simexlme: SIMEX algorithm for linear mixed effects models

In SurvDisc: Discrete Time Survival and Longitudinal Data Analysis

Description

Implementation of the SIMEX algorithm for measurement error models according to Cook and Stefanski.

Usage

simexlme(model, model.model, SIMEXvariable, respvar, grpvar, corform, measurement.error,
  measurement.error.resp, lambda = c(0.5, 1, 1.5, 2), B = 100,
  fitting.method = "quadratic", jackknife.estimation = "quadratic")

Arguments

model	naive model
model.model	dataframe containing all variables in the model
SIMEXvariable	character name of the variable with measurement error. Assumed to be the baseline measurement.
respvar	character name of the response variable. The response is assumed to represent a change from baseline.
grpvar	character name of the grouping variable for the random effects in the model.
corform	formula for the correlation of residual errors within groups. see example
measurement.error	The known standard deviation of measurement errors for SIMEXvariable.
measurement.error.resp	The known stadard deviaiton for respvar
lambda	vector of lambdas for which the simulation step should be done
B	number of iterations for each lambda
fitting.method	fitting method for extrapolation. Only linear or quadratic are recommended.
jackknife.estimation	specifying the extrapolation method for jackknife variance estimation.

Details

See documentation for mcsimex function. This function for lme models was adapted from that function, which is designed to handle linear and generalized linear models, but not lme models. In this function, the measurement error variable must be the baseline value of some measurement and the response is the change from baseline in the same measurement. There is assumed to be one value of this baseline measurement per level of the grouping variable in the mixed effect model. The correlation between the measurement errors for two response values within a subject is assumed to be equal to be equal to the variance of baseline divided by the sum of the variance of baseline and variance of post-baseline errors. For example, for a study measuring the effect of some weight loss treatment, the grouping variable could be subject, the baseline weight is the covariate with measurement error and the response is change from baseline in weight.

Value

An object of class 'simex' which contains:

coefficients	the corrected coefficients of the SIMEX model
SIMEX.estimates	the estimates for every lambda
model	the naive model
measurement.error	the known error standard deviations for SIMEXvariable
B	the number of iterations
extrapolation	the model object of the extrapolation step
fitting.method	the fitting method used in the extrapolation step
residuals	the residuals of the main model
fitted.values	the fitted values of the main model
call	the function call
variance.jackknife	the jackknife variance estimate
extrapolation.variance	the model object of the variance extrapolation
variance.jackknife.lambda	the data set for the extrapolation
variance.asymptotic	the asymptotic variance estimates
theta	the estimates for every B and lambda

Author(s)

John Lawrence,[e-mail], Jianjin Xu, Wolfgang Lederer, Heidi Seibold

References

Cook, J.R. and Stefanski, L.A. (1994) Simulation-extrapolation estimation in parametric measurement error models. Journal of American Statistical Association, 89, 1314 – 1328

See Also

simex,lme

Examples

set.seed(1234)
data("simGFRdata")
simGFR=simGFR[is.element(simGFR$time,c(1:12)/4) & is.element(simGFR$PID,c(1:80)*100),]

fm2=nlme::lme.formula(fixed = cfb ~ time + x1:time + trt + trt:time + trt:x1:time + 0,
                data = simGFR, random = ~time | PID,
                correlation = nlme::corCompSymm(0.5,form = ~time | PID, fixed = TRUE),
                control=nlme::lmeControl(returnObject=TRUE))

(s1 = simexlme(model=fm2, model.model=simGFR[,c("cfb","PID","time","x1","trt")],
                SIMEXvariable="x1",respvar="cfb",grpvar="PID",corform="~time | PID",
                measurement.error=res.sd,measurement.error.resp=res.sd,
                lambda = c(0.5,2),B = 2, fitting.method = "linear",
                jackknife.estimation = FALSE))

plot(s1)

#values of fixed effects used to simulate data
c(fixed.time,fixed.trt,fixed.leGFR,fixed.trttime,fixed.leGFRtrt)

#naive estimates
fm2$coefficients$fixed

#SIMEX corrected estimates
s1$coefficients

Example output

Naive model:
nlme::lme.formula(fixed = cfb ~ time + x1:time + trt + trt:time +     trt:x1:time + 0, data = simGFR, random = ~time | PID, correlation = nlme::corCompSymm(0.5,     form = ~time | PID, fixed = TRUE), control = nlme::lmeControl(returnObject = TRUE))

SIMEX-Variables: x1
Number of Simulations: 2

Coefficients:
       time          trt      time:x1     time:trt  time:x1:trt  
   -0.51693     -0.03342      0.10798      0.14614     -0.03106  

[1] -0.6447911 -0.0478315  0.1333391  0.2186963 -0.0458998
       time         trt     time:x1    time:trt time:x1:trt 
-0.52505122 -0.04379181  0.10882352  0.20892427 -0.04492911 
       time         trt     time:x1    time:trt time:x1:trt 
-0.51693485 -0.03342354  0.10797836  0.14614182 -0.03106170 

SurvDisc documentation built on May 2, 2019, 9:12 a.m.