Skip to content

[R-meta] Meta-analysis with missing data

3 messages · Mekides Gardi, Gabriele Midolo, Wolfgang Viechtbauer

Mekides Gardi
# 
Dear All,

Seeking help on meta-analysis. I am conducting a meta-analysis on the  
response of two cereal crops to elevated CO2. Most of the studies  
didn't report SD or SE. I have the mean values for both the control  
and experimental group and sample size as well.
I read on literature there is still a way to do meta-analysis without  
SD or SE. Could you provide me with guides on how to do it in R with  
metafor package?

Thanks in advance.

Mekides
Gabriele Midolo
# 
Dear Mekides,

I think this is a common issue in agricultural and ecological
meta-analyses. But I am not sure the solution to your problem depends on
metafor. However, based on my experience, there are few options to be
considered (with the following order): 1) to contact the authors of the
primary studies and ask them to provide the SD of the means; 2) to compute
the missing SD via the ?impute_SD? function of the R package metagear
(Lajeunesse, 2016), in case few values are missing; 3) you might consider
to perform an "unweighed meta-analysis" including also the studies without
SD to compare results with a meta-analysis including studies with available
SD only? See an example in e.g. of Humbert et al. 2016 Global Change
Biology, 22(1), 110-120. (Figure 1).

Hope this helps,
Gabriele.


Il gio 28 mar 2019, 16:06 Mekides Gardi <mekides.gardi at uni-hohenheim.de> ha
scritto:

  
  


  


      

    

    

      
      

        


  

        

      Wolfgang Viechtbauer
    

    

      
      #
    

  


  

      
If the outcome measure is a response ratio (log ratio of means), then one approach we have used before is to look at the studies that do report both means and SDs and see what the (average or median) coefficient of variation is in those studies. For the studies that do not report SDs, we then imputed the sampling variances using this value. See:

Hoeksema, J. D., Bever, J. D., Chakraborty, S., Chaudhary, V. B., Gardes, M., Gehring, C. A., Hart, M. M., Housworth, E. A., Kaonongbua, W., Klironomos, J. N., Lajeunesse, M. J., Meadow, J., Milligan, B. G., Piculell, B. J., Pringle, A., R?a, M. A., Umbanhowar, J., Viechtbauer, W., Wang, Y.-W., Wilson, G. W. T., & Zee, P. C. (2018). Evolutionary history of plant hosts and fungal symbionts predicts the strength of mycorrhizal mutualism. Communications Biology, 1(1), 116.

One can also do quite fancy modeling to deal with this issue:

https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2017-October/000333.html

Also relevant in this context is:

Furukawa, T. A., Barbui, C., Cipriani, A., Brambilla, P., & Watanabe, N. (2006). Imputing missing standard deviations in meta-analyses can provide accurate results. Journal of Clinical Epidemiology, 59(1), 7-10.

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Gabriele Midolo
Sent: Friday, 29 March, 2019 15:03
To: Mekides Gardi
Cc: r-sig-meta-analysis at r-project.org
Subject: Re: [R-meta] Meta-analysis with missing data

Dear Mekides,

I think this is a common issue in agricultural and ecological
meta-analyses. But I am not sure the solution to your problem depends on
metafor. However, based on my experience, there are few options to be
considered (with the following order): 1) to contact the authors of the
primary studies and ask them to provide the SD of the means; 2) to compute
the missing SD via the ?impute_SD? function of the R package metagear
(Lajeunesse, 2016), in case few values are missing; 3) you might consider
to perform an "unweighed meta-analysis" including also the studies without
SD to compare results with a meta-analysis including studies with available
SD only? See an example in e.g. of Humbert et al. 2016 Global Change
Biology, 22(1), 110-120. (Figure 1).

Hope this helps,
Gabriele.

Il gio 28 mar 2019, 16:06 Mekides Gardi <mekides.gardi at uni-hohenheim.de> ha
scritto: