[R-meta] [metafor package] Metaanalysis using lmer.
Viechtbauer, Wolfgang (SP)
wolfg@ng@viechtb@uer @ending from m@@@trichtuniver@ity@nl
Mon Jul 2 16:03:49 CEST 2018
I was in the process of writing when Michael's response arrived. I'll just add my answer to his.
First of all, the subject is a bit misleading -- you are asking about the lme4 package, not metafor.
I do not know what 'yi' stands for, but I wonder how appropriate the use of lmer() is for analyzing your data. This may be worth reading:
But leaving this aside, your model seems rather overparameterized. For example, for "Lead_construct" there are only two levels. Trying to estimate intercept and slope variances over these two levels is asking a bit much.
Also, if either the intercept or slope variance is estimated to be close to 0, the value of the correlation is irrelevant -- the correlation between a variable and something that appears to be constant is in essence arbitrary. But before I would worry about this, I think you first need to consider simplifying your model.
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-
>project.org] On Behalf Of Michael Dewey
>Sent: Monday, 02 July, 2018 15:57
>To: Marcel Lars Meyert; r-sig-meta-analysis using r-project.org
>Subject: Re: [R-meta] [metafor package] Metaanalysis using lmer.
>You seem to have a large number of grouping variables (the ones to the
>right of the | symbol) which means you have few observations within any
>given cell of your design. I think you either need more data or a
>On 02/07/2018 12:19, Marcel Lars Meyert wrote:
>> Hello everyone,
>> Thank you so much for accepting me to this newsletter.
>> As I am fairly new to metaanalysis and especially metaanalysis with R,
>> have a question concering the usage of the lmer-command.
>> The Multilevelanalysis consists of first and second level variables.
>> The first level variable here is: Diff_comp and the second level
>> variables: Lead_construct, Out_alpha, Meta_N, Pub_type, Country.
>> I want to use the lmer-command as following:
>> Test_analysis = lmer(yi ~ Diff_comp + (Diff_comp | Lead_construct) +
>> (Diff_comp | Out_alpha) + (Diff_comp | Meta_N) + (Diff_comp | Pub_type)
>> + (Diff_comp | Country), data = dat)
>> As a result I get following message:
>> Warning message:
>> In optwrap(optimizer, devfun, getStart(start, rho$lower, rho$pp), :
>> convergence code 1 from bobyqa: bobyqa -- maximum number of function
>> evaluations exceeded
>> Additionally, if I print it, I get the following result:
>> Linear mixed model fit by REML ['lmerMod']
>> Formula: yi ~ Out_Breuer + (Out_Breuer | Lead_construct) + (Out_Breuer
>> Out_alpha) + (Out_Breuer | Meta_N) + (Out_Breuer | Pub_type) +
>> (Out_Breuer | Country)
>> Data: dat
>> REML criterion at convergence: 103.1168
>> Random effects:
>> Groups Name Std.Dev. Corr
>> Meta_N (Intercept) 0.000e+00
>> Out_Breuer 6.178e-07 NaN
>> Out_alpha (Intercept) 2.694e-03
>> Out_Breuer 2.507e-02 -0.97
>> Pub_type (Intercept) 3.535e-01
>> Out_Breuer 9.909e-02 -1.00
>> Country (Intercept) 2.835e-07
>> Out_Breuer 1.066e-07 -1.00
>> Lead_construct (Intercept) 2.586e-01
>> Out_Breuer 6.867e-02 -1.00
>> Residual 3.100e-01
>> Number of obs: 165, groups: Meta_N, 41; Out_alpha, 31; Pub_type, 5;
>> Country, 2; Lead_construct, 2
>> Fixed Effects:
>> (Intercept) Out_Breuer
>> -0.45167 0.09203
>> convergence code 1; 0 optimizer warnings; 0 lme4 warnings
>> My question: Why is the correlation almost always -1.00 ?. Did I use
>> command right ?. As I already wrote, I am fairly new to all this and
>> just want to check, whether I am on the right path, before I present
>> false results. If any more information is needed, please tell me and I
>> will respond as soon as possible.
>> Thank you so much in Advance.
>> Best Wishes
>> Marcel Meyert
More information about the R-sig-meta-analysis