Experimental disclosure and its moderators: a meta-analysis
The definitive meta-analysis of writing about difficult experiences. Pooling 146 randomised studies with a random-effects model, expressive writing has a positive and statistically significant effect - and an average effect size of r = .075, considerably smaller than earlier syntheses reported.
If you read one paper on whether journaling works, read this one - and be ready for the number to be small.
By 2006 two meta-analyses already suggested expressive writing helped: Smyth 1998 and Frisina et al. 2004. Both used fixed-effects models, which assume every study is estimating one identical underlying effect and consequently limit how far conclusions generalise. A great many more studies had also been published since.
Frattaroli collected 146 randomised studies and reanalysed the question with random-effects models.
The verdict: experimental disclosure is effective, with a positive and significant average effect size of r = .075.
That is a small number, and it is the most trustworthy one in this literature. For scale, Frisina et al. reported d = 0.19 and Smyth reported larger effects still. When a bigger, better-specified analysis lands lower than its predecessors, the lower figure is usually closer to the truth.
The paper also identifies a set of moderators - the conditions under which writing works better or worse. That is the practically useful half: the question is not really does journaling work but for whom, about what, and how often.
How to hold this honestly. A small average effect across 146 studies is entirely compatible with journaling being genuinely worthwhile - it is nearly free, has few side effects, and the moderators suggest some people benefit considerably more than the average. It is not compatible with journaling being transformative for everyone.
r = .075 is very small - smaller than the earlier Smyth 1998 and Frisina 2004 estimates, which used fixed-effects models that limited how far their results generalised. This is the correction, and it corrects downward. It does not identify a single reliable protocol; instead it finds that a number of moderators change the result, which is another way of saying the effect depends heavily on circumstances.
Articles citing this study
Summarized by Tiago Martins · Last updated: · How we write these
