plotSurface2d: Plot bivariate posterior predictive exposure response surface

In jantonelli111/NLinteraction: Bayesian variable selection in multivariate semi-parametric regression models

Description

This function takes in an object which is a model from the NLint function and plots the bivariate exposure response curve and pointwise posterior standard deviations for exposures j1 and j2, keeping the remaining exposures at quantile_rest and the covariates at their mean

Usage

plotSurface2d(
  NLmod,
  X,
  C,
  j1,
  j2,
  gridLength_j1 = 20,
  gridLength_j2 = 20,
  minDist = Inf,
  quantile_rest = 0.5,
  ...
)

Arguments

NLmod	The fitted model
X	an n by p matrix of exposures to evaluate. These should be continuous.
C	An n by q matrix of additional covariates to adjust for.
j1	The exposure for which we will plot the exposure response function
j2	The second exposure which we will fix at a certain value. This should be a secondary exposure of interest in the sense that we want to see if the exposure response curve of j1 is different at different levels of j2 (i.e interaction)
gridLength_j1	The number of points to evaluate the curve at for the first exposure. More points make for a smoother curve, but also increase the computation time associated with calculating the curve
gridLength_j2	The number of points to evaluate the curve at for the first exposure. More points make for a smoother curve, but also increase the computation time associated with calculating the curve
minDist	If the euclidean distance between all observed data points is greater than this number for any point on the plot, we wont' plot the surface there. The default is infinity so that all points on the grid are plotted.
quantile_rest	The quantile at which we will fix the remaining exposures
...	Plotting parameters to be passed on such as ylim, ylab, main, etc.

Examples

n = 200
p = 10
pc = 1

sigma = matrix(0.3, p, p)
diag(sigma) = 1
X = rmvnorm(n, mean=rep(0,p), sigma = sigma)

C = matrix(rnorm(n*pc), nrow=n)

TrueH = function(X) {
  return(0.5*(X[,2]*X[,3]) - 0.6*(X[,4]^2 * X[,5]))
}

Y = 5 + C + TrueH(X) + rnorm(n)

NLmod = NLint(Y=Y, X=X, C=C)

par(mfrow=c(1,2), pty='s')
plotSurface2d(NLmod = NLmod, X=X, C=C, j1=4, j2=5,
              gridLength_j1=20, gridLength_j2 = 20, quantile_rest = 0.5)

jantonelli111/NLinteraction documentation built on Aug. 7, 2020, 10:02 p.m.