random: Specify a random intercept model in a GAMLSS formula

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/random.r

Description

They are two functions for fitting random effects wthin a GAMLSS model, random() and re().

The function random() is based on the original random() function of Trevor Hastie in the package gam. TIn our version the function has been modified to allow a "local" maximum likelihood estimation of the smoothing parameter lambda. This method is equivalent to the PQL method of Breslow and Clayton (1993) applied at the local iterations of the algorithm. In fact for a GLM model and a simple random effect it is equivalent to glmmPQL() function in the package MASS see Venables and Ripley (2002). Venables and Ripley (2002) claimed that this iterative method was first introduced by Schall (1991). Note that in order for the "local" maximum likelhood estimation procedure to operate both argument df and lambda has to be NULL.

The function re() is an interface for calling the lme() function of the package nlme. This gives the user the abilty to fit comlpicated random effect models while the assumtion of the normal distribution for the response variable is relaxed. The theoretical justification cames again from the fact that this is a PQL method, Breslow and Clayton (1993).

Usage

1
2
3
4
random(x, df = NULL, lambda = NULL, start=10)

re(fixed = ~1, random = NULL, correlation = NULL, method = "ML",
          level = NULL, ...)

Arguments

x

a factor

df

the target degrees of freedom

lambda

the smoothing parameter lambda which can be viewed as a shrinkage parameter.

start

starting value for lambda if local Maximul likelihood is used.

fixed

a formula specify the fixed effects of the lme() model. This, in most cases can be also included in the gamlss parameter formula

random

a formula or list specufying the random effect part of the model as in lme() function

correlation

the correlation structure of the lme() model

method

which method, "ML" (the default), or "REML"

level

this argument has to be set to zero (0) if when use predict() you want to get the marginal contribution

...

this can be used to pass arguments for lmeControl()

Details

The function random() can be seen as a smoother for use with factors in gamlss(). It allows the fitted values for a factor predictor to be shrunk towards the overall mean, where the amount of shrinking depends either on lambda, or on the equivalent degrees of freedom or on the estimated sigma parameter (default). Similar in spirit to smoothing splines, this fitting method can be justified on Bayesian grounds or by a random effects model. Note that the behavier of the function is different from the original Hastie function. Here the function behaves as follows: i) if both df and lambda are NULL then the PQL method is used ii) if lambda is not NULL, lambda is used for fitting iii) if lambda is NULL and df is not NULL then df is used for fitting.

Since factors are coded by model.matrix() into a set of contrasts, care has been taken to add an appropriate "contrast" attribute to the output of random(). This zero contrast results in a column of zeros in the model matrix, which is aliased with any column and is hence ignored.

The use of the function re() requires knowledge of the use of the function lme() of the package nlme for the specification of the appropriate random effect model. Some care should betaken whether the data set is

Value

x is returned with class "smooth", with an attribute named "call" which is to be evaluated in the backfitting additive.fit() called by gamlss()

Author(s)

For re() Mikis Stasinopoulos and Marco Enea and for random() Trevor Hastie (amended by Mikis Stasinopoulos),

References

Breslow, N. E. and Clayton, D. G. (1993) Approximate inference in generalized linear mixed models. Journal of the American Statistical Association 88, 9???25.

Chambers, J. M. and Hastie, T. J. (1991). Statistical Models in S, Chapman and Hall, London.

Pinheiro, Jose C and Bates, Douglas M (2000) Mixed effects models in S and S-PLUS Springer.

Rigby, R. A. and Stasinopoulos D. M. (2005). Generalized additive models for location, scale and shape,(with discussion), Appl. Statist., 54, part 3, pp 507-554.

Rigby, R. A., Stasinopoulos, D. M., Heller, G. Z., and De Bastiani, F. (2019) Distributions for modeling location, scale, and shape: Using GAMLSS in R, Chapman and Hall/CRC. An older version can be found in https://www.gamlss.com/.

Schall, R. (1991) Estimation in generalized linear models with random effects. Biometrika 78, 719???727.

Stasinopoulos D. M. Rigby R.A. (2007) Generalized additive models for location scale and shape (GAMLSS) in R. Journal of Statistical Software, Vol. 23, Issue 7, Dec 2007, https://www.jstatsoft.org/v23/i07/.

Stasinopoulos D. M., Rigby R.A., Heller G., Voudouris V., and De Bastiani F., (2017) Flexible Regression and Smoothing: Using GAMLSS in R, Chapman and Hall/CRC.

(see also https://www.gamlss.com/).

Venables, W. N. and Ripley, B. D. (2002) Modern Applied Statistics with S. Fourth edition. Springer.

See Also

gamlss, gamlss.random

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
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
#------------- Example 1 from Pinheiro and Bates (2000) page 15-----------------
# bring nlme
library(nlme)
data(ergoStool)
# lme model
l1<-lme(effort~Type, data=ergoStool, random=~1|Subject, method="ML")
# use random() 
t1<-gamlss(effort~Type+random(Subject), data=ergoStool )
# use re() with fixed effect within re()
t2<-gamlss(effort~re(fixed=~Type, random=~1|Subject), data=ergoStool )
# use re() with fixed effect in gamlss formula
t3<-gamlss(effort~Type+re(random=~1|Subject), data=ergoStool )
# compare lme fitted values with random
plot(fitted(l1), fitted(t1))
# compare lme fitted values with random
plot(fitted(l1), fitted(t2))
lines(fitted(l1), fitted(t3), col=2)
# getting the fitted coefficients 
getSmo(t2)
#-------------------------------------------------------------------------------
## Not run: 
#-------------Example 2 Hodges data---------------------------------------------
data(hodges)
plot(prind~state, data=hodges)
m1<- gamlss(prind~random(state), sigma.fo=~random(state), nu.fo=~random(state), 
            tau.fo=~random(state), family=BCT, data=hodges)
m2<- gamlss(prind~re(random=~1|state), sigma.fo=~re(random=~1|state), 
            nu.fo=~re(random=~1|state), tau.fo=~re(random=~1|state), family=BCT, 
            data=hodges)
# comparing the fitted effective degrees of freedom
m1$mu.df
m2$mu.df
m1$sigma.df
m2$sigma.df
m1$nu.df
m2$nu.df
m1$tau.df
m2$tau.df
# random effect for tau is not needed
m3<- gamlss(prind~random(state), sigma.fo=~random(state), nu.fo=~random(state),  
            family=BCT, data=hodges, start.from=m1)
plot(m3)
# term plots work for random but not at the moment for re()
op <- par(mfrow=c(2,2))
term.plot(m3, se=TRUE)
term.plot(m3, se=TRUE, what="sigma")
term.plot(m3, se=TRUE, what="nu")
par(op)
# getting information from a fitted lme object
coef(getSmo(m2))
ranef(getSmo(m2))
VarCorr(getSmo(m2))
summary(getSmo(m2))
intervals(getSmo(m2))
fitted(getSmo(m2))
fixef(getSmo(m2))
#  plotting 
plot(getSmo(m2))
qqnorm(getSmo(m2))
#----------------Example 3 from Pinheiro and Bates (2000) page 42---------------
data(Pixel)
l1 <- lme(pixel~ day+I(day^2), data=Pixel, random=list(Dog=~day, Side=~1),
          method="ML")
# this will fail 
#t1<-gamlss(pixel~re(fixed=~day+I(day^2), random=list(Dog=~day, Side=~1)), 
#           data=Pixel)
# but this  is working 
t1<-gamlss(pixel~re(fixed=~day+I(day^2), random=list(Dog=~day, Side=~1), 
                    opt="optim"), data=Pixel)
plot(fitted(l1)~fitted(t1))
#---------------Example 4 from Pinheiro and Bates (2000)page 146----------------
data(Orthodont)
l1 <- lme(distance~ I(age-11), data=Orthodont, random=~I(age-11)|Subject,
           method="ML")

t1<-gamlss(distance~I(age-11)+re(random=~I(age-11)|Subject), data=Orthodont)
plot(fitted(l1)~fitted(t1))
# checking the model
plot(t1)
wp(t1, ylim.all=2)
# two observation fat try LO
t2<-gamlss(distance~I(age-11)+re(random=~I(age-11)|Subject,  opt="optim", 
     numIter=100), data=Orthodont, family=LO)
plot(t2)
wp(t2,ylim.all=2)
# a bit better but not satisfactory Note that  3 paramters distibutions fail
#------------example 5 from Venable and Ripley (2002)--------------------------
library(MASS)
data(bacteria)
summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 | ID,
                family = binomial, data = bacteria))
s1 <- gamlss(y ~ trt + I(week > 2)+random(ID), family = BI, data = bacteria)
s2 <- gamlss(y ~ trt + I(week > 2)+re(random=~1|ID), family = BI, 
             data = bacteria)
s3 <- gamlss(y ~ trt + I(week > 2)+re(random=~1|ID, method="REML"), family = BI, 
             data = bacteria)
# the esimate of the random effect sd sigma_b 
sqrt(getSmo(s1)$tau2)
getSmo(s2)
getSmo(s3)
#-------------Example 6 from Pinheiro and Bates (2000) page 239-244-------------
# using corAR1()
data(Ovary)
# AR1 
l1 <- lme(follicles~sin(2*pi*Time)+cos(2*pi*Time), data=Ovary, 
          random=pdDiag(~sin(2*pi*Time)), correlation=corAR1())
# ARMA
l2 <- lme(follicles~sin(2*pi*Time)+cos(2*pi*Time), data=Ovary, 
          random=pdDiag(~sin(2*pi*Time)), correlation=corARMA(q=2))
# now gamlss
# AR1 
t1 <- gamlss(follicles~re(fixed=~sin(2*pi*Time)+cos(2*pi*Time), 
                         random=pdDiag(~sin(2*pi*Time)),
                         correlation=corAR1()), data=Ovary)
plot(fitted(l1)~fitted(t1))
# ARMA
t2 <- gamlss(follicles~re(fixed=~sin(2*pi*Time)+cos(2*pi*Time), 
                          random=pdDiag(~sin(2*pi*Time)),
                          correlation=corARMA(q=2)), data=Ovary)
plot(fitted(l2)~fitted(t2))
AIC(t1,t2)
wp(t2, ylim.all=1)
#-------------------------------------------------------------------------------  

## End(Not run)

gamlss documentation built on March 31, 2021, 5:10 p.m.