[R-sig-ME] Nested and crossed random effects

[Reinhold Kliegl]

I predict you will receive constructive feedback if you provide a
dataframe with (simulated) data.
Reinhold Kliegl

On Thu, Jul 30, 2009 at 5:09 AM, Patrick Onyango<pogola at princeton.edu> wrote:
> Dear All,
> I am in transition from SPSS to R and so I am trying to read as much as I
> can to address some of my immediate statistical needs. And so, as with most
> transitions, I am plagued by haste that some of you may find quite sublime;
> but I ask for leniency.
>
> I am trying to model a response variable, response,  as a function of the
> following fixed terms: a, b, c, d, and e; and 3 random effects: f, g, h such
> that g denotes the ID of my sampling subjects sampled over time and the
> subjects are distributed in 5 groups, here denoted by f. An important note
> is that I sampled g as part of a pair with h such that overall I have 131
> samples involving 39 different gs and 45 different hs where h may be or may
> not be at 2 levels of two of the fixed terms, let's for convenience call
> those terms d and e, each with levels 1 and 2. The number of samples among h
> are pretty uneven.
>
> The main idea is to find out what g does given the level of h; and how is
> that influenced by a,b,c as well as vary by the IDs f,g,h where f is the
> highest level of random effects and g and h should probably be crossed
> terms?
>
> This is what I did:
> model<-lmer(response~a+b+c+d+e+(1|f)+(1|f/g)+(1|g)+(1|h), method='ML')
>
> but got the following warning message
> Error: length(f1) == length(f2) is not TRUE
> In addition: Warning messages:
> 1: In g:f:
>  numerical expression has 131 elements: only the first used
> 2: In g:f:
>  numerical expression has 131 elements: only the first used
> 3: In h:f :
>  numerical expression has 131 elements: only the first used
> 4: In h:f :
>  numerical expression has 131 elements: only the first used
>
> Then I tried
> model<-lmer(response~a+b+c+d+e+(1|f)+(1+f|g)+(1|g)+(1|h), method='ML') only
> making one change at the nested term; and got the following Error message
>
> In mer_finalize(ans) : singular convergence (7)
>
> Further, the output from the latter call found a perfect correlation for the
> intercept and slope of g indicating overparametization; and so I sought to
> simplify the model by removing the correlation term and assuming
> homoscedasticity for g in respect to f by replacing the nested term in the
> previous call with (1|f:g), and also adding corr = FALSE in the call
>
> But got the following
> Error: length(f1) == length(f2) is not TRUE
>
> I go in your records as probably having the longest post. An achievement
> indeed.
>
> I need and will be most thankful for help in figuring how to handle my data
> and in deciphering the error messages.
>
> Many thanks in advance; and please let me if further clarifications would
> suffice.
> Patrick
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>