## Loading required package: grid
## Loading 'meta' package (version 3.5-1).
Let's start easy and do a proof-of-concept meta-analysis of my own semsat1a and semsat2 studies:
# meta <- matrix(c(263,1847.21,54.06,264,1729.43,50.99,
# 243,2261.24,94.44,243,2240.65,95.98, 243,2339.60,96.55,243,2206.07,89.30),
# nrow=3, ncol=6, byrow=TRUE,
# dimnames=list(c('semsat1','semsat2NN','semsat2NV'),
# c('Exp.n','Exp.m','Exp.se', 'Ctrl.n','Ctrl.m','Ctrl.se')))
meta <- matrix(c(66, 1766.4, 104.89, 66, 1586.85, 89.17, 81, 2141.39, 142.54,
81, 2163.05, 158.47, 81, 2267.05, 169.95, 81, 2063.75, 128.8), nrow = 3,
ncol = 6, byrow = TRUE, dimnames = list(c("semsat1", "semsat2NN", "semsat2NV"),
c("Exp.n", "Exp.m", "Exp.se", "Ctrl.n", "Ctrl.m", "Ctrl.se")))
meta <- data.frame(meta)
meta.1 <- metacont(Exp.n, Exp.m, Exp.se, Ctrl.n, Ctrl.m, Ctrl.se, studlab = list("semsat1",
"semsat2NN", "semsat2NV"), data = meta, comb.random = TRUE, prediction = TRUE,
sm = "MD")
meta.1
## MD 95%-CI %W(fixed) %W(random)
## semsat1 179.55 [146.34; 212.76] 49.42 33.79
## semsat2NN -21.66 [-68.08; 24.76] 25.30 33.10
## semsat2NV 203.30 [156.86; 249.74] 25.28 33.10
##
## Number of studies combined: k=3
##
## MD 95%-CI z p.value
## Fixed effect model 134.6 [ 111.294; 158.0] 11.302 < 0.0001
## Random effects model 120.8 [ -9.929; 251.5] 1.811 0.0701
## Prediction interval [-1551.774; 1793.4]
##
## Quantifying heterogeneity:
## tau^2 = 12878.6162; H = 5.43 [3.78; 7.81]; I^2 = 96.6% [93%; 98.4%]
##
## Test of heterogeneity:
## Q d.f. p.value
## 58.98 2 < 0.0001
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
metacum(meta.1, pooled = "fixed")
##
## Cumulative meta-analysis (Fixed effect model)
##
## MD 95%-CI p.value tau^2
## Adding semsat1 (k=1) 179.6 [146.3; 212.8] < 0.0001 0
## Adding semsat2NN (k=2) 111.4 [ 84.4; 138.4] < 0.0001 19818.7127
## Adding semsat2NV (k=3) 134.6 [111.3; 158.0] < 0.0001 12878.6162
##
## Pooled estimate 134.6 [111.3; 158.0] < 0.0001 12878.6162
## I^2
## Adding semsat1 (k=1)
## Adding semsat2NN (k=2) 97.9%
## Adding semsat2NV (k=3) 96.6%
##
## Pooled estimate 96.6%
##
## Details on meta-analytical method:
## - Inverse variance method
forest(metacum(meta.1, pooled = "fixed"))
Let's start by loading in and preprocessing our data:
download.file("https://raw.githubusercontent.com/faulconbridge/semsatMeta/master/effectSizes.csv",
"effectSizes.csv", "wget", extra = "--no-check-certificate")
effectSizes <- read.csv("effectSizes.csv", header = TRUE, sep = ",")
repetitions <- effectSizes[!is.na(effectSizes$s.e), ]
View(repetitions)
meta <- metacont(n.e, m.e, s.e, n.c, m.c, s.c, studlab = c(paste(repetitions$Article,
repetitions$Experiment)), data = repetitions, comb.random = TRUE, prediction = TRUE,
sm = "MD")
metacum(meta, pooled = "fixed")
##
## Cumulative meta-analysis (Fixed effect model)
##
## MD 95%-CI p.value tau^2
## Adding Cohene 1 (k=1) 21.00 [-0.2003; 42.20] 0.0522 0
## Adding Black 1 (k=2) 26.46 [24.6805; 28.24] < 0.0001 0
## Adding Black 2 (k=3) 65.90 [64.8339; 66.97] < 0.0001 1884.9907
## Adding Tian 1a (k=4) 64.85 [63.8003; 65.91] < 0.0001 1821.7964
## Adding Tian 1b (k=5) 63.28 [62.2473; 64.31] < 0.0001 1737.3567
## Adding Tian 2 (k=6) 62.23 [61.2120; 63.24] < 0.0001 1656.8091
## Adding Tian 3 (k=7) 56.97 [56.0037; 57.93] < 0.0001 1718.6276
## Adding Black 1 (k=8) 56.97 [56.0028; 57.93] < 0.0001 1716.6435
## Adding Wetherill 1 (k=9) 57.07 [56.1065; 58.03] < 0.0001 1733.1865
## Adding Wetherill 3a (k=10) 57.04 [56.0727; 58.00] < 0.0001 1735.4165
## Adding Wetherill 3b (k=11) 57.10 [56.1359; 58.06] < 0.0001 1748.0063
##
## Pooled estimate 57.10 [56.1359; 58.06] < 0.0001 1748.0063
## I^2
## Adding Cohene 1 (k=1)
## Adding Black 1 (k=2) 0.0%
## Adding Black 2 (k=3) 99.9%
## Adding Tian 1a (k=4) 99.9%
## Adding Tian 1b (k=5) 99.9%
## Adding Tian 2 (k=6) 99.9%
## Adding Tian 3 (k=7) 99.9%
## Adding Black 1 (k=8) 99.8%
## Adding Wetherill 1 (k=9) 99.8%
## Adding Wetherill 3a (k=10) 99.8%
## Adding Wetherill 3b (k=11) 99.8%
##
## Pooled estimate 99.8%
##
## Details on meta-analytical method:
## - Inverse variance method
metacum(meta, pooled = "random")
##
## Cumulative meta-analysis (Random effects model)
##
## MD 95%-CI p.value tau^2
## Adding Cohene 1 (k=1) 21.00 [-0.2003; 42.20] 0.0522 0
## Adding Black 1 (k=2) 26.46 [24.6805; 28.24] < 0.0001 0
## Adding Black 2 (k=3) 45.65 [-3.9748; 95.27] 0.0714 1884.9907
## Adding Tian 1a (k=4) 40.44 [-1.7434; 82.62] 0.0602 1821.7964
## Adding Tian 1b (k=5) 37.33 [ 0.5208; 74.14] 0.0468 1737.3567
## Adding Tian 2 (k=6) 36.11 [ 3.3102; 68.90] 0.0309 1656.8091
## Adding Tian 3 (k=7) 32.19 [ 1.2978; 63.08] 0.0411 1718.6276
## Adding Black 1 (k=8) 34.38 [ 5.0573; 63.70] 0.0216 1716.6435
## Adding Wetherill 1 (k=9) 48.98 [21.0423; 76.91] 0.0006 1733.1865
## Adding Wetherill 3a (k=10) 43.23 [16.4406; 70.02] 0.0016 1735.4165
## Adding Wetherill 3b (k=11) 55.29 [29.4427; 81.14] < 0.0001 1748.0063
##
## Pooled estimate 55.29 [29.4427; 81.14] < 0.0001 1748.0063
## I^2
## Adding Cohene 1 (k=1)
## Adding Black 1 (k=2) 0.0%
## Adding Black 2 (k=3) 99.9%
## Adding Tian 1a (k=4) 99.9%
## Adding Tian 1b (k=5) 99.9%
## Adding Tian 2 (k=6) 99.9%
## Adding Tian 3 (k=7) 99.9%
## Adding Black 1 (k=8) 99.8%
## Adding Wetherill 1 (k=9) 99.8%
## Adding Wetherill 3a (k=10) 99.8%
## Adding Wetherill 3b (k=11) 99.8%
##
## Pooled estimate 99.8%
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
forest(metacum(meta, pooled = "random"))
