Studying the link between mental illness and violence seems a sure way to
- get attention from the mass media, and
- contradict some other study, and not necessarily one published that long ago.
A pretty old review (and not in some high impact venue) by Stuart (2003) seems to me to still hold water 15 years later with respect to a few general observations:
Definitive statements are difficult to make and it is equally possible to find recent literature supporting the conclusions that the mentally ill are no more violent, they are as violent, or they are more violent than their nonmentally ill counterparts. [...]
Because of the significant methodological challenges faced by researchers in this field, the nature of this association remains unclear. For example, violence has been difficult to measure directly, so that researchers have often relied on official documentation or uncorroborated selfreports. The prevalence of violence has been demonstrated to differ dramatically depending on the source. [...]
And it goes on to talk about confounders as the perennial problem of epidemiological studies.
Stuart also highlights a (highly cited) US-sample study of the "no-link" variety (at least if substance-abuse is controlled for), Steadman et al. 1998 ... and I certainly found some newer ones myself, Elbogen & Johnson (2009) or Anderson et al. 2012, just to mention some picked up by the press.
On the other hand, the Oxford group of Fazel has been publishing study after study on European samples including on Sweden (2015) or UK, Finland and Netherlands (2017) that do find a link between depression and violence... and these have also been picked up the press. There are older studies in Scandinavian countries which concur; Hodgins et al. (1996) for instance.
So is this a trend? And (the really tough question) is there a meta-analysis of such studies that does moderator analysis (e.g. by meta-regression) to determine which study parameters matter most with respect to the finding? Is it the case that the Europeans are less violent overall so that the "inherent" violence of the mentally ill "stands out" in Europe but "gets lost in the noise" of the higher society violence in the US? (It certainly is the case the US is more violent than Sweden.) Or is it some other factor that is dominant, e.g. is depression more conductive to violence than other forms of severe mental illness (makes sense to ask since 2 of 3 US studies I've looked at aggregate severe mental illnesses — SMIs). To give you an example of the kind of meta-analysis I'm hoping to find, for ADHD prevalence it seemed that there might be geographical differences between US and Europe, but they turned out to be methodological.
I found a paper by Van Dorn et al. (2012) reanalyzing the same data as Elbogen & Johnson, and coming to a European-style conclusion of a link between SMI and violence. The difference in methodology came down (in this case) to:
important is spatio-temporal contiguity (i.e., the causal factor must be spatially near the effect).
So merely lifetime SMI prevalence wasn't a good predictor but recent symptoms (within 12 months) were. That and the fact that the control group of Elbogen & Johnson apparently was not clean enough (of substance abusers) apparently tipped the scales.
But Eblogen didn't quote concede the field, and in a 2016 study on the exact same data (using 12 months window this time) but with more attention to confounders... found again in favor of their original findings:
Mediation analyses informed by the I3 model revealed that the positive association between SMI and violence was by and large indirect. Statistically controlling for impellance and disinhibition completely wiped out any positive relationship between SMI and violence.
This is somewhat similar result to what Anderson et al. (2012) obtained on different data after controlling for impulsivity and "present orientation" (using just two simple questions for these two factors!)