SurvDisc (version 0.1.1)

simexlme: SIMEX algorithm for linear mixed effects models

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.

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

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.

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

# NOT RUN {
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
# }