na.omit.data.frame.mixmeta function

Handling Missing Values in mixmeta Models

These method functions exclude rows corresponding to units with invalid missing pattern from model frames of class "data.frame.mixmeta". This guarantees the correct handling of missing values while fitting meta-analytical models.

## S3 method for class 'data.frame.mixmeta'
na.omit(object, ...)

## S3 method for class 'data.frame.mixmeta'
na.exclude(object, ...)

Arguments

• object: an object of class "data.frame.mixmeta".
• ...: further arguments passed to or from other methods.

Details

A model frame of class "data.frame.mixmeta" is produced by mixmeta. A call to na.omit or na.exclude removes from the model frame the rows corresponding to studies with invalid missing pattern. In addition, a na.action attribute is added to the model frame, namely a numeric vector corresponding to the removed rows and class "omit" or "exclude", respectively. This information is used by naresid and napredict to deal with missing values in functions such as fitted, residuals, predict and blup, among others.

The definition of missing, identifying an invalid missing pattern, is different in meta-analytical models performed through mixmeta if compared to other regression functions such as lm or glm, in particular for the multivariate case. Specifically, while a unit is removed if at least an observation for one predictor is missing, partially missing outcomes do not prevent the unit to contribute to estimation (see mixmeta). Specific methods na.omit and na.exclude for class "data.frame.mixmeta" allow this different definition.

Returns

These functions returns the model frame object with rows corresponding to units with invalid missing pattern being removed. They also add the related na.action attribute as explained above.

Author(s)

Antonio Gasparrini <antonio.gasparrini@lshtm.ac.uk >

See Also

See na.action, naresid and napredict. See model.frame. See mixmeta-package for an overview of the package and modelling framework.

Examples

# INPUT MISSING VALUES IN PREDICTOR AND ONE RESPONSE
data <- berkey98
data[2,1] <- data[4,3] <- NA
data

# RUN THE MODEL
model <- mixmeta(cbind(PD,AL) ~ pubyear, S=data[5:7], data=data, method="ml")

# SUMMARIZE: NOTE THE NUMBER OF STUDIES AND OBSERVATIONS
summary(model)
df.residual(model)

# EXTRACT THE MODEL FRAME WITH na.pass
model.frame(model, na.action="na.pass")
# EXTRACT THE MODEL FRAME WITH na.omit (DEFAULT)
model.frame(model, na.action="na.omit")

# COMPARE WITH DEFAULT METHOD FOR na.omit
frame <- model.frame(model, na.action="na.pass")
na.omit(frame)
class(frame)
class(frame) <- "data.frame"
na.omit(frame)

# WITH na.exclude
residuals(model)
residuals(update(model, na.action="na.exclude"))