

   FFiitt aa GGeenneerraalliizzeedd NNoonnlliinneeaarr MMiixxeedd MMooddeell

        gnlmm(y, distribution="normal", mu=NULL, shape=NULL, linear=NULL,
             nest=NULL, pmu=NULL, pshape=NULL, psd=NULL, exact=F, wt=1,
             delta=1, shfn=F, scale=NULL, points=10, common=F,
             envir=sys.frame(sys.parent()), print.level=0, typsiz=abs(p),
             ndigit=10, gradtol=0.00001, stepmax=10*sqrt(p%*%p),
             steptol=0.00001, iterlim=100, fscale=1)

   AArrgguummeennttss::

          y: A response vector for uncensored data, a two col-
             umn matrix for binomial data or censored data,
             with the second column being the censoring indica-
             tor (1: uncensored, 0: right censored, -1: left
             censored), or an object of class, response (cre-
             ated by `restovec') or repeated (created by
             `rmna').

   distribution: Either a character string containing the name
             of the distribution or a function giving the -log
             likelihood and calling the location and shape
             functions.

         mu: A user-specified function of `pmu', and possibly
             `linear', giving the regression equation for the
             location. This may contain a linear part as the
             second argument to the function. It may also be a
             formula beginning with ~, specifying a either lin-
             ear regression function for the location parameter
             in the Wilkinson and Rogers notation or a general
             function with named unknown parameters. If none is
             supplied, the location is taken to be constant
             unless the linear argument is given.

      shape: A user-specified function of `pshape', and possi-
             bly `linear' and/or `mu', giving the regression
             equation for the dispersion or shape parameter.
             This may contain a linear part as the second argu-
             ment to the function and the location function as
             last argument (in which case `shfn' must be set to
             TRUE). It may also be a formula beginning with ~,
             specifying either a linear regression function for
             the shape parameter in the Wilkinson and Rogers
             notation or a general function with named unknown
             parameters. If none is supplied, this parameter is
             taken to be constant unless the linear argument is
             given. This parameter is the logarithm of the
             usual one.

     linear: A formula beginning with ~, specifying the linear
             part of the regression function for the location
             parameter or list of two such expressions for the
             location and/or shape parameters.

       nest: The variable classifying observations by the unit
             upon which they were observed. Ignored if y has
             class, response.

        pmu: Vector of initial estimates for the location
             parameters.  If `mu' is a formula with unknown
             parameters, their estimates must be supplied
             either in their order of appearance in the expres-
             sion or in a named list.

     pshape: Vector of initial estimates for the shape parame-
             ters.  If `shape' is a formula with unknown param-
             eters, their estimates must be supplied either in
             their order of appearance in the expression or in
             a named list.

        psd: Initial estimate of the standard deviation of the
             normal mixing distribution.

      exact: If TRUE, fits the exact likelihood function for
             continuous data by integration over intervals of
             observation, i.e. interval censoring.

         wt: Weight vector.

      delta: Scalar or vector giving the unit of measurement
             (always one for discrete data) for each response
             value, set to unity by default. Ignored if y has
             class, response. For example, if a response is
             measured to two decimals, delta=0.01. If the
             response is transformed, this must be multiplied
             by the Jacobian. The transformation cannot contain
             unknown parameters. For example, with a log trans-
             formation, `delta=1/y'. (The delta values for the
             censored response are ignored.)

       shfn: If true, the supplied shape function depends on
             the location (function). The name of this location
             function must be the last argument of the shape
             function.

      scale: The scale on which the random effect is applied:
             identity, log, logit, reciprocal, or exp.

     points: The number of points for Gauss-Hermite integration
             of the random effect.

     common: If TRUE, `mu' and `shape' must both be functions
             with, as argument, a vector of parameters having
             some or all elements in common between them so
             that indexing is in common between them; all
             parameter estimates must be supplied in `pmu'.  If
             FALSE, parameters are distinct between the two
             functions and indexing starts at one in each func-
             tion.

      envir: Environment in which model formulae are to be
             interpreted or a data object of class, repeated,
             tccov, or tvcov.  If `y' has class `repeated', it
             is used as the environment.

     others: Arguments controlling `nlm'.

   DDeessccrriippttiioonn::

        `gnlmm' fits user-specified nonlinear regression equa-
        tions to one or both parameters of the common one and
        two parameter distributions (binomial, beta binomial,
        double binomial, mult(iplicative) binomial, Poisson,
        negative binomial, double Poisson, mult(iplicative)
        Poisson, gamma count, Consul generalized Poisson, loga-
        rithmic series, geometric, normal, inverse Gauss,
        logistic, exponential, gamma, Weibull, extreme value,
        Cauchy, Pareto, Laplace, and Levy; all but the bino-
        mial-based distributions may be right and/or left cen-
        sored).

        Nonlinear regression models can be supplied as formulae
        where parameters are unknowns. Factor variables cannot
        be used and parameters must be scalars. (See `fin-
        terp'.)

        The intercept of the location regression has a nor-
        mally-distributed random effect. This normal mixing
        distribution is computed by Gauss-Hermite integration.

        The `scale' of the random effect is the link function
        to be applied. For example, if it is `log', the sup-
        plied mean function, `mu', is transformed as
        log(exp(mu)+sd), where sd is the random effect parame-
        ter.

        It is recommended that initial estimates for `pmu' and
        `pshape' be obtained from `gnlr'.

   VVaalluuee::

        A list of class gnlr is returned.  The printed output
        includes the -log likelihood (not the deviance), the
        corresponding AIC, the maximum likelihood estimates,
        standard errors, and correlations. A list is returned
        that contains all of the relevant information calcu-
        lated, including error codes.

   AAuutthhoorr((ss))::

        J.K. Lindsey

   SSeeee AAllssoo::

        `finterp', `fmr', `glm', `glmm', `gnlr', `gnlr3', `lm'.

   EExxaammpplleess::

        library(gnlm)
        # data objects
        sex <- c(0,1)
        sx <- tcctomat(sex)
        dose <- matrix(rpois(20,10),nrow=2)
        dd <- tvctomat(dose)
        # vectors for functions
        dose <- as.vector(t(dose))
        sex <- c(rep(0,10),rep(1,10))
        nest <- rbind(rep(1,10),rep(2,10))
        y <- exp(0.2+0.3*dose+0.5*sex)+rgamma(20,2,5)+rep(rnorm(10),rep(2,10))
        y <- restovec(matrix(y, nrow=2), nest=nest)
        reps <- rmna(y, ccov=sx, tvcov=dd)
        #
        # log linear regression with gamma distribution
        mu <- function(p) exp(p[1]+p[2]*sex+p[3]*dose)
        print(z <- gnlr(y, dist="gamma", mu=mu, pmu=c(1,0,0), pshape=1))
        gnlmm(y, dist="gamma", mu=mu, nest=nest, pmu=z$coef[1:3],
             pshape=z$coef[4], psd=0.1, points=3)
        # or equivalently
        gnlmm(y, dist="gamma", mu=~exp(b0+b1*sex+b2*dose), nest=nest,
             pmu=z$coef[1:3], pshape=z$coef[4], psd=0.1, points=3, envir=reps)
        # or with identity link
        print(z <- gnlr(y, dist="gamma", mu=~sex+dose, pmu=c(1,0,0), pshape=1))
        gnlmm(y, dist="gamma", mu=~sex+dose, nest=nest, pmu=z$coef[1:3],
             pshape=z$coef[4], psd=0.1, points=3)
        # or
        gnlmm(y, dist="gamma", mu=~b0+b1*sex+b2*dose, nest=nest, pmu=z$coef[1:3],
             pshape=z$coef[4], psd=0.1, points=3, envir=reps)
        #
        # nonlinear regression with gamma distribution
        mu <- function(p) p[1]+exp(p[2]+p[3]*sex+p[4]*dose)
        print(z <- gnlr(y, dist="gamma", mu=mu, pmu=c(1,1,0,0), pshape=1))
        gnlmm(y, dist="gamma", mu=mu, nest=nest, pmu=z$coef[1:4],
             pshape=z$coef[5], psd=0.1, points=3)
        # or
        mu2 <- function(p, linear) p[1]+exp(linear)
        gnlmm(y, dist="gamma", mu=mu2, linear=~sex+dose, nest=nest,
             pmu=z$coef[1:4], pshape=1, psd=0.1, points=3)
        # or
        gnlmm(y, dist="gamma", mu=~b4+exp(b0+b1*sex+b2*dose), nest=nest,
             pmu=z$coef[1:4], pshape=z$coef[5], psd=0.1,
             points=3, envir=reps)
        #
        # include regression for the shape parameter with same mu function
        shape <- function(p) p[1]+p[2]*sex
        print(z <- gnlr(y, dist="gamma", mu=mu, shape=shape, pmu=c(1,0,0,1),
             pshape=rep(1,2)))
        gnlmm(y, dist="gamma", mu=mu, shape=shape, nest=nest,
             pmu=z$coef[1:4], pshape=z$coef[5:6], psd=0.1, points=3)
        # or
        gnlmm(y, dist="gamma", mu=mu, shape=shape, nest=nest, pmu=z$coef[1:4],
             pshape=z$coef[5:6], psd=0.1, points=3, envir=reps)
        # or
        gnlmm(y, dist="gamma", mu=~b4+exp(b0+b1*sex+b2*dose), shape=shape,
             nest=nest, pmu=c(z$coef[4],z$coef[1:3]),
             pshape=z$coef[5:6], psd=0.1, points=3, envir=reps)

