Law enforcement response to mental health crises: citizen risk factors and preferences for social policy^*

* An earlier version of this paper was presented at the 2014 American Society of Criminology annual meetings. The authors thank Marlon Marion, Ryan Stafford, Robin Clough, James Foutch, and Dylan Waite for research assistance.

Abstract

Despite having the potential to affect cooperation and support for law enforcement, community preferences for police training are generally unstudied. This paper seeks to understand (1) general community preferences for police training for responding to mental health crises, (2) where the public prioritizes this mental health crisis training relative to racial bias training, and (3) whether these preferences vary depending on respondent demographics and risk factors for police contact. Survey data from a general sample of Portland, Oregon residents (N = 259) shows public support for police training on responding to mental health crises, but the public as a whole is split over whether mental health, or racial bias training, should be prioritized as the best use of police training time. Logistic regression analyses demonstrate three main findings regarding the impact of respondent characteristics on preferences for police training: (1) black respondents appear concerned that, by drawing attention to mental health crises, police will be more limited in their responses to racial bias; (2) respondents facing the greatest, and those with the least, amount of risk in police/citizen encounters are most supportive of mental health crisis training; and (3) mental illness itself does not appear to affect support for police mental health crisis training. These differences in support for training need to be understood before law enforcement can cultivate community support for police officers and their training practices.

Keywords:

• Mental illness
• race
• police training policies

Notes

* An earlier version of this paper was presented at the 2014 American Society of Criminology annual meetings. The authors thank Marlon Marion, Ryan Stafford, Robin Clough, James Foutch, and Dylan Waite for research assistance.

1. The fliers, and any online blurbs about the survey, indicated that respondents who took this online survey would be entered into a drawing for a $50 gift card as an incentive for their participation.

2. Analyses of the two groups (those taking the survey online vs. those recruited face-to-face for a paper version of the survey) showed no significant differences in support for police training, which suggests that using the different recruitment strategies did not bias results.

3. For the multivariate analyses, ‘other/mixed’ race respondents who identified as black plus some other category were combined with those who solely self-reported African American. Supplemental analyses were conducted before making this change, however. These analyses were conducted to ensure that the black/other race respondents were similar to the solely black/African American respondents. Results suggest that the two groups were quite similar in their characteristics, mental health history, and support for police training. The conclusions in this paper would not differ if these two groups were kept separate. This suggests that, by combining these two groups, the results and conclusions are not being altered.

4. Responses to the first two questions (‘mental illness training will improve police interactions with individuals with mental illness’ and ‘mental illness training will improve police relations with all’) are negatively skewed, and responses to the third question (‘increasing mental illness training for the police will negatively impact their interactions with racial minorities’) are positively skewed. The final dependent variable (mental illness training should be prioritized over racial bias training) is not skewed, but is leptokurtic, with a large concentration of respondents choosing the neutral response (neither agree nor disagree). Due to non-normal distributions, and because the research is primarily interested in understanding the predictors of ‘support’ vs. ‘opposition’ to police training, the decision was made to convert the dependent variables into dichotomous measures and to conduct logistic regression, although supplemental analyses retained the original metric and ran OLS regression. Any differences between the two analyses are reported in endnotes.

5. Because these four statements represent distinct concepts/ideas regarding police training, they were kept separate rather than combined into one index for analysis.

6. Because the sample was designed to recruit a sufficiently large number of racial/ethnic minorities, and was not focused on recruitment of self-reported respondents with mental illness, our final sample of black respondents who resort mental illness is small (N = 5). Consequently, our sample does not allow for full consideration of the interaction of black*mentally ill statuses. Supplemental analyses suggest that this group of respondents (black individuals who also report mental illness) are unlikely to support increased mental health crisis training for police officers (e.g., they are about half as likely as non-black respondents with mental illness to say mental health crisis training is ‘good for individuals with mental illness’ and ‘good for all’), but these results are not presented in Table 2 due to concerns about drawing conclusions from this small number of cases.

7. Supplemental analyses considered whether these outcomes are better modeled as continuous rather than dichotomous variables. Retaining the original coding for the various dependent variables and employing OLS regression, we found that the results of the OLS regression analyses did not differ substantially from the logistic regression analyses presented in Table 2, and the primary conclusions are the same, regardless of whether the dependent variable is treated as a continuous or a dichotomous outcome.

8. In supplemental models where the neutral responses are eliminated from the analyses – so the logistic models are only comparing ‘disagree’ responses with ‘agree’ responses – the interaction term for arrest*mental illness is significantly positive (B = 3.11, odds ratio = 22.36, p < .05) in Model 2, suggesting that respondents who are at highest risk for encountering the police while experiencing a mental health crisis are most likely to favor prioritizing mental illness training over racial bias training for law enforcement officers. All other comparisons between this supplemental analysis and Table 2 logistic regression results are substantially the same.