Advanced search
1 file | 3.97 MB

Fixed versus random effects models for fMRI meta-analysis.

Han Bossier (UGent)
(2015)
Author
Organization
Abstract
Meta-analyses for brain imaging are gaining attention given the increasing amount of fMRI studies and the need for synthesis and integration of data across studies. Standard meta-analyses are not well adapted to summarize peaks of brain activation. In this paper we look at the performance of current meta-analysis methods and investigate the effect of pooling subjects at the study level on the outcome. We do this by combining a fixed, ordinary least squares versus mixed effects pooling method with (1) a vote-counting procedure, (2) a fixed- and (3) a random effects meta-analysis. The fMRI data consists of 300 subjects. We split the group in 2, using a group analysis of 150 subjects as a benchmark. The other group of 150 subjects is divided into 10 smaller studies. Each result of a meta-analysis is overlaid on the group-analysis to calculate the false positive rate, power and overlap (i.e. spatial accuracy). Results show the most beneficial effects when pooling subjects through a mixed effects analysis regardless of the meta-analysis. Unfortunately, the highest observed power of any meta-analysis does not exceed 50%, which indicates challenges in reproducing fMRI results.

Downloads

  • Presentation HBossier.pdf
    • full text
    • |
    • open access
    • |
    • PDF
    • |
    • 3.97 MB

Citation

Please use this url to cite or link to this publication:

Chicago
Bossier, Han. 2015. “Fixed Versus Random Effects Models for fMRI Meta-analysis.” In .
APA
Bossier, H. (2015). Fixed versus random effects models for fMRI meta-analysis. Presented at the Joint Statistical Meetings.
Vancouver
1.
Bossier H. Fixed versus random effects models for fMRI meta-analysis. 2015.
MLA
Bossier, Han. “Fixed Versus Random Effects Models for fMRI Meta-analysis.” 2015. Print.
@inproceedings{8552047,
  abstract     = {Meta-analyses for brain imaging are gaining attention given the increasing amount of fMRI studies and the need for synthesis and integration of data across studies. Standard meta-analyses are not well adapted to summarize peaks of brain activation. In this paper we look at the performance of current meta-analysis methods and investigate the effect of pooling subjects at the study level on the outcome. We do this by combining a fixed, ordinary least squares versus mixed effects pooling method with (1) a vote-counting procedure, (2) a fixed- and (3) a random effects meta-analysis. The fMRI data consists of 300 subjects. We split the group in 2, using a group analysis of 150 subjects as a benchmark. The other group of 150 subjects is divided into 10 smaller studies. Each result of a meta-analysis is overlaid on the group-analysis to calculate the false positive rate, power and overlap (i.e. spatial accuracy). Results show the most beneficial effects when pooling subjects through a mixed effects analysis regardless of the meta-analysis. Unfortunately, the highest observed power of any meta-analysis does not exceed 50\%, which indicates challenges in reproducing fMRI results.},
  author       = {Bossier, Han},
  language     = {eng},
  location     = {Seattle},
  title        = {Fixed versus random effects models for fMRI meta-analysis.},
  url          = {http://www.amstat.org/meetings/jsm/2015/},
  year         = {2015},
}