itsadug (version 2.0)

check_resid: Inspect residuals of regression models.

Description

Inspect residuals of regression models.

Usage

check_resid(model, AR_start = NULL, split_pred = NULL, ask = TRUE,
  select = 1:4)

Arguments

model

A regression model, resulting from the functions gam or bam, or lm, glm, lmer, or glmer.

AR_start

Defaults to NULL. Only use this when the model was run in an old versions of package mgcv and the function cannot retrieve the used AR.start values from the model. When an error is shown with newer versions of mgcv

split_pred

A names list indicating time series in the data.

ask

Logical: whether or not to show the plots one by one. Defaults to TRUE. When set to FALSE, make sure to have specified sufficient rows and columns to show the X plots. Alternatively, use select to plot only specific plots.

select

Vector or numeric value indicating which plots to return (see Notes). Defaults to 1:4 (all).

Note

• Plot 1: distribution of residuals with QQ norm plot.
• Plot 2: distribution of residuals with density plot.
• Plot 3: ACF plot of residuals. In case an AR1 model is included, the gray lines indicate standard residuals, and the thick black lines indicate AR1 corrected residuals.
• Plot 4 (optional): In case thesplit_predpredictors are specified an ACF plot averaged over the time series is produced. dashed lines indicate the maximum and minimum time series (w.r.t. lag 2), the solid lines the 25% and 75% quantiles (w.r.t. lag 2) and the bars the mean of all time series.

See the examples on how to specify a selection of these plots.

See Also

Other Model evaluation: diagnostics, plot_modelfit

Examples

data(simdat)

# Add start event column:
simdat <- start_event(simdat, event=c("Subject", "Trial"))
head(simdat)

# bam model with AR1 model (toy example, not serious model):
m1 <- bam(Y ~ Group + te(Time, Trial, by=Group), 
   data=simdat, rho=.5, AR.start=simdat$start.event)

# Warning, no time series specified:
check_resid(m1)

# Time series specified, results in a "standard" ACF plot, 
# treating all residuals as single time seriesand,
# and an ACF plot with the average ACF over time series:
check_resid(m1, split_pred=list(Subject=simdat$Subject, Trial=simdat$Trial))
# Note: residuals do not look very good.
# Alternative (results in the same, see help(acf_resid) ):
check_resid(m1, split_pred="AR.start")

# Define larger plot window (choose which line you need):
dev.new(width=8, height=8) # on windows or mac
quartz(,8,8)               # on mac
x11(width=8, height=8)     # on linux or mac

par(mfrow=c(2,2), cex=1.1)
check_resid(m1, split_pred="AR.start", ask=FALSE)