Gauss-Hermite Integration
One-dimensional Gauss-Hermite integration
Gauss-Hermite integration approximates a specific integral as follows
in which and are weights and abscissas for N-point Gauss-Hermite integration, see Abramowitz and Stegun (1972). N-point integration is exact for all polynomials of degree 2N-1, see Dahlquist and Bjorck (1974). This can for instance be used to approximate the mean of a function of a normally distributed random variable with mean and variance :
Two-dimensional Gauss-Hermite integration
One-dimensional Gauss-Hermite integration can readily be extended to two dimensions. The following principal result in two dimensions is more or less given in Jäckel (2005) for the standard bivariate normal distribution with correlation parameter :
in which
and
as given in Jäckel (2005) .
Jäckel (2005) discusses other Gauss-Hermite approximations to the two-dimensional integral, but found that the approximation given above generally gives the most accurate results. For the general bivariate normal distribution with means and variances the integral can be approximated by means of
The product can be very small, especially when many quadrature points are used, thus wasting possibly precious calculation time. This can be remedied by pruning, i.e. by dropping combinations of with very small values of the product .
Maximum likelihood for the LNN model with two-dimensional Gauss-Hermite integration
Denote non-consumption on day for individual as . The conditional likelihood, i.e. given random effects and , of a non-consumption on day equals, with the inverse of the logit function
The conditional likelihood of a positive intake > 0 equals, with the density of the normal distribution
The conditional likelihood contribution for individual is the product of the individual contributions for each day. The marginal likelihood contribution for individual is obtained by integrating over the possible values of and . Since the pair follows a bivariate normal distribution, the likelihood contribution for individual can be approximated by means of two-dimensional Gauss-Hermite integration. Individually based covariables, such as sex or age, imply that and must be used instead of and . The likelihood must be optimized by means of some general optimization routine.