[R-sig-ME] [FORGED] Types of residuals in lme4

[Burcu Darst]

Thank you, Rolf and James! That is very helpful to know!

On Jul 5, 2018, at 8:39 PM, James Uanhoro <james.uanhoro using gmail.com<mailto:james.uanhoro using gmail.com>> wrote:

In addition to Rolf Turner's point, you can get the random effects using
the ranef() function.

ranef(test)$DBID and ranef(test)$familyID should do it.

James


On 07/05/2018 08:40 PM, Rolf Turner wrote:
On 06/07/18 02:10, Burcu Darst via R-sig-mixed-models wrote:
Dear All,

I am using lme4 to get residuals from a model that has two random
intercepts, as shown below:

test = lmer(continuousOutcome ~ age + sex + (1|DBID) + (1|familyID),
data, na.action = na.exclude)


I’ve tried extracting all of the following types of residuals, but
the only differences I observe between these approaches are due to
scaling; i.e., residuals do not differ by residual type.

resids = as.data.table(residuals(test,type = "pearson", scaled = TRUE))
resids = as.data.table(residuals(test,type = "working", scaled = TRUE))
resids = as.data.table(residuals(test,type = "response", scaled = TRUE))
resids = as.data.table(residuals(test,type = "deviance", scaled = TRUE))
resids = as.data.table(residuals(test,type = "pearson"))
resids = as.data.table(residuals(test,type = "working"))
resids = as.data.table(residuals(test,type = "response"))
resids = as.data.table(residuals(test,type = "deviance"))


Is this an expected result when using lme4 to obtain residuals from
mixed models? I want to ensure that the residuals I am obtaining are
observation level (which they appear to be) and that they account for
the two random intercepts (which I believe they do, since they differ
if I exclude one of the random intercepts). Also, is it possible to
get higher level residuals from lme4, such as individual or family
level residuals (which are the two random intercepts included in my
model)?

I would greatly appreciate any help!


I am no expert --- and younger and wiser heads may correct me --- but
it is my understanding that "types" of residuals are relevant only in
the context of *generalised* linear models (mixed or "straight").  For
linear models (mixed or "straight") a residual is a residual is a
residual.  (And a caterpillar is a tractor. :-) )

cheers,

Rolf Turner




	[[alternative HTML version deleted]]