[R-meta] min sample size for three random variables in meta-analysis

Thu, Jul 28, 2022 3:41 AM

Dear Jose,

With this little data, you are not going to get accurate estimates of the variance components (in fact, the model may even be overparameterized, but I cannot easily check this without a fully reproducible example). Hence, I wouldn't put much trust into the findings.

In general, I am not aware of any useful rules though that one could use to decide on an appropriate model complexity in this context. In the multilevel literature, one sometimes finds suggestions that there should be at least a certain number (e.g., 10) of upper-level units (e.g., schools) before one can consider using random effects at that level. But these rules are typically overly simplistic and have not been examined in the meta-analytic context as far as I know.

You might have to stick to a simpler (e.g., standard random-effects model) for this analysis and note that phylogenetic relatedness could not be accounted for due to the small sample size.

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On
Behalf Of Jose Valdebenito
Sent: Friday, 22 July, 2022 16:50
To: Michael Dewey
Cc: r-sig-meta-analysis at r-project.org
Subject: Re: [R-meta] min sample size for three random variables in meta-analysis

Dear Michael,
Thanks for your reply. By sample size I mean small number of primary
studies. k means number of effect sizes.
In the example below there were 10 effect sizes, 7 studies, 9 species
ID, and 9 species associated with a vcv value.

head(df, 10)

species_2 | spp | study.ID | L_d | L_v
Grus_americana | Grus_americana | p.b.4 | 0.000 | 0.135
Loxia_leucoptera | Loxia_leucoptera | p.b.15 | 1.057 | 0.068
Centrocercus_urophasianus | Centrocercus_urophasianus | p.b.60 | 0.384 | 0.021
Pica_nuttalli | Pica_nuttalli | p.b.8 | -2.042 | 0.037
Acrocephalus_paludicola | Acrocephalus_paludicola | 17 | 0.000 | 0.068
Acrocephalus_paludicola | Acrocephalus_paludicola | 17 | -4.978 | 0.334
Lepidothrix_coronata | Lepidothrix_coronata | 20 | -2.043 | 0.088
Erithacus_rubecula | Erithacus_rubecula | 35 | -1.723 | 0.401
Fringilla_coelebs | Fringilla_coelebs | 35 | -2.697 | 0.111
Parus_caeruleus | Parus_caeruleus | 35 | 4.862 | 0.559

Thanks,
Jose
On Fri, 22 Jul 2022 at 08:19, Michael Dewey <lists at dewey.myzen.co.uk> wrote:

Dear Jose

Unfortunately this is a plain text list and your post in HTML has got
scrambled so as to be almost unreadable.

Just to clarify, when you say small sample size do you mean a small
number of primary studies or that individual studies were based on small
numbers? Is k the number of primary studies?

Michael

On 21/07/2022 15:42, Jose Valdebenito wrote:

Hi,

I am conducting a (multi-level) meta-analysis with some sample sizes that
are a bit small.

I was wondering if there is a kind of rule of thumb regarding minimum
sample size number, and the maximum number of random variables I can add to
the models.

I would like to add 3 (as suggested in Cinar et al. Methods Ecol Evol.
2022;13:383?95), but I am afraid that might incur in overparameterization
if I have, say k=10?

These are the random variables:
-Phylogenetic relatedness
-Study ID
-Species ID

Also, would it be any different in a meta-regression?

Here I paste the model and dataset so you can see:

m2 <- rma.mv(L_d, L_v,
                         data = df,
                          random = list(~1|species_2,
                                                ~1|study.ID,
                                                ~1|spp),
                           R = list(species_2 = vcv), digits = 3)

head(df, 10)

species_2                                 spp
            study.ID         L_d         L_v
Grus_americana                       Grus_americana
p.b.4             0.000       0.135
Loxia_leucoptera                      Loxia_leucoptera
p.b.15          1.057        0.068
Centrocercus_urophasianus    Centrocercus_urophasianus   p.b.60
0.384        0.021
Pica_nuttalli                              Pica_nuttalli
         p.b.8            -2.042      0.037
Acrocephalus_paludicola          Acrocephalus_paludicola       17
     0.000      0.068
Acrocephalus_paludicola          Acrocephalus_paludicola       17
       -4.978      0.334
Lepidothrix_coronata                Lepidothrix_coronata              20
              -2.043     0.088
Erithacus_rubecula                   Erithacus_rubecula                35
               -1.723      0.401
Fringilla_coelebs                       Fringilla_coelebs
  35                 -2.697      0.111
Parus_caeruleus                       Parus_caeruleus                    35
                4.862      0.559

Thanks in advance,
Joe

[R-meta] min sample size for three random variables in meta-analysis

Thread (4 messages)