Are people with mental illness more violent or aggressive, or more likely to commit crimes than those without mental illness? Two recent papers investigate different aspects of this question.
In the January 2011 issue of Psychiatric Services, Fisher and colleagues investigate data from the Massachusetts Department of Mental Health, and found that people who had been diagnosed with a “severe and persistent psychiatric disorder” were two-thirds more likely than the general population to be arrested within a one-year period. Arrest rates were significantly higher for all crimes, but particularly high for assault and battery on a police officer, a felony (odds ratio 5.96, or about 6 times more likely), and “crimes against public decency” (odds ratio 4.72). While the data only reflect arrests (and not convictions, which would be fewer, since some charges were undoubtedly dropped), and say nothing about whether a person was actively involved in treatment at the time of his or her arrest, it does portray the severely mentally ill as more likely to engage with the criminal justice system.
A separate study, published last month in PLOS One, examined reports of medication-related adverse events to the FDA. The authors looked at 484 drugs and the rates with which they had been associated with “violence-related symptoms.” All medications had some such symptoms reported, but certain classes of drugs were associated with more frequent violent events than would occur by chance alone. In particular, varenicline (Chantix) was most frequently associated with reports of violence, with a PRR (proportional reporting ratio) of 18.0. (This means that the proportion of violent events by patients on Chantix was 18 times greater than the proportion of violent events reported for all other drugs). Other medications shown to be associated with violent behaivor included antidepressants (average PRR = 8.4) and psychostimulants (average PRR = 6.9).
It should be pointed out that the authors of the PLOS study have served as expert witnesses in criminal cases involving psychiatric drugs, and work for the Institute for Safe Medication Practices, a nonprofit group dedicated to exposing medication risks. Furthermore, the reports of adverse events to the FDA may suffer from “attribution error”: when an adverse event like a violent act occurs, we search for possible causes, and medications– particulalry psychiatric meds– are frequent culprits, when in fact there may be no causal relationship. Nevertheless, the large numbers of events, and the relative consistency within drug classes, should give us reason for concern.
But even with these limitations, these studies unfortunately show that criminal activity may indeed be more common among the mentally ill, and we need to exercise caution when using medications when the risk of aggression is high, to avoid making matters worse.
Thought Broadcast 
I find the mention of the author's expert testimony as implying some kind of conflict of interest interesting, since at trial opposing attorneys would undoubtedly attempt to portray their participation in this study as evidence that they are biased witnesses, not to be believed. The key here, though, is like you say, there may be no causal relationship. Indeed only an association has been established. No mechanism of causality has been revealed. Such violent behaviors are complex and generally require intent. These are not automatons acting robot-like against their own will. Do you ever wonder why we don't hear of patients taking moxapoxabazepine or whatever suddenly dressing themselves in purple and waving Liberian flags while standing on their heads. It generally involves behaviors where there is criminal potential or possibility of financial compensation, and more work (and money) for lawyers, not to mention expert witnesses.