rp.flm.test: Goodness-of fit test for the functional linear model using...

Description Usage Arguments Value Author(s) References Examples

View source: R/rp.flm.test.R

Description

Tests the composite null hypothesis of a Functional Linear Model with scalar response (FLM),

H_0: Y = <X, β> + ε vs H_1: Y != <X, β> + ε.

If β=β_0 is provided, then the simple hypothesis H_0: Y = <X, β_0> + ε is tested. The way of testing the null hypothesis is via a norm (Cramer-von Mises or Kolmogorov-Smirnov) in the empirical process indexed by the projections.

No NA's are allowed neither in the functional covariate nor in the scalar response.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
rp.flm.test(
  X.fdata,
  Y,
  beta0.fdata = NULL,
  B = 1000,
  n.proj = 10,
  est.method = "pc",
  p = NULL,
  p.criterion = "SICc",
  pmax = 20,
  type.basis = "bspline",
  projs = 0.95,
  verbose = TRUE,
  same.rwild = FALSE,
  ...
)

Arguments

X.fdata

functional observations in the class fdata.

Y

scalar responses for the FLM. Must be a vector with the same number of elements as functions are in X.fdata.

beta0.fdata

functional parameter for the simple null hypothesis, in the fdata class. The argvals and rangeval arguments of beta0.fdata must be the same of X.fdata. If beta0.fdata=NULL (default), the function will test for the composite null hypothesis.

B

number of bootstrap replicates to calibrate the distribution of the test statistic.

n.proj

vector with the number of projections to consider.

est.method

estimation method for β, only used in the composite case. There are three methods:

  • list("\"pc\"") if p is given, then β is estimated by fregre.pc. Otherwise, p is chosen using fregre.pc.cv and the p.criterion criterion.

  • list("\"pls\"") if p is given, β is estimated by fregre.pls. Otherwise, p is chosen using fregre.pls.cv and the p.criterion criterion.

  • list("\"basis\"") if p is given, β is estimated by fregre.basis. Otherwise, p is' chosen using fregre.basis.cv and the p.criterion criterion. Both in fregre.basis and fregre.basis.cv, the same basis for basis.x and basis.b is considered.

p

number of elements for the basis representation of beta0.fdata and X.fdata with the est.method (only composite hypothesis). If not supplied, it is estimated from the data.

p.criterion

for est.method equal to "pc" or "pls", either "SIC", "SICc" or one of the criterions described in fregre.pc.cv. For "basis" a value for type.CV in fregre.basis.cv such as GCV.S.

pmax

maximum size of the basis expansion to consider in when using p.criterion.

type.basis

type of basis if est.method = "basis".

projs

a fdata object containing the random directions employed to project X.fdata. If numeric, the convenient value for ncomp in rdir.pc.

verbose

whether to show or not information about the testing progress.

same.rwild

wether to employ the same wild bootstrap residuals for different projections or not.

...

further arguments passed to create.basis (not rangeval that is taken as the rangeval of X.fdata).

Value

An object with class "htest" whose underlying structure is a list containing the following components:

Author(s)

Eduardo Garcia-Portugues (edgarcia@est-econ.uc3m.es) and Manuel Febrero-Bande (manuel.febrero@usc.es).

References

Cuesta-Albertos, J.A., Garcia-Portugues, E., Febrero-Bande, M. and Gonzalez-Manteiga, W. (2017). Goodness-of-fit tests for the functional linear model based on randomly projected empirical processes. arXiv:1701.08363. https://arxiv.org/abs/1701.08363

Garcia-Portugues, E., Gonzalez-Manteiga, W. and Febrero-Bande, M. (2014). A goodness-of-fit test for the functional linear model with scalar response. Journal of Computational and Graphical Statistics, 23(3), 761–778. http://dx.doi.org/10.1080/10618600.2013.812519

Examples

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
## Not run: 
# Simulated example

set.seed(345678)
t <- seq(0, 1, l = 101)
n <- 100
X <- r.ou(n = n, t = t, alpha = 2, sigma = 0.5)
beta0 <- fdata(mdata = cos(2 * pi * t) - (t - 0.5)^2, argvals = t,
               rangeval = c(0,1))
Y <- inprod.fdata(X, beta0) + rnorm(n, sd = 0.1)

# Test all cases
rp.flm.test(X.fdata = X, Y = Y, est.method = "pc")
rp.flm.test(X.fdata = X, Y = Y, est.method = "pls")
rp.flm.test(X.fdata = X, Y = Y, est.method = "basis", 
            p.criterion = fda.usc::GCV.S)
rp.flm.test(X.fdata = X, Y = Y, est.method = "pc", p = 5)
rp.flm.test(X.fdata = X, Y = Y, est.method = "pls", p = 5)
rp.flm.test(X.fdata = X, Y = Y, est.method = "basis", p = 5)
rp.flm.test(X.fdata = X, Y = Y, beta0.fdata = beta0)

# Composite hypothesis: do not reject FLM
rp.test <- rp.flm.test(X.fdata = X, Y = Y, est.method = "pc")
rp.test$p.values.fdr
pcvm.test <- flm.test(X.fdata = X, Y = Y, est.method = "pc", B = 1e3,
                      plot.it = FALSE)
pcvm.test

# Estimation of beta
par(mfrow = c(1, 3))
plot(X, main = "X")
plot(beta0, main = "beta")
lines(rp.test$beta.est, col = 2)
lines(pcvm.test$beta.est, col = 3)
plot(density(Y), main = "Density of Y", xlab = "Y", ylab = "Density")
rug(Y)

# Simple hypothesis: do not reject beta = beta0
rp.flm.test(X.fdata = X, Y = Y, beta0.fdata = beta0)$p.values.fdr
flm.test(X.fdata = X, Y = Y, beta0.fdata = beta0, B = 1e3, plot.it = FALSE)

# Simple hypothesis: reject beta = beta0^2
rp.flm.test(X.fdata = X, Y = Y, beta0.fdata = beta0^2)$p.values.fdr
flm.test(X.fdata = X, Y = Y, beta0.fdata = beta0^2, B = 1e3, plot.it = FALSE)

# Tecator dataset

# Load data
data(tecator)
absorp <- tecator$absorp.fdata
ind <- 1:129 # or ind <- 1:215
x <- absorp[ind, ]
y <- tecator$y$Fat[ind]

# Composite hypothesis
rp.tecat <- rp.flm.test(X.fdata = x, Y = y, est.method = "pc")
pcvm.tecat <- flm.test(X.fdata = x, Y = y, est.method = "pc", B = 1e3,
                       plot.it = FALSE)
rp.tecat$p.values.fdr[c(5, 10), ]
pcvm.tecat

# Simple hypothesis
zero <- fdata(mdata = rep(0, length(x$argvals)), argvals = x$argvals,
              rangeval = x$rangeval)
rp.flm.test(X.fdata = x, Y = y, beta0.fdata = zero)
flm.test(X.fdata = x, Y = y, beta0.fdata = zero, B = 1e3)

# With derivatives
rp.tecat <- rp.flm.test(X.fdata = fdata.deriv(x, 1), Y = y, est.method = "pc")
rp.tecat$p.values.fdr
rp.tecat <- rp.flm.test(X.fdata = fdata.deriv(x, 2), Y = y, est.method = "pc")
rp.tecat$p.values.fdr

# AEMET dataset

# Load data
data(aemet)
wind.speed <- apply(aemet$wind.speed$data, 1, mean)
temp <- aemet$temp

# Remove the 5% of the curves with less depth (i.e. 4 curves)
par(mfrow = c(1, 1))
res.FM <- depth.FM(temp, draw = TRUE)
qu <- quantile(res.FM$dep, prob = 0.05)
l <- which(res.FM$dep <= qu)
lines(aemet$temp[l], col = 3)

# Data without outliers
wind.speed <- wind.speed[-l]
temp <- temp[-l]

# Composite hypothesis
rp.aemet <- rp.flm.test(X.fdata = temp, Y = wind.speed, est.method = "pc")
pcvm.aemet <- flm.test(X.fdata = temp, Y = wind.speed, B = 1e3,
                       est.method = "pc", plot.it = FALSE)
rp.aemet$p.values.fdr
apply(rp.aemet$p.values.fdr, 2, range)
pcvm.aemet

# Simple hypothesis
zero <- fdata(mdata = rep(0, length(temp$argvals)), argvals = temp$argvals,
              rangeval = temp$rangeval)
flm.test(X.fdata = temp, Y = wind.speed, beta0.fdata = zero, B = 1e3,
         plot.it = FALSE)
rp.flm.test(X.fdata = temp, Y = wind.speed, beta0.fdata = zero)

## End(Not run)

fda.usc documentation built on Feb. 18, 2020, 1:07 a.m.