The use of evidence-based practice (EBP) has become a well-established goal in social services, health care, and mental health services. Varied research designs (ranging from qualitative descriptions to various kinds of quasi-experimental research) can provide initial support for a promising practice. However, services are usually designated as evidence-based only if rigorous, randomized controlled trials (RCTs) demonstrate that the people they serve have better outcomes (for example, longer periods in stable housing) than similar people who did not receive such services. Because RCTs randomly assign participants to either experimental or alternative services and compare their outcomes, they can provide strong evidence that a practice actually causes good outcomes. The highest form of evidence, meta-analysis, combines data from several RCTs of the same intervention (Institute of Medicine, 2001; Sackett et al., 2000).
Although we highlight instances of higher level evidence, such as RCTs, a growing literature contests uncritical acceptance of EBP criteria. Critics note that RCTs test interventions in populations and conditions rarely matched in the real world (as when studies of services for people who have experienced homelessness and psychiatric disabilities exclude those with co-occurring disorders); understudied approaches may not attract funding and thus have little chance to develop an evidence base; and EBPs often use a limited notion of what works (Sanderson, 2004; Tanenbaum, 2005). These critiques resonate with advocates of recovery-oriented services who worry that peer services may be undermined by random assignment and that experimental methods and existing measures fail to capture meaningful but hard to measure processes and goals (Anthony, Rogers, & Farkas, 2003; Clay, 2006; Jewell, Davidson, & Rowe, 2006).
As the uses of EBP expand from aiding clinical decision-making to guiding policy, further objections have been registered by those who understand policy-making as a process that incorporates considerations of power and values (e.g., the power to set research agendas; assumptions about valued outcomes) in addition to evidence-based understandings of what works (Nixon, Walker, & Baron, 2002). These concerns have particular relevance to consumer involvement in research, policy, and planning, where available metrics for assessing outcomes may not take appropriate measure of the goal of changing the balance of power between consumers and those who shape the policies that affect them.
 Advocacy groups described here are documented in media coverage or in Web sites or reports. We suspect much consumer activism goes unrecorded.
 Treatment services for alcohol and drug addiction have for decades relied on a large mutual support component represented in 12-step groups such as Alcoholics Anonymous and Narcotics Anonymous that often serve as an alternative, sometimes as an adjunct, to professionally delivered treatment (White 1998). In many programs, peer support is an integral part of addiction treatment for people who have experienced homelessness (Galanter 2000).
 RCTs are needed to make a more definitive case for effectiveness by ruling out the possibility that unmeasured characteristics related to case manager assignment might account for the equivalent outcomes.
 Studies of self-help groups show those who attend have symptom outcomes similar to or better than non-attendees (Kaufman, Schulberg, & Schooler, 1994; Powell et al., 2001; Moos et al., 2001). Greater commitment to the group is associated with symptom improvements (Galanter, 1988; Raiff, 1984). Participants have the same or lower rates of hospitalization (Kurtz, 1988; Galanter, 1988; Kennedy, 1989; Rappoport, 1993; Edmondson, Bedell, & Gordon, 1984; Trainor et al., 1997; Trainor & Tremblay, 1992), and shorter duration of stay (Solomon & Draine, 2001). They are more socially integrated as indexed by larger social networks (Rappaport et al., 1985; Roberts & Rappaport 1989), more pursuit of work and education (Kaufman, 1995), more involvement in formal social roles (Zimmerman et al., 1991); and they have better social functioning (Carpinello, Knight, & Janis, 1991; Galanter, 1998; Kaufman, 1995; Markowitz, DeMasi, & Carpinello, 1996).
 Initial analyses showed improvement in well-being over time for both experimental and control groups, with significantly greater improvement in a subset of COSP sites, but failed to confirm the overall hypothesis. However, these analyses also showed that some participants crossed over from COSP to usual services during the course of the study. Because conventional intent-to-treat analysis, which makes comparisons based on original group assignment, regardless of whether those randomly assigned to experimental and control groups actually participated in their assigned programs, may understate effects of program involvement, subsequent analyses employed an as treated approach. These analyses compared groups that actually participated in experimental or control services. Propensity scoring was used to divide the sample into homogeneous strata and compare outcomes for experimental and control participants within strata to avoid mistaking outcome differences that may occur because experimental and control programs attract different kinds of participants for those that result from program differences.
 Most consumers remained stably housed throughout the study, limiting the likelihood of showing significant effects of COSP on homelessness (Jean Campbell, personal communication, 8/8/06; Gregory Teague, personal communication, 8/8/06). Thus whether COSPs effectively address homelessness may remain an open question.
 We use the term integrated housing development, to refer to a type of supportive housing that accommodates diverse housing-needy individuals among whom those with mental illness are not the majority in single apartment or SRO units. Services are available on site but from a provider separate from the housing manager, and tenure is not contingent on sobriety or service involvement (Hopper & Barrow 2003).
 After one year, participants in the Housing First Program spent 85 percent of their time stably housed, compared with less than 25 percent for participants in the services-as-usual group (Tsemberis, Gulcur, & Nakae, 2004) and the effect endured: After two years, housing first participants still spent approximately 80 percent of their time stably housed, compared with only 30 percent for the control group. Rates of homelessness decreased dramatically for Housing First tenants, who had spent approximately 55 percent of the period before baseline literally homeless, dropping to 12 percent at one year, and less than 5 percent after two years. Reductions in homelessness were significantly slower and less dramatic for the control group, who were homeless about 50 percent of the time at baseline, 27 percent at one year, and 25 percent after two years.
 Among its early applications, needle exchange programs designed to mitigate needle-related infectious disease transmission were prominent (Langendam et al., 2001). More recent applications include jail diversion programs to reduce the psychological, economic, and socially hazardous conditions incurred during incarceration (Klein, 1997).
 The stages of change are precontemplation, contemplation, preparation, action, and maintenance.