Blurb::
Specify the type of model discrepancy

Description::
After the model parameters are calibrated, the difference between the data and
the calibrated model, i.e. the model discrepancy, is calculated

\f[
\delta_i(x_j) = d_i(x_j) - M_i(\theta^*, x_j).
\f]

Each \f$\delta_i\f$ corresponds to a different regression model. These 
regression models must all be either Gaussian process or polynomial models, and
they are functions of the configuration variable \f$x\f$. The order
of the trend function may be selected using the \c correction_order command
by specifying \c constant, \c linear, or \c quadratic.

Note that for Dakota 6.9 and earlier, this keyword only applies to discrepancy 
calculations for scalar responses. For field responses, a Gaussian process
model with a quadratic function is used by default.
