# Code to illustrate inconsistency between AIC and summary output

library(metafor)

# setwd("c:\\My Documents\\Consulting\\SC&IT\\Hasintha Wijesekara\\data")
#---------------------------------------------------------------------------------------------
# Data management section
data0=read.table("Final final data set_Hasintha.txt",header=T,sep="\t",row.names = NULL )
head(data0)
names(data0)

data0=escalc(measure="MD",m1i=treat,sd1i=sdt,n1i=nt,
             m2i=control,sd2i=sdc,n2i=nc,vtype="HO",
             append=T,data=data0) 

# Remove 1 extreme observation
data1=subset(data0,Citation!="Pérez-de-Mora et al., 2006")
nrow(data1)

# Remove records with missing values for random effects - metafor fails to execute otherwise
data2=subset(data1,!is.na(data1$rate) & !is.na(data1$years))
nrow(data2)
nrow(na.omit(data2[,-14])) # removing clay variable too

# Base model
f1=rma.mv(yi,vi,mods = ~rate + years + depth.mid , random = ~ 1 | Citation/soil/rate/years, data=data2,method="ML",na.exclude=T)
summary(f1)

# Add problem variable
f2=rma.mv(yi,vi,mods = ~rate + years + depth.mid + biosolid.type, random = ~ 1 | Citation/soil/rate/years, data=data2,method="ML",na.exclude=T)
summary(f2)

AIC(f1,f2)

# Inconsistency is that based on AIC being much lower model 2 with the addition of the new variable biosolid.type
# is a much better model and with a 200 reduction in AIC it should be a very significant variable.
# But in the summary output the two levels for biosolid.type are not very different to each other nor are they 
# significantly different to the reference category

table(data2$biosolid.type)

# So AIC says biosolid.type  is an important variable, the summary table says it is not.
# They should be in agreement, so what is going on?
# Am I missing something?

# Thanks,
# Kim