Link to full text was posted on Twitter: https://dl.dropboxusercontent.com/u/15691907/Tajika et al. (2015).pdf
Replication and contradiction of highly cited research papers in psychiatry: 10-year follow-up
Aran Tajika, Yusuke Ogawa, Nozomi Takeshima, Yu Hayasaka, Toshi A. Furukawa
The British Journal of Psychiatry Jul 2015, DOI: 10.1192/bjp.bp.113.143701
Contradictions and initial overestimates are not unusual among highly cited studies.
However, this issue has not been researched in psychiatry.
To assess how highly cited studies in psychiatry are replicated by subsequent studies.
We selected highly cited studies claiming effective psychiatric treatments in the years 2000 through 2002.
For each of these studies we searched for subsequent studies with a better-controlled design, or with a similar design but a larger sample.
Among 83 articles recommending effective interventions, 40 had not been subject to any attempt at replication, 16 were contradicted, 11 were found to have substantially smaller effects and only 16 were replicated.
The standardised mean differences of the initial studies were overestimated by 132%.
Studies with a total sample size of 100 or more tended to produce replicable results.
Caution is needed when a study with a small sample size reports a large effect.