refund (version 0.1-23)

lpfr: Longitudinal penalized functional regression

Description

Implements longitudinal penalized functional regression (Goldsmith et al., 2012) for generalized linear functional models with scalar outcomes and subject-specific random intercepts.

Usage

lpfr(
  Y,
  subj,
  covariates = NULL,
  funcs,
  kz = 30,
  kb = 30,
  smooth.cov = FALSE,
  family = "gaussian",
  method = "REML",
  ...
)

Arguments

Y

vector of all outcomes over all visits

subj

vector containing the subject number for each observation

covariates

matrix of scalar covariates

funcs

matrix or list of matrices containing observed functional predictors as rows. NA values are allowed.

kz

dimension of principal components basis for the observed functional predictors

kb

dimension of the truncated power series spline basis for the coefficient function

smooth.cov

logical; do you wish to smooth the covariance matrix of observed functions? Increases computation time, but results in smooth principal components

family

generalized linear model family

method

method for estimating the smoothing parameters; defaults to REML

...

additional arguments passed to gam to fit the regression model.

Value

fit

result of the call to gam

fitted.vals

predicted outcomes

betaHat

list of estimated coefficient functions

beta.covariates

parameter estimates for scalar covariates

ranef

vector of subject-specific random intercepts

X

design matrix used in the model fit

phi

list of truncated power series spline bases for the coefficient functions

psi

list of principal components basis for the functional predictors

varBetaHat

list containing covariance matrices for the estimated coefficient functions

Bounds

list of bounds of a 95% confidence interval for the estimated coefficient functions

Details

Functional predictors are entered as a matrix or, in the case of multiple functional predictors, as a list of matrices using the funcs argument. Missing values are allowed in the functional predictors, but it is assumed that they are observed over the same grid. Functional coefficients and confidence bounds are returned as lists in the same order as provided in the funcs argument, as are principal component and spline bases.

References

Goldsmith, J., Crainiceanu, C., Caffo, B., and Reich, D. (2012). Longitudinal penalized functional regression for cognitive outcomes on neuronal tract measurements. Journal of the Royal Statistical Society: Series C, 61(3), 453--469.

Examples

# NOT RUN {
# }
# NOT RUN {
##################################################################
# use longitudinal data to regress continuous outcomes on
# functional predictors (continuous outcomes only recorded for
# case == 1)
##################################################################

data(DTI)

# subset data as needed for this example
cca = DTI$cca[which(DTI$case == 1),]
rcst = DTI$rcst[which(DTI$case == 1),]
DTI = DTI[which(DTI$case == 1),]


# note there is missingness in the functional predictors
apply(is.na(cca), 2, mean)
apply(is.na(rcst), 2, mean)


# fit two models with single functional predictors and plot the results
fit.cca = lpfr(Y=DTI$pasat, subj=DTI$ID, funcs = cca, smooth.cov=FALSE)
fit.rcst = lpfr(Y=DTI$pasat, subj=DTI$ID, funcs = rcst, smooth.cov=FALSE)

par(mfrow = c(1,2))
matplot(cbind(fit.cca$BetaHat[[1]], fit.cca$Bounds[[1]]),
  type = 'l', lty = c(1,2,2), col = c(1,2,2), ylab = "BetaHat",
  main = "CCA")
matplot(cbind(fit.rcst$BetaHat[[1]], fit.rcst$Bounds[[1]]),
  type = 'l', lty = c(1,2,2), col = c(1,2,2), ylab = "BetaHat",
  main = "RCST")


# fit a model with two functional predictors and plot the results
fit.cca.rcst = lpfr(Y=DTI$pasat, subj=DTI$ID, funcs = list(cca,rcst),
  smooth.cov=FALSE)

par(mfrow = c(1,2))
matplot(cbind(fit.cca.rcst$BetaHat[[1]], fit.cca.rcst$Bounds[[1]]),
  type = 'l', lty = c(1,2,2), col = c(1,2,2), ylab = "BetaHat",
  main = "CCA")
matplot(cbind(fit.cca.rcst$BetaHat[[2]], fit.cca.rcst$Bounds[[2]]),
  type = 'l', lty = c(1,2,2), col = c(1,2,2), ylab = "BetaHat",
  main = "RCST")
# }