# Do I really need correlation coefficients when computing effect sizes from pretest-posttest studies?

I am computing effect sizes from various studies for a meta-analysis. Most of those studies follow a typical pretest-posttest design (repeated measures), thus I compute the effect size as a "Standardized Mean Change".

I am using the metafor package in R to compute these effect sizes, but it asks for the correlation coefficient (𝑟, ri in metafor) between pre and posttest scores.

My formula is something like this:

escalc(data = df, measure = "SMCR", m1i = mean_pre, m2i = mean_post,
sd1i = sd_pre, sd2i = sd_post, ni = n_subjects, append = T)
# Error in escalc.default(data = df, measure = "SMCR"...  :
#     Cannot compute outcomes. Check that all of the required
#     information is specified via the appropriate arguments.


→ It requires ri (correlation coefficient).

The problem is that no study ever reports correlation coefficients for a pretest-posttest design. And I don't think I can compute it from the summary statistics (M, SD). How should I proceed?

• You could carry out a sensitivity analysis using a range of plausible values and see how that affects the outcome. You could try contacting the authors to see if any of them can tell you the value of $r$. – mdewey Jul 12 '17 at 16:38
• An alternative option is that you can calculate the effect sizes and their variances yourself, circumventing the defaults in metafor. Ideally you should seek to include the correlation coefficients if possible, but if not you can work to calculate $d_z$ and its variance. – Matt Barstead Jul 13 '17 at 23:23
• How did you compute standardized mean change and what is SMCR ? What are n_sizes for various studies. – Subhash C. Davar Jul 18 '17 at 16:08
• @subhashc.davar: SMCR is the standardized mean change abbreviation in library(metafor). I am not sure about the precise calculations happening behind. n in this case is the number of subjects in both pretest and posttest (as it should be exactly the same sample size). If you have various studies, the effect size is calculated independently for each study. – iNyar Jul 19 '17 at 11:46
• Are you interested in meta-analysis or background process ? What are the results ? – Subhash C. Davar Jul 20 '17 at 5:45