Hello!
I am calculating effect sizes for a research synthesis of the literature (e.g. Hedges' g), but many of the original publications do not provide the sample size (N) by subgroups. I see that the formula requires this N to calculate effect sizes from means and standard deviations, or from t, or F....
- Can I assume that the N by subgroup must be the Total N divided by the number of subgroups/cells (e.g. high vs low levels)?
- Is there a way to calculate from other statistics, that does not require this subgroup N?
- Any other suggestion? I could contact the authors for each of the cases, but I risk them not responding or now knowing anymore...
Many thanks in advance!
Hello Miriam,
I'd always try contacting authors, in the hope that you'd get the information you need.
It may be possible to infer the number of cases per group (especially when there are only two groups involved in a comparison), if the authors have included mean & SD for each group and overall. However, this isn't foolproof, as the precision with which the summary statistics are reported can affect your ability to find the right answer.
Attached is an excel worksheet which you may use as an estimator for two-group comparisons. You'll enter the correct total N, and the Mean & SD for each group, and the total (orange cells).
Then, try different combinations of group n values (such that each try for n1 and n2 sums to N), and look at the final two cells (discrepancy for total Mean estimate and total SD estimate). Whichever combination of group n values yields the lowest discrepancy is likely the correct values for group sizes.
Good luck with your work.
The workbooks of Meta-Essentials calculate effect sizes based on t-value or F-values.
You may freely download Meta-Essentials.
Typically, the power is optimal when all groups have the same sample size (one exception: binomial models with considerably different proportions in the groups; here it would be better to have more data in groups with proportions closer to 0 or 1).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size