jasonwgriffin
commented
3 years ago Thank you for your input and feedback! However, this particular line of code is specific to Cohen's d not the subgroups. Currently, sample_size reflects the total number of participants in a study, which for Cohen's d - the standardized mean difference between 2 groups - assumes an equal number of participants in each group (i.e., sample_size/2). IN contrary, for correlations, the sample size is not divided by 2 at the outset. This ensures that for Cohen's d, the variance is computed appropriately given the 2 groups' sample size (see below). For other metrics, like correlation, the sample size is not divided by 2 because of the within-subject nature of that metric; this also ensures that the computation of the variance for correlation is correct within the compute_variance() function (see below)
compute_variance <- function(sample_size, effect_size, es_type, con_table){
if(es_type == "d"){
return(round(((sample_size+sample_size)/((sample_size)*(sample_size))) + ((effect_size^2)/(2*(sample_size+sample_size))),5))
} else if (es_type == "r"){
return(1/(sample_size-3))
}else if (es_type == "or"){
con_table <- data.frame(a = con_table[1],
b = con_table[2],
c = con_table[3],
d = con_table[4])
return((1/con_table$a)+(1/con_table$b)+(1/con_table$c)+(1/con_table$d))
}
}After replicating your parameters, I do notice that the summary output is printing the sample size per group overall (i.e., 33, which is sample_size/2); however, the compute_variance() functions takes sample_size/n_groups as an input to compute the variance for each subgroup assuming they are the same size (i.e., 11 when n_groups = 3). In conclusion, I believe that everything is being calculated correctly, but the output is confusing, I am going to change the summary output of subgroup_power() to clearly show the sample size of group 1 and 2, as well as the sub sample sizes within group 1 and group 2.
Please let me know if you disagree; I of course want this package to be as accurate and correct as possible.
mcweenysean
Hi,
First, thank you for putting together this package - it will come in super handy. I think I may have caught a bug in the
subgroup_power()function relating to sample size. Both in the R version I am using (0.2.1) and on the git page, these lines appear:This is true if and only if you are using 2 groups. When I enter a sample size of 66 with three groups, the expected sample size output shows as 33 per group. I don't know the math super well or what
compute_variance()does exactly, but downstream after dividing by 2, you dividesample_sizeagain by n_groups when calculating variance. Assuming the first argument incompute_variance()is sample size, this would greatly increase variance since you have divided the sample size twice now.Please let me know if anything I said is incorrect or you have any questions. Thank you!