Hi All, I would like to get some advice on including the following article in meta-analysis. I have requested data from authors but have not been successful. The article is open access so hopefully it is accessible. https://www-ncbi-nlm-nih-gov/pmc/articles/PMC4921225/ The report the intercepts from Hierarchical Linear Modeling Analyse in table 3 for one of the outcomes of importance. Would these values be able to be used as the mean in analysis, with the SE converted to SD? An example of their reported data is below. Post-Treatment Fixed Effect Beta Coefficient SE t-ratio df p-value Level 1 Intercept CS Rot 15.092 1.586 9.518 90 <0.001 CS ExtRot 14.301 2.030 7.045 90 <0.001 NCs Rot 17.219 1.421 12.122 90 <0.001 NCs ExtRot 13.695 1.813 7.553 90 <0.001 Level 1 Linear d CS Rot ?0.609 0.116 ?5.268 90 <0.001 CS ExtRot ?0.625 0.143 ?4.365 90 <0.001 NCs Rot ?0.702 0.141 ?4.968 90 <0.001 NCs ExtRot ?0.878 0.120 ?7.318 90 <0.001 Level 1 Quadratic d CS Rot 0.014 0.003 4.256 688 <0.001 CS ExtRot 0.017 0.004 4.875 688 <0.001 NCs Rot 0.014 0.004 3.441 688 <0.001 NCs ExtRot 0.025 0.004 5.936 688 <0.001 Thanks, Scott.
[R-meta] Beta Coefficients (Predicted Means?) For Inclusion in Meta-Analysis
2 messages · Scott Tagliaferri, Röver, Christian
Dear Scott, from the article it sounds like the quoted "intercept" parameters would in fact correspond to the estimated mean outcome (disability score): "The beta coefficients for the intercept are the expected scores for the outcome measure at the data collection point." The corresponding "data collection point" then is shown in each table, so that the numbers you quote should then correspond to the mean score at post-treatment time point. Along with each coefficient estimate, they quote a standard error (SE), which might be the number you'd actually need for a meta analysis (?). In case you in fact require standard deviations instead (e.g. for subsequent calculations), deriving these from the quoted standard errors would probably be tricky, since they have fitted a somewhat complex hierarchical model here. Not sure whether one could possibly reverse-engineer at least an estimate of these from Table 6 (Are the numbers in brackets supposed to be standard deviations? Maybe one could derive a pooled standard deviation from these.). Cheers, Christian
On Sat, 2021-01-30 at 08:00 +1100, Scott Tagliaferri wrote:
Hi All, I would like to get some advice on including the following article in meta-analysis. I have requested data from authors but have not been successful. The article is open access so hopefully it is accessible. https://www-ncbi-nlm-nih-gov/pmc/articles/PMC4921225/ The report the intercepts from Hierarchical Linear Modeling Analyse in table 3 for one of the outcomes of importance. Would these values be able to be used as the mean in analysis, with the SE converted to SD? An example of their reported data is below. Post-Treatment Fixed Effect Beta Coefficient SE t-ratio df p-value Level 1 Intercept CS Rot 15.092 1.586 9.518 90 <0.001 CS ExtRot 14.301 2.030 7.045 90 <0.001 NCs Rot 17.219 1.421 12.122 90 <0.001 NCs ExtRot 13.695 1.813 7.553 90 <0.001 Level 1 Linear d CS Rot ?0.609 0.116 ?5.268 90 <0.001 CS ExtRot ?0.625 0.143 ?4.365 90 <0.001 NCs Rot ?0.702 0.141 ?4.968 90 <0.001 NCs ExtRot ?0.878 0.120 ?7.318 90 <0.001 Level 1 Quadratic d CS Rot 0.014 0.003 4.256 688 <0.001 CS ExtRot 0.017 0.004 4.875 688 <0.001 NCs Rot 0.014 0.004 3.441 688 <0.001 NCs ExtRot 0.025 0.004 5.936 688 <0.001 Thanks, Scott.
_______________________________________________ R-sig-meta-analysis mailing list R-sig-meta-analysis at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis