Combining summary estimates and CI's from multiple metafor() models

[Ryan Byrnes]

Sorry for the confusing subject line but I'm conducting a meta-analysis where I am analyzing multiple response variables (1, 2, 3, 4) against livestock grazing systems (a, b and c). I am interested in producing some summary forest plots using the metafor package and the forest() function that would display the summary statistics of the estimate and confident interval from summary(model). Ideally, I would be able to have combined plots variable by system. So, one plot for system a, that would display the overall model estimate and CI of each response variable. I had been able to do this in the past but am unable to do this with my current coding:

estimates.CN <- c(coef(m1.SOC.cn),

                  coef(m1.BD.cn), 

                  coef(m1.CN.cn),

                  coef(m1.TN.cn))

variances.CN <- c(vcov(m1.SOC.cn),

                  vcov(m1.BD.cn),

                  vcov(m1.CN.cn),

                  vcov(m1.TN.cn))

# create vector with labels

labels.bd <- c("Soil Organic Carbon", 

               "Bulk Density", 

               "C:N Ratio", 

               "Total Nitrogen")

# forest plots

forest(estimates.CN, 

       variances.CN, 

       slab=labels.CN)

Where these series of errors and warnings are porduced:

Error in forest.default(estimates.CN, variances.CN, slab = labels.CN) : 

  Length of 'yi' does not match the length of 'vi', 'sei', or the ('ci.lb', 'ci.ub') pairs.

In addition: Warning messages:

1: In sqrt(vi) : NaNs produced

2: In yi - qnorm(alpha/2, lower.tail = FALSE) * sqrt(vi) :

  longer object length is not a multiple of shorter object length

3: In sqrt(vi) : NaNs produced

4: In yi + qnorm(alpha/2, lower.tail = FALSE) * sqrt(vi) :

  longer object length is not a multiple of shorter object length

I've been trying to diagnose this for the better half of a day and not understanding exactly the inner workings of metafor() is making this task difficult. Does anyone have any suggestions? My individual response variable model are fine and their forest plots are as well fine.

Thanks in advance!

Ryan