management systems allow the database structure to be changed at any time. The problem is that, for data elements that are added, the cells of records already entered will be null. As
a result, the structure of the database should not be substantially changed
after data have been entered, and this structure influences how data can be
10. Effective Ecological Modeling: Data Issues
199