A random effects model was appropriately chosen (between study variation is allowed) as the analysis showed evidence of heterogeneity across studies that was of borderline significance (p = 0.049).
The majority of between-study variation was attributed to studies that defined cases on serology.
Between-study variation was also lower within subgroups of tumor grade and tumor size.
The overall estimate for this effect was 0.77 (95% CI, 0.66 0.88) and significant between-study variation was found.
Furthermore, the between-study variation was markedly lower when taking tumor size into account, and somewhat lower within subgroups of grade.
A mixed logit model with a random intercept (i.e., to account for between-study variation) was applied to assess hemoglobin effects on pregnancy outcome.
Because of the presence of heterogeneity across studies, a random-effects model, which considered both within- and between-study variation, was used to compute the summary risk estimate.
For GLUT1, the overall between-study variation was large as well, but substantially smaller for studies investigating membranous staining only and the best evidence studies (these study groups again largely overlapped).
If heterogeneity was present, between studies variation was then estimated as follows: (tau^{2} = frac{Q - (k - 1)}{{sum w_{i} - frac{{sum w_{1}^{2} }}{{sum w_{1} }}}}) if Q k 1 or 0 otherwise.
Reasons based upon heterogeneity (or between-study variation) were the most commonly cited (46/69; 66.7%).
The error term in a fixed effects model represents only within-study variation, and between-study variation is ignored.
