The Role of Neighborhood Factors and Community Stigma in Predicting Community Participation Among Persons With Psychiatric Disabilities

Institutional review board approval was received from all participating institutions. We recruited community member participants between March 2014 and December 2015. Approximately 200 community members were surveyed in each area, with a minimum of 50 respondents per zip code (with the exception of one area where safety concerns limited street outreach) by trained research assistants, who distributed flyers in public locations. Eligibility criteria included being 18 or older, speaking English well enough to complete interviews, and living or working within the targeted area. After providing informed consent, participants completed questionnaires, either administered orally by the research assistant or completed independently—depending on preference. Interviews lasted approximately five to ten minutes, and compensation was $10.

Within each site, 50–60 participants were recruited from two types of housing treatment facilities: independent scattered-site and congregate housing. Participants were recruited from four public-sector clinics and eight housing agencies between March 2014 and December 2015 via announcements at common areas at the site (for example, housing programs’ community meetings) and flyers distributed to potentially eligible participants by housing agencies. When potential participants were not recruited at their residence, they were asked the name and address of their housing agency; its location within a targeted zip code was confirmed with a map and its category confirmed by using New York City’s housing vacancy update (18). Interviews were scheduled after eligibility was determined and informed consent was obtained. Of the persons approached who met inclusion criteria, 13% refused to participate. In addition, 6% of persons who were likely aware of the study at the recruitment sites did not express interest in participating. We estimated that roughly 80% of eligible individuals participated. Interviews lasted from one to two hours, and compensation was $30.

Demographic data were collected with a questionnaire including age, gender, race-ethnicity, education, zip code of residence, and residence or employment within the targeted area.

Stigmatizing attitudes were measured with the Attitudes Toward Mental Illness Scale (AMIS) (19), an 11-item measure scored on a 1- to 5-point scale, with higher overall scores indicating more stigma. AMIS consists of two factors, negative stereotypes (Cronbach’s α=.66) and recovery and outcomes (Cronbach’s α=.69).

Stigmatizing behaviors were measured with the Reported and Intended Behavior Scale (RIBS) (20) and Mental Illness Microaggressions Scale–Perpetrator Version (MIMS-P) (11). The RIBS includes eight items of intended interaction with individuals with mental illness, with four items regarding intended behavior (for example, “In the future, I would be willing to live with someone with a mental health problem”). Internal consistency in our sample was adequate (Cronbach’s α=.74). The MIMS-P includes 17 items measuring microaggression experiences perpetrated toward individuals with mental illness (a 14-item version of the scale was included in this study). Each item is rated on a scale from 1 to 4, with higher scores indicating more microaggressions. Internal consistency was high in the sample (Cronbach’s α=.85).

Demographic data included age, race-ethnicity, education, length of time in the current residence, age at first hospitalization, and number of previous hospitalizations. Psychiatric diagnosis was assessed, with consent, via a review of the clinical face sheet from participant’s charts.

Psychiatric symptoms were measured with the Brief Psychiatric Rating Scale (21), a rating scale for 24 different psychiatric symptoms. We used a subscale-scoring approach (22), dividing items into five factors: affect, positive symptoms, negative symptoms, resistance, and activation.

Physical community integration was measured with a modified 19-item version of the External Community Integration Scale (ECI) (23), assessing number of days in the past two weeks when activities were completed outside the household (for example, going to a movie and attending religious services). Internal consistency was acceptable in our sample (Cronbach’s α=.57). Social community integration was measured with a 12-item scale (24) assessing frequency of interactions between the participant and community members (for example, borrowing items and social outings). Two versions were administered: one related to interactions with neighbors (Cronbach’s α=.85) and the other to interactions with family and friends (Cronbach’s α=.88). Civic engagement was measured with the Social Capital Short Form (25), assessing involvement in activities indicative of citizenship (for example, community service and voting). Internal consistency was poor (Cronbach’s α=.45), with low item endorsement; thus we dichotomized this variable (none versus any involvement).

Social functioning was assessed with the Quality of Life Scale (QLS) (26), a 21-item rating scale completed after a semistructured interview. For this study, we were interested in three of the four QLS factor scores most related to social functioning: interpersonal relations measures the frequency of recent social contacts, instrumental functioning measures vocational functioning, and common objects and activities measures participation in common societal activities (for example, reading a newspaper).

When a reported response reflected the impossible or improbable (for example, length of time in the community that exceeded the participant’s age), responses were considered missing. Mean imputation was used to include scales in cases when specific item values were missing. Scales were not included in cases when more than 50% of items were missing.

Items were developed based on five identified categories from a focus group study of microaggression experiences: invalidation, assumption of inferiority, fear of mental illness, shaming of mental illness, and second-class citizen (27). Items were created on the basis of participant responses related to interactions with community members (for example, “When I am in public, other community members avoid being physically close to me because of my mental illness”) and scored on a 1–4 scale, with higher scores indicating more microaggression experiences. The scale demonstrated good internal consistency (Cronbach’s α=.83).

We first conducted an exploratory factor analysis to reduce the large number of community participation variables. We included all community participation variables except for citizenship, which was excluded because of low internal consistency. The Kaiser-Meyer-Olkin measure of sampling adequacy was .71, above the recommended value of .5 (28), indicating that the sample size was appropriate. The scree plot indicated a three-factor solution, and principal-components analysis with direct oblimin rotation was employed. Using a cutoff factor loading of .4, factor 1 (social community participation) included QLS interpersonal relationships, SCI with neighbors, and SCI with friends and family; factor 2 (vocational involvement) consisted of the QLS instrumental functioning subscale only; and factor 3 (physical community participation) included the QLS common objects and activities factor and the ECI. We combined scales that loaded on the same factors by converting them to standard scores and summing them.

Neighborhood-level data were gathered linking participants to characteristics drawn from the 2010 U.S. Census. For a large number of participants (14% of the community sample), we did not have enough data to identify census tract because of inaccuracy of reported cross streets or unwillingness to disclose addresses. Because zip code–level data were available for all participants, all neighborhood data were gathered at the zip code level. Neighborhood characteristics were first analyzed with exploratory factor analysis to create composite variables. Variables were divided into two sets for factor analysis to obtain adequate indicators of sampling adequacy.

Set 1 included unemployment rate, median income, percentage of families below the federal poverty level, percentage of high school graduates, percentage receiving public assistance, and percentage of female-headed households for each zip code. These characteristics were included together because they have typically been used in previous neighborhood disadvantage research (29,30). The Kaiser-Meyer-Olkin measure of sampling adequacy was adequate (28). Because of low communality, unemployment rate was removed. Communalities for all remaining items were above .6, indicating that individual items shared common variance with other items. Principal-components analysis with direct oblimin rotation using Kaiser normalization was employed with a two-factor solution (following examination of the scree plot), and the resulting factors were named socioeconomic disadvantage and vocational disadvantage.

Set 2 included the percentage of vacant housing units, residential stability (percentage of residents moving within the past five years), percentage of foreign-born individuals, GINI index, population density by square mile, and conservative voting habits. The GINI index is a measure of unequal income distribution, ranging from 0 to 1, with higher values indicating greater inequality. Vacant housing, residential stability, and percentage of foreign-born individuals have been used as indicators of neighborhood disadvantage in previous studies (29,31). Population density, conservative voting, and the GINI index were chosen on the basis of prior work indicating that political affiliation and “family orientation” (high proportion of single-family units) are predictors of interest (6). All variables were drawn from the 2010 census at the zip code level, except for conservative voting, which was determined at the congressional district level with the percentage of individuals who voted for the 2012 Republican presidential nominee. Because of low communality, residential stability was removed from the analysis. A two-factor solution was determined, and the factors were named suburban values and income inequality.

Hierarchical linear modeling (HLM) was first employed because of expected variation at the neighborhood level. Unconditional models were estimated to examine intraclass correlation coefficients (ICCs); ICCs provide the proportion of variance at level 2 (neighborhood) to the total variance of the model and are used to assess whether HLM is necessary. Across hypotheses, ICCs ranged from .01 to .1; this determined that there was not enough variance at the neighborhood level to continue with HLM, and the model was reduced to a linear regression analysis. Summary scores on the AMIS, RIBS, and MIMS-P at the zip code level are provided in Table 2.

Table 3 presents results of a regression analysis examining the relationship between neighborhood factors and endorsed stigma by community members. Partial support was found for the hypothesis that neighborhood disadvantage would be associated with more community stigma. Socioeconomic disadvantage was associated with more microaggressions endorsed by community members (MIMS-P total). Community members in areas with more “suburban values” (less dense and more politically conservative) also reported fewer recovery-oriented attitudes (AMIS recovery) and more microaggressions.

Conversely, vocational disadvantage was associated with more positive community member attitudes toward mental illness (AMIS recovery), and residential instability was also associated with more positive attitudes and fewer microaggressions endorsed by community members. A significant negative relationship was also found between income inequality and scores on the AMIS, such that individuals living in areas with greater inequality reported fewer negative stereotypes. However, community members in these areas also endorsed fewer recovery-oriented attitudes toward mental illness.

No significant relationship was noted between neighborhood variables and microaggressions perceived by participants with psychiatric disabilities. Exploratory analyses indicated, however, that, despite their being located in the same communities, participants in congregate housing perceived significantly more microaggressions than participants in independent scattered-site housing (β=.19, p≤.01). We thus explored whether there was an interaction between housing and neighborhood factors in predicting perceived stigma with product terms for housing category and each neighborhood predictor. Significant interactions were noted between housing and vocational disadvantage (β=–.36, p<.05) and between housing and suburban values (β=.31, p<.05). Participants in independent scattered-site housing perceived more microaggressions in communities with more vocational disadvantage, while participants in congregate housing perceived more microaggressions in neighborhoods with more suburban values.

Table 4 provides results from multiple regression analyses. Contrary to expectations, community member microaggressions had a significant positive relationship to vocational involvement when the analysis controlled for negative symptoms, such that individuals with psychiatric disabilities living in areas with more community member–reported microaggressions demonstrated higher levels of vocational involvement. However, when housing type was included in the analyses, a significant interaction between neighborhood microaggressions and housing type was found for physical community participation and vocational involvement. For individuals in congregate housing, living in areas with higher levels of community microaggressions was positively related to vocational outcomes and participation in routine activities in the community, whereas the reverse was found for participants living in independent scattered-site housing in these communities. Figure 1 illustrates the nature of the interaction for physical community participation.

In multiple regression analysis including perceived microaggressions, housing, and negative symptoms, the overall model significantly predicted community participation (social community participation, F=6.68, df=3 and 324, p<.01, r²=.06; physical community participation, F=14.45, df=3 and 324, p<.01, r²=.06; physical community participation, F=14.45, df=3 and 324, p<.01, r²=.12; and vocational involvement, F=3.99, df=3 and 324, p<.01, r²=.03). Individually, perceived microaggressions had a statistically significant relationship with social community participation (β=–.13, p<.05) and with physical community participation (β=–.15, p<.01) but not with vocational involvement, when the analysis controlled for negative symptoms. Individuals who perceived more community microaggressions demonstrated significantly less social and physical community participation. When housing type (scattered site or congregate) was added to the model, it was significantly associated only with physical community participation, such that individuals living in scattered-site housing demonstrated higher levels of physical community participation (β=–.17, p<.01).