Many people do things to make themselves feel safe in their homes, such as lock their doors, turn on lights, have a dog, use security systems, and participate in Neighborhood Watch. All of these behaviors are related to crime prevention. This study uses weighted data from the 2,080 respondents to the 2009 Anchorage Community Survey to determine relationships between these “self-protective behaviors” and perceptions of safety, social engagement, and collective efficacy. (Collective efficacy refers to the social cohesion of a group and the willingness of members to intervene on each other’s behalf.)
It examines the assumptions that (a) people will engage in more self-protective behaviors if they generally do not feel safe; (b) those who interact a lot with others in their communities are more likely to carry out self-protective behaviors that involve working with others (specifically, attending Neighborhood Watch meetings and developing a signal for “danger” with neighbors); and (c) people who rate their neighborhoods high on measures of social cohesion and informal social control will be more likely to engage in self-protection activity that involves working with others. Research about such assumptions can assist law enforcement, policymakers, and planners in understanding how and why people engage in certain behaviors, and can also help in developing strategies and determining allocation of resources for crime prevention in neighborhhoods.
Data and Methods
Data for this study come from the 2009 Anchorage Community Survey (ACS)—a mixed-mode (mail and Internet) survey of adult heads-of-household living in the Anchorage municipality. The survey included questions designed to measure, among other things, experiences of criminal victimization, perceptions of safety, feelings about quality of life and perceptions of social cohesion in respondents’ neighborhoods, involvement in community activities and groups, satisfaction with municipal services, and attitudes toward the criminal justice system.
A total of 4,702 people were included in the sample. Many of the addresses used to contact respondents were unusable for various reasons (n=546), and 15 people on the mailing list were deceased. After a data collection period of several weeks, during which four items were mailed sequentially to respondents (a notification letter, the questionnaire and $2 bill incentive, a reminder postcard, and another questionnaire), a total of 2,080 completed surveys were received. The response rate was 50.8 percent.
A recognized problem with mail surveys (and indeed, broadly-sampled community surveys in general) is large variation in response rates across different sub-groups in the population. Women, white people, and older people are considerably more likely than men, racial and ethnic minorities, and young people to respond to surveys. This is a problem because data from surveys are not really representative of the population if certain groups are overrepresented and certain groups are underrepresented. One way to deal with this is to over-sample from groups that you expect will have low response rates. Another approach, which was used here, is to weight the data after the survey has been administered. Briefly, weighting involves using data, such as the U.S. Census, to take into account the demographic characteristics of the surveyed population (in this case, adults living in Anchorage), and then comparing these to the demographic characteristics of the people who responded to the survey. The responses from people whose demographic characteristics are overrepresented compared to the whole population (for example, females, whites, and older people) are given less weight—that is, they count “less”—relative to responses from people whose demographic characteristics are underrepresented in the survey data. Using weighted data allows researchers to be more confident that their findings can be generalized to the whole population, and are not just descriptions of the people who sent back their surveys.
The dependent variables (what we are trying to explain) concern self-protective behavior. A question on the ACS showed respondents a “list of things people may do for self-protection or to feel more secure in their homes and neighborhoods,” and asked them to check all that applied. As shown in Table 1, the most common self-protective behavior, selected by almost all survey participants, was “lock doors at night and when you are away from home.” There were nine other self-protective behaviors included on the list: “lock doors during the day and when you are at home,” “have a dog,” “keep a firearm in the home,” “use a security system on vehicle(s),” “use a home security system,” “attend Neighborhood Watch meetings,” “develop a signal for ‘danger’ with neighbors,” “have outside/automatic lights to deter prowlers,” and “take self-defense lessons.”
The final dependent variable was derived from a composite measure of the ten self-protection behavior variables. It is essentially a count of the number of behaviors each respondent reported doing to feel more secure in their homes and neighborhoods. This count ranges from 0 to 10; the mean was 4.0 (see Table 1).
There are four independent variables (things related to or correlated with the dependent variables) used in the analyses: (1) perception of safety, (2) social engagement, (3) social cohesion, and (4) informal social control. Perception of safety is a composite measure of responses to ten questions in the ACS asking people how safe they feel in a variety of scenarios in their neighborhoods, both during the day and at night (see Table 2). People reported feeling least safe while walking along trails at night, and most safe at home alone during the day. The value of the composite measure ranges from 1 (if the response to all ten scenarios was “very unsafe”) to 4 (if all ten responses were “very safe”). The mean score of the weighted sample was 3.29, that is, between “reasonably safe” and “very safe.”
Social engagement is also a composite measure derived from respondents’ reported involvement in different types of local organizations: religious groups; local political organizations; block groups, tenant associations, or community councils; business or civic groups; ethnic or nationality clubs; and Neighborhood Watch groups. As shown in Table 2, 44.1 percent of respondents were involved in a church, synagogue, mosque, temple, or other religious organization, while only 4.1 percent said they were involved with a local ethnic or nationality club. Over 40 percent of survey participants reported no involvement in any of the listed organizations. The social engagement variable ranges from 0 to 6, with 0 indicating no reported involvement in the listed organizations, and 6 indicating involvement in all six of the listed organizations. The mean score of the weighted sample was 0.91; the average respondent was involved in just under one of the listed organizations.
Social cohesion and informal social control are derived from ten questions first developed by Sampson, Raudenbush and Earls in their collective efficacy research associated with the Project on Human Development in Chicago Neighborhoods. The five questions on the ACS concerning social cohesion included “People in my neighborhood can be trusted,” and “Mine is a close-knit neighborhood,” and the five questions measuring informal social control included “One or more of my neighbors could be counted on to intervene if a fight broke out in front of their home,” and “One or more of my neighbors could be counted on to intervene if children were spray painting graffiti on a local building.” Table 2 shows the means and standard deviations for the ten measures. The mean score of the weighted sample on the social cohesion composite is 3.52; the mean score of the weighted sample on the informal social control composite measure is 3.59. These scores fall between “neither agree nor disagree” and “agree” on a five-point Likert scale.
Additional variables used included age, gender, race, gross household income, and home ownership (see Table 3). Age is a continuous variable that represents respondents’ replies to the question “How old were you on your last birthday?” The mean age of the weighted sample was 48.5 years old, with a standard deviation of 14.5 years. For some of the analyses discussed in this article, age was collapsed into age ranges, as shown in Table 3. Gender is a dichotomous (only 2 values) variable where male=1 and female=0; 44.2 percent of the weighted sample was male. Race is also dichotomous. Whites, which comprised 79.6 percent of the weighted sample, were coded as 1, and all other racial groups were coded as 0. Gross household income was measured as income ranges; respondents were asked to indicate into which category their gross household income fell: less than $20,000, $20,000 to $34,999, $35,000 to $49,999, $50,000 to $74,999, $75,000 to $99,999, and $100,000 or more. Over one-third of the weighted sample reported a gross household income of at least $100,000, and slightly over twelve percent of the weighted sample said their household income was below $35,000. According to the Anchorage Economic Development Corporation, the median household income in Anchorage in 2009 was $70,151. Home ownership was coded as homeowner=1 and renter=0. Homeowners comprised 80.3 percent of the weighted sample.
A number of analyses were conducted to explore the relationships between the dependent and independent variables. The results from the univariate analysis are discussed in the above section of this article and are presented in Tables 1, 2 and 3. The next step was bivariate analysis, which involves looking for relationships between a dependent variable and an independent variable using tabular analysis. For example, are homeowners more likely than renters to report that they keep a firearm in the home for protection? Is household income related to whether a survey respondent has a security system for their vehicle? To facilitate this approach, many of the independent variables were collapsed into a smaller number of categories: the three composite measures of perception of safety, social cohesion, and informal social control were each collapsed into three categories: “low,” “medium,” and “high” based on their standard deviations. These categories allow for comparison among respondents who scored unusually low, unusually high, and close to average on these composite measures. As Table 4 shows, most respondents fall into the “medium” category. The fourth composite measure, social engagement, was also collapsed into three categories: “low,” no reported participation in any of the listed organizations; “medium,” participation in one organization; and “high,” participation in two or more of the listed organizations.
For the series of bivariate analyses, all relationships between each dependent variable and each independent variable were examined for strength of the relationship and direction, if appropriate. Direction can be either positive or negative; a positive direction indicates that as one variable increases, the other increases. A negative direction indicates that as one variable increases, the other decreases. The relationships were also examined for statistical significance. An interesting thing about statistical significance is that the likelihood of finding it depends a lot on the size of the sample. The smaller the sample, the stronger the relationship between two variables has to be for it to be statistically significant. With a large sample, such as that of the Anchorage Community Survey, even weak and unmeaningful relationships can be statistically significant. In this study, all but one of the 90 bivariate analyses were statistically significant. So this doesn’t really tell us anything. It is more important to look at the strength of the relationships. To do this, two statistics which are measures of the proportionate reduction in error (PRE) were used: Gamma when the independent variable had more than one category, and Cramer’s V when the independent variable had only two categories (such was the case for gender, race, and home ownership). PRE measures tell us how much better our ability to predict the value of the dependent variable becomes if we know the value of the independent variable. For example, if we know the gender of a person, are we better able to predict whether that person takes self-defense lessons than if we have no knowledge of the person’s gender? Gamma is a directional PRE measure, which means that it tells us the strength of the association as well as the direction of the association. To illustrate, suppose we are interested in the relationship between gross household income and attendance at Neighborhood Watch meetings. Gamma will show whether there is a relationship between these two variables, how strong it is, and whether respondents from households with larger incomes are more likely to go to Neighborhood Watch meetings than people from households with smaller incomes.
Findings and Discussion
Tables 5 and 6 show the results of the bivariate analyses. Numbers represent the weighted percentage of respondents who answered that they did perform the assorted self-protective behaviors. For example, in Table 5, 97.5 percent of respondents who scored “low” in social cohesion reported locking their doors at night and while away from home. Percentages should be read across by rows to get a sense of whether the independent variable is related to the dependent variable. The greater the differences in the percentages, the more likely that there is a meaningful relationship. A more statistically valid approach is to look at the PRE measure (Gamma or Cramer’s V). The closer this measure is to 1.00 or -1.00, the stronger the relationship between the two variables. A good rule-of-thumb is that the relationship is weak if the PRE measure is less than .100, moderate if it is between .100 and .200, moderately strong if it is between .200 and .300, and strong if it is above .300. Also, in the case of Gamma, the sign of the measure (positive or negative) indicates the direction of the relationship.
An examination of Table 5 shows that while there appear to be patterns based on the percentages, social cohesion is not strongly related to any of the ten self-protective behaviors, though it is moderately related to locking doors during the day and while at home, and using a home security system. As respondents reported higher levels of social cohesion, their tendency to lock their doors during the day and while at home declined. It may be that because they are more trusting of their neighbors, they are less concerned about daytime intrusions into their homes. Conversely, respondents were more likely to use a home security system if they scored higher on social cohesion. This may be an illustration of Robert Frost’s observation that “good fences make good neighbors.”
Respondents who scored low on social cohesion were more likely, compared to those who scored higher, to report taking self-defense lessons, while the survey participants who scored high on social cohesion were more likely to attend Neighborhood Watch meetings and develop a signal for “danger” with their neighbors than those who scored lower. These findings are not surprising. People who feel connected to others would be expected to rely upon others to some extent for their personal safety, while those who feel less connected might be inclined to seek more independent approaches.
Informal Social Control
Informal social control was strongly and negatively related to taking self-defense classes. Respondents who disagreed that their neighbors would be likely to intervene in the event of occurrences like children spray painting graffiti or skipping school to hang out on a local corner were more likely to report taking self-defense lessons. Another negative relationship, though not as strong, is evident between informal social control and locking doors during the day and while at home. Respondents who scored lower on informal social control were also more likely to have outside or automatic lights to deter prowlers and to use a security system on their vehicles. All these relationships are in the expected direction. Those who believe their neighbors are not likely to intervene in different situations are more apt to take matters of defending themselves and their property into their own hands.
All of the strong or moderately strong relationships between social engagement and self-protective behaviors were positive, that is, the more organizations respondents said they were involved with (zero, one, or two or more) the more likely they were to do all the listed self-protective behaviors. An exception was having a dog, which varied little depending on the amount of reported social engagement. Not unexpectedly, the strongest associations were with attending Neighborhood Watch meetings and developing a signal for “danger” with neighbors. Because it uses a composite measure, this analysis does not differentiate among involvement in say, a religious community or business group, and involvement in a more explicitly neighborhood-level group (such as a tenant association). It is possible that social engagement with particular organizations is associated with some self-protective behaviors, but not with others.
Perception of Safety
The general pattern is that perception of safety is negatively associated with most of the self-protective behaviors. Respondents who said they feel safer are less likely to do things like lock their doors, either during the day or at night, at home or when away, and take self-defense lessons. Though the association is somewhat weaker, survey participants who feel safer are also less likely to use outside or automatic lights or a security system for their vehicles, or develop a signal for “danger” with neighbors.
There was little difference between women and men in the survey with respect to most of the reported self-protective behaviors (see Table 6). However, women were more likely than men (80.5% versus 71.3%) to say they lock their doors when at home and during the day, and men were more likely than women to report keeping a firearm in the home (59.1% versus 45.2%).The relationship of gender to (1) door locking at home and during the day and (2) keeping a firearm is moderate in both cases.
Survey respondents from minority racial groups differed from white respondents not in terms of how many of the ten self-protective behaviors they reported doing, but which ones. As shown in Table 6, compared to white respondents, minority respondents were more likely to lock their doors during the day and when at home (84.0% versus 74.5%), take self-defense lessons (12.1% versus 7.9%), attend Neighborhood Watch meetings (9.8% versus 6.9%), and develop a signal for “danger” with neighbors (8.5% versus 2.3%). White respondents were more likely than minority respondents to have a dog (50.1% versus 37.8%), use outside or automatic lights (54.3% versus 49.0%), and keep a firearm (52.8% versus 46.1%). With the exception of developing a signal for “danger” with neighbors, all of these relationships are weak, and should thus be interpreted with caution. It appears these behaviors could be categorized as things you do versus things you buy. A relationship to income may be connected with these categories.
The self-protective behaviors reported by homeowners and renters in the survey were moderately different. Renters reported doing fewer of the ten possible behaviors, and were much less likely than homeowners in the survey to have outside or automatic lights (39.6% versus 57.0%), keep a firearm in the home (39.1% versus 54.6%), have a dog (31.6% versus 51.7%), or use security systems at their homes (8.6% versus 24.6%). Home ownership is a reasonable proxy for income; it is possible that the greater one’s income, the more one can spend on security systems and firearms. Another explanation is that renters have little incentive to install security systems, particularly in their homes, as they do not own the property. Also, renters might not be allowed to install a security system, or may see no need to do so if the rental property already has some form of crime prevention implemented by the landlord. With respect to dogs, many renters are prohibited under conditions of rental agreements from owning dogs, so it is not surprising that they typically did not report keeping a dog for self-protection. Homeowners and renters in the survey reported similar levels of door locking, whether during the day or at night, or when at home or away, and both groups had the same likelihood of developing a signal for “danger” with neighbors. Homeowners who answered the survey were more than twice as likely as renters to say they attend Neighborhood Watch meetings (8.4% versus 3.6%); this was, however, a weak association.
Some of the self-protective behaviors seemed to increase with age, such as locking doors at night and while away (even though this is a behavior practiced by just about everyone in the sample), using outside or automatic lights, attending Neighborhood Watch meetings, and developing a signal for “danger” with neighbors. The latter two behaviors involve working with others; it might be that older people have lived in their homes longer, and thus are more likely to know their neighbors and work with them to feel safer in their homes and communities, or that they have more time (presumably because they are less likely to be raising children and working full-time compared to younger adults) to devote to interaction with their neighbors. Locking doors during the day and while at home and taking self-defense lessons appeared to decrease with age. Older people may be reluctant to partake of self-defense classes because they feel less physically able compared to younger people (there is a large drop in those reporting to have taken self-defense lessons between the age groups of 45-54 and 55-64). Dog ownership reached a peak among those aged 45-54 years, but dropped among older respondents. A similar pattern was observed for use of security systems on vehicles, and to a lesser extent, use of security systems in the home. While only a weak relationship, firearm ownership increased with age up to a point; the oldest respondents had the lowest reported level of keeping a firearm (43.9%).
Gross Household Income
The strongest relationship involving income was a negative one—people who reported the highest household incomes were least likely to work with neighbors to develop a “danger” signal. Those with incomes in the middle range ($50,000-$74,999) were the most likely to do this. A moderately weak and negative association was seen between taking self-defense lessons and gross household income; respondents from the households with the lowest incomes had the highest reported participation in that activity. Household income was moderately and positively associated with keeping a firearm in the home, and having security systems on a vehicle or at one’s home. The only non-statistically-significant relationship in all the bivariate analyses was gross household income and attending Neighborhood Watch meetings.
This study looked at the relationships between self-protective behaviors and how safe people feel, how much they interact with others in their community, and how they rate their neighborhood on measures of social cohesion and informal social control. It also examined the assumptions that (a) people will engage in more self-protective behaviors if they generally do not feel safe; (b) those who interact a lot with others in their communities are more likely to carry out self-protective behaviors that involve working with others; and (c) people who rate their neighborhoods high on measures of social cohesion and informal social control will be more likely to engage in self-protection activity that involves collective activity.
This analysis validated two of the three assumptions tested in the research. Feelings of safety are inversely related to self-protective behavior. The less safe one feels, the more one will do things to feel safer (or the more one does things to feel safer, the less safe one feels).
Involvement in local organizations is strongly and positively related to engaging in collective self-protective behavior, that is, getting involved with Neighborhood Watch and developing a signal for “danger” with one’s neighbors.
The third assumption, that people who report greater feelings of community or think their neighbors will do something if crime or disorder is occurring are more likely to engage in collective self-protective behavior, was partially supported by the research. Social cohesion was moderately and positively related to both attendance at Neighborhood Watch meetings and developing a signal for “danger” with neighbors, while informal social control was not related to participation in Neighborhood Watch in any apparent way, and was negatively associated with developing a signal for “danger” with neighbors.
Crime prevention advocates recognize that the police and other agencies of the criminal justice system have limited resources to adequately deal with crime after the fact, let alone prevent it before it happens. Thus, a goal that is common to most crime prevention programs is to encourage individuals to take more personal responsibility for their safety. This study suggests a variety of strategies for those who wish to promote more self-protective behavior. First, if fear of crime is associated with self-protective behavior, it might make sense to target particularly fearful groups for interventions aimed at increasing the use of these behaviors. Whether this is ethical or even effective at reducing crime is unclear. There is some evidence from other studies that teaching people how to do things that supposedly will make them safer can actually make people more afraid of being a crime victim. Also, the relationship between fear and actual risk of victimization is inverse; for example, members of the most fearful groups in American society are often the least likely to be victimized. Directing scarce crime prevention resources at the people at the lowest risk of victimization is not efficient. To quote British crime prevention researchers Mike Hough and Nick Tilley, this is not “getting the grease to the squeak.”
This study shows a strong relationship between social cohesion and participation in local organizations and collective self-protective behavior. This suggests that a fruitful approach would be to direct efforts toward those who are already somewhat socially connected to others with the belief that those persons are more likely to engage in crime prevention activities. However, given that much research indicates that social connectedness in general seems to provide protection against criminal victimization, focusing more resources on connected individuals is also not especially efficient.
In reviewing the results of this study, it is necessary nonetheless to consider the limitations of the survey, particularly the difficulty in surveying hard to reach segments of the population, and the impact this may have on estimates. As noted earlier, weighting the data helps address some of these limitations.
Future research on this question should incorporate other possibly explanatory variables, such as victimization experience, household income, education, length of time in one’s current home, and attitudes towards the efficacy of the criminal justice system. While the social cohesion and informal social control measures were generally poor at explaining self-protective behavior, the inclusion of other variables measuring neighborhood-level factors, such as social and physical disorder and reported crime, would help in understanding if a person’s immediate environment affects whether that individual engages in self-protective behaviors, and if so, what type.
Sharon Chamard is an associate professor with the Justice Center.