Tools for Those Who Summarize the Evidence Base
I intend to conduct a meta analysis of quantitative studies which explore the predictors (barriers and facilitators) related to attendance and adherence. I am a bit confused as to whether 'r' is the correct effect size index to be used. My background is healthcare, so am not very good with statistics.
My question specifically is that we have 2 independent variables -
We need to measure the relationship between predictor (e.g. pain) with adherence as well as attendance. What other factors need to be considered?
Any guidance would be useful.
Seems like attendance and adherence are both continuous variables. Most people would put them into the r metric, as you intuited. What happens in the studies? Do researchers commonly report the associations in terms of r or standardized beta weights? Or do they report odds ratios? If the latter, then going to OR is probably conventional. (Though a purist would argue it should still be r.)
Hope that helps!
Thanks for your quick reply Blair.
The aim is to study the relationship between 1) attendance and predictors (say for e.g. pain, self efficacy etc.) and 2) adherence and predictors (which can be same or different as attendance) and 3) if possible, study relationship between attendance and adherence.
So we have 2 dependent variables attendance and adherence, while there will probably be 10-15 independent variables (say for e.g. pain, self efficacy etc.)
I am going to say what I am proposing and then you can correct me if I am wrong. This is copied from my protocol which I am preparing -
The data extracted from included studies will be presented in a table highlighting comparison and critique of the studies. Narrative summary of the results of each study will be provided. Thus a matrix of themes which can act as barriers or facilitators of attendance and adherence will be created.
In addition, for each independent variable, “effect size r” (Correlation between independent variable and adherence) will be calculated. If the effect size r (in the form of Pearson, point-biserial, or Phi coefficient) is not presented in the study, data will be extracted to calculate an r from F (1 degree of freedom in the numerator), t, chi square, means and standard deviations, tables of counts, or exact p values (and using the Z associated with the exact p, divided by the square root of n, to equal Phi). If results are described in text simply as “nonsignificant”, then r value will be designated as 0. Effect size r represents both the strength and direction of association between independent variable and adherence (ranging from -1 to +1) (Durlak 2009).
But as I understand, we have to apply the random effects model for pooling of individual results, though am not sure how to do that at this stage?
I have looked at a few studies and most of them report some form of 'r', though am not sure if multiple regression is used when other independent variables are used, how it would affect 'r'?
Look forwards to your opinion.
You'll want to use the Fisher r-to-z conversion for analyses (and convert z-to-r for interpretation). Sources like Lipsey & Wilson (2001, Practical Meta-Analysis) will lead you through the problems. They also offer macros for meta-analysis for use with SPSS, SAS, and Stata.The macros are free (but of course getting access to any of those statistics programs may have considerable cost). There's a no-cost solution in Viechtbauer's program suite metafor, written for R. Check out our discussions of these issues on this site.
Thanks Blair. I do have access to SPSS through my University, so can you tell me where I can download macros for this calculation. I think I could do without the steep learning curve of R. You have been truly helpful in guiding me through this maze and I guess as I proceed, I will have a lot more questions coming your way.