#Sample data on which to test the code.
#Calculate the necessary mean, standard deviation, and sample size for all three datasets.
#You could have your data in columns, too; this is just how I set up the sample data.
#The objects you've created will be used as follows:
#mes(meanA, meanB, standarddeviationA, standarddeviationB, samplesizeA, samplesizeB)
mes(m1,m2,sd1,sd2,l1,l2) #compares data1 and data2
mes(m1,m3,sd1,sd3,l1,l3) #compares data1 and data3
mes(m2,m3,sd2,sd3,l2,l3) #compares data2 and data3
Results for the last comparision (2 vs. 3) look like this:
EFFECT SIZE CALCULATION (FOR SINGLE INPUT) Mean Differences ES: d [ 95 %CI] = 0.37 [ -0.8 , 1.55 ] var(d) = 0.29 p-value(d) = 0.5 U3(d) = 64.6 % CLES(d) = 60.44 % Cliff's Delta = 0.21 g [ 95 %CI] = 0.35 [ -0.75 , 1.45 ] var(g) = 0.25 p-value(g) = 0.5 U3(g) = 63.71 % CLES(g) = 59.79 % Correlation ES: r [ 95 %CI] = 0.18 [ -0.44 , 0.69 ] var(r) = 0.07 p-value(r) = 0.55 z [ 95 %CI] = 0.19 [ -0.47 , 0.84 ] var(z) = 0.09 p-value(z) = 0.55 Odds Ratio ES: OR [ 95 %CI] = 1.97 [ 0.23 , 16.61 ] p-value(OR) = 0.5 Log OR [ 95 %CI] = 0.68 [ -1.45 , 2.81 ] var(lOR) = 0.96 p-value(Log OR) = 0.5 Other: NNT = 4.79 Total N = 14
For my application, I used Cohen's d (the first entry, highlighted in bold above). For additional information on effect sizes (not just Cohen's d), Cohen (1992) is a very useful reference.