Lifestyle-Exposure Theory as a Framework to Analyze Victimization of People Experiencing Homelessness

ABSTRACT

Segmentation analysis and logistic regression were used to test the probabilistic connection between exposure to high-risk situations and victimization events, as posited by Lifestyle-exposure theory, in a sample of homeless people. The results support the hypothesis put forward. First, those who had engaged in risky behaviors had suffered victimization events to a greater extent. Second, this was particularly true for participants who had done so more frequently or had engaged in a wider range of such behaviors. The highest risk profile included those who had been arrested on several occasions and also reported having used drugs during the previous month or, otherwise, had served a sentence different from prison in the past. Implications of these findings are discussed considering that homeless people’s engagement in risky behaviors, as well as, in general, their greater degree of exposure to situations in which risk of victimization is high, often stem from the situation they are going through. A major conclusion is that any effort to eradicate violence against homeless people should contemplate strategies for combating homelessness itself.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ Interested readers will find all the information about the survey in INE (2012a).

² The survey questionnaire can be found at INE (2012b).

³ Even with the limitations they imply – which will be discussed in detail below—, these variables have been deemed as the best available indicators of the individual’s involvement in unlawful behavior and, accordingly, of a deviant lifestyle.

⁴ Cases with Z-scores greater than |3.0|.

⁵ The authors qualify that this cannot be applied to all professions.

⁶ Cases with Z-scores over |3,0|.

⁷ It should be remembered that this category is not limited to the unemployed, although this was the predominant response (almost 75% of the sample).

⁸ Although the movements from one category to another for the ordinal variables “time spent homeless” and “usual degree of alcohol consumption” cannot be considered as equivalent, these variables were introduced in the logistic regression model as quantitative rather than categorical variables. As proposed by Agresti (2010: 75), a comparative analysis was conducted among the values of Akaike and Bayesian information criteria – AIC and BIC, respectively – for a set of four complete models including one or both variables as categorical and both as quantitative. Following the criteria suggested by Hilbe (2011: 70 and 73), the best-fit model was the latter. In addition, the models in which these variables were introduced as categorical using the lowest value of the scale as the reference category showed gradual increases in the coefficients and odds ratios (OR) for each successive category.

⁹ At the bivariate level, sex was not related to victimization [OR_Man = 1.16, p = 0.076]. A significant relationship between sex and victimization was only found when the variables “have you ever been reported to the police?” and “have you ever been arrested?” were introduced in the regression model. Both of them were significantly related to sex: male participants had been reported to the police [once: OR_Man = 2.29, p < 0.001; more than once: OR_Man = 3.23, p < 0.001], as well as arrested [OR_Man = 2.57, p < 0.001; more than once: OR_Man = 3.53, p < 0.001] to a greater degree. These variables were also related to victimization [reported to the police once: OR = 1.89, p < 0.001; reported to the police more than once: OR = 3.40, p < 0.001; arrested once: OR = 1.75, p < 0.001; arrested more than once: OR = 3.39, p < 0.001]. Therefore, it seems we have found a suppression effect: the abovementioned variables are suppressors with respect to the relationship between sex and victimization experiences (Ato and Vallejo 2011; Jacob et al. 2003, Lancaster 1999).

¹⁰ To this effect, observations with standardized Pearson residuals higher than |2| (Hosmer David, Stanley, and Sturdivant 2013: 360 and 370), |DfBetas| > 2/√n (Belsley, Kuh, and Welsch 2004: 28) – cutoff point: |0,038|—, and D > 0.5 (Cook and Weisberg 1999: 358) have been deemed problematic. As can be seen, in no case was the threshold for D exceeded, 20 cases were above the cutoff point for residuals and 15 for differences in betas. When the analyses were repeated after excluding, for purely analytical purposes, the problematic observations, results were essentially convergent with the original findings.

¹² For ordinal variables, a value can only be aggregated to another one in the segmentation if they are consecutive in the scale. The ordinal variables Have you ever been arrested? and Have you ever been reported to the police? have three response options: No, Yes, once, and Yes, more than once. Accordingly, the results must be interpreted as follows: the node “No” or less refers to the values of the variable up to No. This node only includes the No responses because there are no lower values. Higher than “No” through “Yes, once” refers to the response Yes, once. Higher than “Yes, once” refers to the response Yes, more than once. Finally, Higher than “No” refers to both the responses Yes, once and Yes, more than once..

¹³ Risk estimate. Also known as overall rate of incorrect classification, it determines the predictive capacity of the segmentation. Its calculation is based on the probability of making mistakes in predicting the dependent variable with the information provided by the independent variables introduced in the segmentation. It indicates the proportion of cases of the sample which have been incorrectly classified with the information of the terminal nodes (Escobar 2007: 69).

¹⁴ Standard error.

¹⁵ Relative risk reduction. This is a relative measure of the reduction in error that results from the segmentation analysis. It indicates the proportional reduction in the risk of making mistakes in predicting the dependent variable with the information provided by the independent variables. With regard to its calculation, see Escobar (2007: 70).

¹¹ The variables “Have you ever been reported to the police?” and “Have you ever been arrested?,” particularly the latter, are both related to having been in prison. Thus, as regards the former: Chi-square(2) = 830.657, p < 0.001; Cramer’s V = 0.492, symmetric Lambda = 0.131. With regard to the second one: Chi-square(2) = 1243.988, p < 0.001; Cramer’s V = 0.602, symmetric Lambda = 0.241.

Additional information

Notes on contributors

Patricia Puente Guerrero

Patricia Puente Guerrero, Ph.D., is an Assistant Professor of Psychology at the University of Extremadura, in Spain. Her thesis focused on victimization of people experiencing homelessness. She has also studied stereotypes, prejudice, and discrimination against this particularly vulnerable group, as well as their criminalization, and has addressed the role all these factors play in homeless people’s suffering of direct, structural, and cultural violence.