Get model predictions for differences between conditions.
get_difference(
model,
comp,
cond = NULL,
rm.ranef = TRUE,
se = TRUE,
sim.ci = FALSE,
f = 1.96,
return.n.posterior = 0,
print.summary = getOption("itsadug_print")
)
A named list with the two levels to compare.
A named list of the values to use for the other predictor terms. Variables omitted from this list will have the closest observed value to the median for continuous variables, or the reference level for factors.
Logical: whether or not to remove random effects. Default is TRUE. Alternatively a vector of numbers with the mdoelterm number of the random effect(s) to remove. (See notes.)
Logical: whether or not to return the confidence interval or standard error around the estimates.
Logical: Using simultaneous confidence intervals or not
(default set to FALSE). The implementation of simultaneous CIs follows
Gavin Simpson's blog of December 15, 2016:
https://fromthebottomoftheheap.net/2016/12/15/simultaneous-interval-revisited/.
This interval is calculated from simulations based.
Please specify a seed (e.g., set.seed(123)
) for reproducable results.
In addition, make sure to specify at least 200 points for each smooth
for the simulations when using simultaneous CI.
Note: in contrast with Gavin Simpson's code, here the Bayesian posterior
covariance matrix of the parameters is uncertainty corrected
(unconditional=TRUE
) to reflect the uncertainty on the estimation of
smoothness parameters.
A number to scale the standard error. Defaults to 1.96, resulting in 95% confidence intervals. For 99% confidence intervals use a value of 2.58.
Numeric: N samples from
the posterior distribution of the fitted model are returned.
Default value is 0 (no samples returned).
Only workes when sim.ci=TRUE
.
Logical: whether or not to print a summary of the
values selected for each predictor.
Default set to the print info messages option
(see infoMessages
).
Returns a data frame with the estimates of the difference and optionally the confidence intervals around that estimate.
Other, not specified effects and random effects are generally canceled
out, when calculating the difference. When the predictors that
specify the conditions to compare are involved in other interactions
or included as random slopes, it may be useful to specify the values
of other predictors with cond
or remove the random effects with
rm.ranef
.
Other Model predictions:
get_coefs()
,
get_fitted()
,
get_modelterm()
,
get_predictions()
,
get_random()
# NOT RUN {
data(simdat)
# first fit a simple model:
m1 <- bam(Y ~ Group+te(Time, Trial, by=Group), data=simdat)
# get difference estimates:
diff <- get_difference(m1, comp=list(Group=c('Adults', 'Children')),
cond=list(Time=seq(0,500,length=100)))
head(diff)
# }
Run the code above in your browser using DataLab