The Analysis of the Relationship between residents’ Perceived Probability of Property Victimization in Dwellings and the Level of Home Security Systems

Abstract

Using the dataset of the first nationwide victimization survey in Azerbaijan, the current paper analyzed whether home security systems have a relationship with the perceived probability of property crime victimization. The ordinal logistic regression analysis was employed. Examination of the distribution of security systems identified that dwelling type, residential location and income had a higher level of significance than the perceived risk of property crime victimization in predicting the level of security systems at home. As to the predictive capacity of security systems in relation to the perceived risk of property crime victimization, the former is statistically significant. Additionally, being a victim of property crime, residential location, age, and perceived neighborhood disorder level predicted the perceived risk of property crime victimization.

Keywords:

• Home security systems
• victimization
• Azerbaijan
• property crime
• ordinal logistic regression

Introduction

The home security system market is a large industry on a global scale, with an estimated value of US$ 26.6 Billion in 2019 (IMARC Group, 2020). Some scholars point out the increasing societal obsession with security and the installation of security systems, such as surveillance, secured entries, gated communities, and fencing (Blakely & Snyder, 1997; Low, 2004).

Does person’s residential characteristics have a relationship with either fear of crime or perceived probability of property crime victimization? The current literature consists of a plethora of research examining (a) the relationship between housing contexts and fear of crime (Alvi et al., 2001; Brantingham & Brantingham, 1991; Palmer et al., 2005), (b) Crime Prevention Through Environmental Design’s (CPTED) effect on fear of crime in neighborhood (Luo et al., 2016; Hedayati-Marzbali et al., 2012) and (c) the importance of geographic location in shaping individuals’ fear of crime (Luo et al., 2016; Ferguson & Mindel, 2007; Wyant, 2008). There are also scholarly works exploring whether a sense of safety within the community influences residents’ defensive or participatory behaviors, such as installing extra locks on doors and windows (e.g. Reid et al., 1998; Yuan & McNeeley, 2016). Finally, some scholars have explored the effects of CCTV upon fear of crime in a public place in Seoul, which reported an inverse relationship between number of CCTV cameras and fear of crime, but no effect on the perceived risk of crime (Cho & Park, 2017).

However, an investigation of the association between home security measures and perceived probability of victimization (PPV hereafter) has been limited, though some scholars (see Evans, 2001; Hirschfield et al., 2004; Vilalta, 2012; Mawby & Simmonds, 2003) have explored the effect of the latter on fear of crime. In addition to the dwelling’s security features and its association with fear of crime, there is scant research exploring the impact of CPTED on a sense of safety among students on campus (e.g. Shariati & Guerette, 2019). In light of the findings of the current literature, one may view the existing state of knowledge on the association between the perceived probability of property victimization and security systems as an interesting puzzle to solve. That is, on one hand, there is a large home security industry globally, one of which premises is to reduce fear. On the other hand, despite a great deal of work on crime prevention, the focus has been mostly on the impact of crime prevention techniques on crime and victimization rates (see Hirschfield et al. 2004; Tseloni et al., 2017), not the perceived probability of property victimization.

Roth (2018) notes that both fear of crime and personal assessment of the risk of victimization can affect one’s decision to adopt precautionary measures, home security systems inclusive. This study makes a modest attempt to discover whether there is any association between the level of home security systems and residents’ perceived risk of being a victim of property crime. Unlike most of the studies cited above, the current analysis relies on data from a national sample covering a large variety of households.

The perceived probability of victimization and fear of crime

The existing body of literature suggests a distinction between fear of crime (an emotional reaction) and risk of victimization (a cognitive assessment) (see Ferguson & Mindel, 2007). Some scholars have used a survey to display the differences between these concepts (Rountree & Land, 1996; Rader, 2004). As highlighted by Pleggenkuhle and Schafer (2018, p. 384), “citizens might express fear of crime, but perceive they are at low risk of being victimized for a variety of reasons.” While it is the risk of victimization the present analysis focuses on, the paragraphs below present various studies (they focus on either fear of crime or perceived risk of victimization). However, it is worth noting the relative imbalance in the literature. That is, Chon and Wilson (2016, p. 311) highlight that “although an abundance of studies on FC exists, the works on PRV are relatively rare”.

Significant work has been done on different individual and community-level factors’ influence on one’s reported levels of either fear of crime or PPV. Women (Chon & Wilson, 2016; Helfgott et al., 2020; Prechathamwong & Rujiprak, 2018; Sutton & Farrall, 2004) and those with prior direct victimization (Chon & Wilson, 2016; Ferguson & Mindel, 2007; Schafer et al., 2006; Russo & Roccato, 2010; Vilalta, 2011a) and vicarious victimization (Özaşçılar & Ziyalar, 2017; Skogan & Maxfield, 1981) have been found to hold significantly higher levels of fear. Some researchers have identified a positive association between fear and age (De Donder et al., 2005; Helfgott et al., 2020), while others have found higher levels of fear of crime among younger respondents (Vilalta, 2011a; Yun et al., 2010). Individual’s socioeconomic status, as well as the level of education, have also been found to correlate with either fear of crime or perceived risk of becoming a victim (Hummelsheim et al., 2011; Kanan & Pruitt, 2002; McGrath & Chananie-Hill, 2011; Roccato et al., 2011). Chon and Wilson (2016), for instance, identified a positive association between socioeconomic status and perceived risk of burglary. In addition to these demographic features, trust in the police and public satisfaction with police service in the neighborhood both have been found to affect fear of crime (Alda et al., 2017; Dammert, 2014; Renauer, 2007; Scheider et al., 2003).

Additionally, there are environmental cues that affect a person’s level of fear, since humans are visually sensitive to their environment and subjectively feel safe or unsafe depending on the physical and social cues in the environment around them (Nasar & Fisher, 1993). Signs such as social and physical uncivilized behavior and disorder, in the form of darkness, unfamiliar environments, and unknown bystanders (LaGrange et al., 1992; Ross & Jang, 2000; Warr, 2000), can induce fear. Therefore, the perceived level of neighborhood disorder is a significant community-level determinant of fear of crime (Chataway & Hart, 2016; Markowitz et al., 2001). Another community-level factor shaping the fear of crime is collective efficacy. By way of definition, collective efficacy refers to the “perception that neighbors can be trusted to engage in social control for the community’s benefit, which results in alleviating the fear of crime among residents” (Gibson et al., 2002, p. 543). Neighborhoods that lack collective efficacy generally generate mistrust among their residents and elevate fear (Brunton-Smith & Sturgis, 2011; Karakus et al., 2010). Members of a community with high levels of mutual trust are likely to share goals of neighborhood safety and to be willing to work together to achieve those goals (Sampson & Raudenbush 1999). However, several studies from the developing world indicate that neighborhood integration and cohesion do not always lead to lower levels of fear, because stories about crime spread rapidly in close-knit neighborhoods (Villarreal & Silva, 2006). There is even evidence suggesting that moving beyond neighborhood conditions, broader insecurity features such as financial and educational insecurities impact upon one’s perceived risk of victimization (Caliso et al., 2020).

Home security systems

Since Newman (1972) presented the role of environmental and housing design in reducing the chances of being a victim of crime, there is an increasing interest in the characteristics of the physical environment and their effect on the fear of crime, though none of them has adopted the perceived probability of being a victim as an outcome variable. One line of studies has focused primarily on particular housing contexts, such as gated communities (Vilalta, 2011b; Wilson-Doenges, 2000) or public housing (Alvi et al., 2001; DeLone, 2008; Palmer et al., 2005; Staunton, 2006). Thus, DeLone (2008) found that living in an elderly-only tower is associated with a reduced level of fear of crime, while residing in mixed-age towers generally elevates fear of crime in Omaha, U.S. Staunton (2006), on the other hand, identified a greater sense of fear of crime among the residents in the private rented sector than those who are owner-occupiers or rent from registered social landlords in Birkenhead, England. In terms of Crime Prevention Through Environmental Design (CPTED) and the fear of crime, studies from Malaysia (Hedayati-Marzbali et al., 2012) and Australia (Minnery & Lim, 2005) have failed to identify a significant relationship between CPTED and fear of crime. While this literature highlights the importance of the physical environment in shaping perceptions of crime and safety, its focus is not on security systems at an individual’s own dwelling.

Instead of the perceived risk of victimization, fear of crime and its relationship with home security has been explored in many studies. Studying the relationship between the fear of crime and home security systems in Mexico, Vilalta (2012) found no effect of a home security system on the reported levels of fear of crime when a person was at home alone. Vilalta (2012) argued that his results demonstrate that having a security system at home may not necessarily translate to feeling more secure. Drawing upon the results from a baseline survey of 289 respondents carried out in Merseyside, north-west England, Hirschfield et al. (2004) failed to identify an association between the number of home security measures and residents’ perceptions of safety within their homes, irrespective of the time of day. In an examination of the fear of burglary and the presence of a number of home security features, Evans (2001) found that the only security feature that reduced the fear of burglary was a door chain. Another study from England found that while target hardening improved residents’ sense of security, the differences between the two groups (experimental and control) were minimal when they answered standard “fear of crime” questions. Finally, in a relatively recent survey analysis by Rader and Haynes (2014) considered certain dwelling features (i.e. installing door locks) as one of the protective behaviors. Their analysis found that protective behaviors mattered only for concerns about crime for others, not for self. To our knowledge, only one paper has looked at the perceived risk of victimization in the context of security systems. Measuring the relationship between security measures and the perceived likelihood of victimization, Shibata and Nakayachi (2022) used an experimental vignette-based approach among 180 Japanese university students, and did not consider a person’s dwelling as their focus. They rather presented scenarios where new security measures are supposedly going to be implemented across the university campus, rather than residential dwellings. Shibata and Nakayachi’s results indicate that when security measures are implemented to deal with serious crimes, the perceived risk tends to increase.

Intriguingly, research has also found that fortification and the installation of security systems may have unintended consequences, such as social isolation and decreased social interaction (Green et al., 2002). Furthermore, security systems may signal the presence of danger. For instance, both Gill and Spriggs (2005) and Ditton (2000) have revealed that closed circuit television in an area makes the location seem more problematic than it was previously, and thus may deter some people from visiting it. Similar findings have been reported by others as well (e.g. Bachman et al., 2011; Schreck et al., 2003).

The context of Azerbaijan

A south Caucasian jurisdiction, Azerbaijan is an oil-rich, economically developing and middle-income country. Following the establishment of the new state in 1991, Azerbaijan has developed both socially and economically. GDP increased dramatically from US$3.9 billion in 1993 to US$75.2 billion in 2014, before plunging to US$54.6 billion in 2021, primarily as a result of oil shocks in the global markets (The World Bank, 2022).

The choice of Azerbaijan as a geographic unit of study is interesting for several reasons. Azerbaijan has not been considered in criminological studies on the fear of crime or PPV, although some other developing countries do feature in the literature on the former (e.g. Adu-Mireku, 2002; Alda et al., 2017; Johnson, 2006; Karakus et al., 2010; Vilalta, 2011a, 2011b, 2012). From the point of view of crime rates, as shown below, Azerbaijan has a considerably lower number of officially registered crimes per 1,000 inhabitants than other societies that have been extensively studied.¹

To the authors’ knowledge, there is no information available about the perceived probability of being a victim of crime in Azerbaijan. However, the scant information available about the assessment of personal safety may provide some glimpses. In Gallup’s Law and Order Index compiled for 2019² (Gallup, 2020), Azerbaijan scored a fairly high value of 91 out of 100 points, placing it 12^th (in the upper quantile) among the 144 countries covered.

Dynamics of crime in Azerbaijan

According to the Criminal Code of the Republic of Azerbaijan, there are 14 main types of property crimes enlisted on Articles 177-189.1 The following offenses are categorized as property crimes: burglary, robbery, fraud, embezzlement, theft. However, an important note must be made here regarding the crime data. The official data does not list every single offense per crime category (i.e. property crime category). Instead, only a limited types of property offenses is shown, as other acquisitive crimes are collapsed into the “other” category.

Historically, property crimes have steadily increased (State Statistics Committee, 2021), but this increase needs to be interpreted with caution. The introduction of emergency hotlines by law enforcement bodies, changing recording practices by the police and other relevant agencies, as well as evolving public attitudes toward crime reporting may have influenced the rates. The latest data available at the time of writing show that the aggregate crime rate per 1,000 population in Azerbaijan stood at 2.60 in 2020³. Property crime (0.9 per 1,000) fraud (0.06 per 1,000), illegal possession of drugs (0.4 per 1,000), traffic violations, (0.23 per 1,000) and hooliganism (0.06 per 1,000) were the five most prevalent officially recorded crimes in the country. Nine thousand four hundred property crimes (0.9 per 1,000) were recorded in 2020.⁴ To put it into the regional context, the data from neighboring state, Russia shows an aggregate crime rate of 14,86 per 1,000 in 2016⁵ (Ministry of Internal Affairs of the Russian Federation, 2017). In Azerbaijan’s neighboring state, Georgia, crimes against property were the most recorded offense in 2017 (14.385, or 3.58 per 1.000) (Ministry of Internal Affairs of the Republic of Georgia (2018). Thus, Azerbaijan, at least on paper, is a notably safer country than some of its neighbors.

Objective of the study

As discussed above, the perceived probability of being a victim and fear of crime are two distinct concepts. There is a burgeoning literature on the relationship between home security systems and fear of crime, but the perceived probability of victimization has been paid lesser attention. Our understanding of the relationship between home security systems and the perceived probability of being a victim is rather limited. In an effort to fill this gap, this study attempts to discover whether there is any association between the level of home security systems and the perceived probability of being a victim of property crime. Furthermore, despite their value, all the studies exploring the impact of security systems on either fear of crime or perceived risk of victimization differ from the current paper. First, the current paper is based on a nationally representative sample covering households with different background characteristics. In fact, because of their focus on a specific location, the research design of previous studies did not control for certain community-level variables that are likely to differ between settlements, such as urban–rural differences and varying neighborhood contexts. This factor is important given that several scholars have noted a greater perceived risk of victimization among urban dwellers than among rural dwellers (e.g. Hummelsheim et al., 2011; Markowitz et al., 2001). Similarly, as noted by Giblin et al. (2012), crime prevention (one of which involves dwelling security features) in rural communities can differ from that of urban areas. Secondly, the analysis by Hirschfield et al. (2004) focused on the effects on the sense of security of a single anti-burglary initiative related to alley-gating. Mawby and Simmonds (2003), on the other hand, conducted an evaluation of “homesafing” in one area only to ascertain the impact of target-hardening mechanisms on worry. The current study, however, takes into account several security features, and categorizes dwellings based on the number of security features in place.

Furthermore, as a by-product, our study may enrich our understanding of the effects of situational crime prevention measures. Although there has been extensive discussion of this question in relation to the impact of situational measures on crime rates (Knights & Pascoe, 2000; Montoya et al., 2016; Nilsson & Estrada, 2006; Piza et al., 2018), relatively limited thought has been given to understanding the impact of situational measures on the perceived risk of victimization. This gap is surprising in both theoretical and practical terms, given the growing security industry, the increased security measures taken for private property and the burgeoning literature on fear of crime.

Two research questions guide this paper:

1. What is the security level across the dwellings?

2. What is the relationship between home security systems and the perceived probability of being a property victim at home?

Methods

Sampling frame and technique

This analysis uses the dataset of the first nationwide victimization survey in Azerbaijan, which was conducted in between July and August 2020 by a local think-tank. As a sampling frame, dual frame (mobile and landline) containing the list of landline telephone numbers of every household and mobile phones number in 8 economic regions⁶ of Azerbaijan was employed in this survey.⁷ The official data from 2021 indicate that the figure for mobile phone number subscription per 100 individuals was 108 (in absolute numbers, 10.817.100), while the figure for landline telephones per 100 households was 64 (in absolute numbers, 1.377.200 households) (State Statistics Committee, 2022). From a representativeness point of view, the figures for mobile phone number subscriptions suggest that the majority is in possession of at least one mobile phone. Both lists, provided by the relevant authority, were said to be complete, but since the researchers did not have full control over the preparation of the lists, we acknowledge the potential exclusion of some residents, which reduces randomness. Nonetheless, to some extent, triangulating both mobile and landline call approaches was effective in reducing sampling bias (see Vicente et al., 2009). The target population of this study comprised the whole population over 18 years of age in Azerbaijan. Unfortunately, the data on population provided by the State Statistics Committee (2020) do not show the population size of citizens over 18 years of age. Instead, it provides data on 15–19 years and older, of which the total number is 7.810.400. This study uses the dataset gathered from a randomly collected nationwide sample (n = 1216). The results are interpreted at the confidence interval of 95% and margin of error set at 3%. No weighting was applied.

As this was a nationwide study, multistage cluster sampling was employed as a sampling technique. There are 8 economic regions covered in this survey, hence, 8 primary sampling units were chosen the first stage. Next, economic regions’ population size relative to the national population size was determined. This proportion was the reference point in determining sample size for each economic region. Subsequently, within each economic region, all urban residence and several rural clusters were randomly chosen to ensure the coverage of both urban and rural residents. In the following stage, dual frame for each economic region was used for cold-calling respondents. Within the sample frame, every 3rd number was called. Once household was contacted, interviewers asked to speak to the person associated with each number who was 18 years. As a further selection principle, “the last birthday” method was used. Response rates across the 8 economic regions that were sampled varied, and coverage and response rates respectively were as follows: Baku (33.2%, 85%), Absheron (7.28%, 95%), Ganja-Qazakh (11.81%, 94%), Shaki-Zakatala (5.61%, 98%), Lankaran (9.06%, 95%), Quba-Khacmaz (6%, 97%), Aran (20%, 88%), Mountainous Shirvan (3.44%, 92%), Upper Karabakh and Kalbajar (2.60%, 88%) and Nakhchivan (1.1%, 93%).

In order to determine the extent to which the respondents in our sample differed from the country population, the sample was compared to the 2019 census data (State Statistics Committee, 2020). Unfortunately, only gender and age indicators can be compared, since they are measured by the census data too. As of the 2019 census data, 49.9% of the Azerbaijani population consisted of men, while females made up the remaining 51.1%. People aged between 15–24 and 25–34 made up 13% and 18.1% of the population respectively. Percentages for other age groups are as follows; 35–44 (14.7%), 45–54 (12.8%), 55–64 (11.4%) and 65 over (7.1%). Note that the official age categories provided by the census and the current sample do not match exactly, thus, this comparison between datasets should be read with care. For instance, while the national census includes individuals aged less than 18, this survey did not do so. Nonetheless, considering sample size and the small margin of difference between census and sample characteristics, findings can be regarded as generalizable to the whole Azerbaijani population.

Data collection method and survey administration

As a data collection method in the survey, computer-assisted telephone interviewing (CATI) was employed. Logistically, given the budget limitations, CATI was significantly more affordable than a face-to-face survey method. It also enabled the coverage of a relatively large number of respondents within a short period of time. A team of 7 surveyors carried out calls. Pollsters were given Lenovo tablets to register the answers of respondents. All the responses were anonymous and confidential. Employing the “cold-calling” method, the survey was conducted on diverse days of the week in order to cover all segments of the population as much as possible. The calls were made between 10:00–21:00. The response rate for the survey was 90% (1350 respondents were contacted).

Survey instrument

The questionnaire was general in nature and created by a local think tank to measure victimization, as well as attitudes to numerous crime and justice-related issues. The original questionnaire was in Azerbaijani, and subsequently back-translated to English.

Variables

Different conceptualizations and measurements of fear of crime have been employed by different scholars. Whereas fear of crime is a physiological and emotional reaction (Henson & Reyns, 2015), perceived risk is considered to be more of a cognitive assessment of one’s likelihood of being victimized (Chadee & Ditton, 2003; Warr, 2000). Franklin et al. (2008) further add that while perceived risk represents the cognitive dimension of fear, worry refers to the affective dimension.⁸ However, several scholars argue that fear and risk are related because fear is both influenced by and influences judgments of risk simultaneously (Franklin et al., 2008; Jackson, 2006; Matthews, 1990).

The current analysis adopts one’s personal judgment of their risk of victimization, since this was the approach employed by the original survey. There was one dependent variable in this analysis, which was operationalized as follows; “How probable is it for you to be victimized through property crime (i.e. burglary/theft and etc.)”. The question had the following response set: none at all (1), low (2) and high (3).

Key independent variable

Home security systems

Operationalization of home security systems was realized through a single question: “Which security measures listed below are in place in the house or apartment flat you live in?” A respondent was allowed to choose multiple response items, which were as follows: CCTV (1), formal surveillance system either inside the home or the site I reside (e.g. security guards) (2), alarm system (3), multipoint locking doors (4) and window security bars (5). Respondents with no measure in place were coded as (0). A respondent was allowed to choose multiple measures.

However, an important point must be clarified here. Since around 70% of the respondents had no security system at all,⁹ it meant that analyzing the predictive capacity of every security item would not allow us to draw conclusions on their role in shaping PPV. Moreover, the authors were not interested in ascertaining the effect of each item separately. Therefore, in the ordinal logistic regression analysis, home security systems were amalgamated into one single variable. Thus, a new code (security level) was created to differentiate households depending on a count of how many (none, one, or two or more) measures are in place. Those choosing no measure were categorized (0), while those having one and multiple measures were categorized (1) and (2), respectively.

Control variables

All control variables incorporated into the analysis are based on previous research. In line with the guidelines of the survey, respondents were allowed to choose one response only among the crime types presented, since only the most recent victimization in the last 3 years was asked in the original survey. For the purposes of the present analysis, only the responses to “property crime (i.e. burglary/theft)” were explored. Thus, this question was coded as a binary variable, where property crime victims were coded (1), while non-victims or victims of other crimes were coded (0). Respondents’ trust in the police was measured through the question which had the following: completely distrust (1), mostly distrust (2), mostly trust (3) and completely trust (4). To operationalize one’s perceived level of disorder, a respondent was asked “Which problems do you observe in your community?”. The response items presented cover both crime-related and non-crime-related neighborhood concerns. The items presented are as follows; (1) widespread poverty and unemployment, (2) lack of social order (noise, binge drinking, lack of formal control etc.), (3) untidy and littered streets, (4) widespread drug use, (5) high crime rate (i.e. theft, looting, violence) and (6) poor social bonding/ties. During the analysis phase, respondents were categorized into 3 three categories based on the number of items chosen. Those choosing 1 and 2 or more problems were coded as (1) and (2), while the remaining as (0).

Regarding socio-demographics, both age and household income level were continuous variables in the original questionnaire, but the responses for both were categorized for the purposes of the analysis. Age was converted to a categorical variable (1 = 18–25 years; 2 = 26–35 years; 3 = 36–45 years; 4 = 46–55 years; 5 = 56–65 years; 6 = 66 years and over), while responses on income were categorized into 5 brackets (1= ₼1–250 AZN; 2= ₼251–500; 3= ₼501–750; 4= ₼751–1000 and ₼1001 + and over). To note, 250 AZN was the national minimum wage in the country as of 2020. Gender was a dummy-coded variable (male-0, female = 1). Unfortunately, the original survey had not asked of educational level, which the authors acknowledge as a limitation. City, county¹⁰ and rural dwellers were coded (1), (2) and (3) respectively. Next, respondents stated the type of dwelling they reside in—apartment flat (1) and detached or semi-attached flat with garden (2). The subsequent question asked whether a respondent lives (1) alone or (2) with someone else.

Analysis

Initially, descriptive statistics (mean score, standard deviation and percentages) for all the variables considered are presented. The next step involves chi-square test of independence between all variables and perceived probability of victimization. Additionally, the chi-square test between the level of security in a dwelling and demographic trait was run as well to explore the distribution of security systems (or lack thereof) across the sample. The third analytical procedure involves ordinal regression. As the dependent variables in both models were ordinal, ordinal logistic regression was chosen. Here, two models are run. While Model 1 includes PPV as a dependent variable, Model 2 explores the predictors of a home security level. While identifying whether it is individuals’ perceived probability of victimization level that affects the dwelling’s security features, or the other way round, is not possible in this cross-sectional data, one more set of models was run, where the level of home security was chosen as a dependent variable.

Results

Descriptive statistics

In general, the probability of being victimized was assessed by the overwhelming majority of the sample as either non-existent or low (M=.36, SD=.597). A significant proportion of the sample does not have any security measure at home (M =  0.40 SD = 0.668). Those residing in a detached or semi-detached house dominated the sample (M = 1.74, SD=.435). Full results are presented in Table 1. Next, bivariate correlations between each independent variable and the dependent variable were explored. Table 2 presents the results of bivariate correlations. Victimization in the last 3 years, household income level, trust in the police, perceived neighborhood disorder level, dwelling security level, age and residential location were correlates of PPV, and all of them had significance levels below 0.005.

Table 1. Descriptive statistics of the variables (n = 1216).

Variables	N	Min	Max	Mean	Med	St.D.
PPV	1216	0	2	0.36	.000	0.597
Security level	1216	0	2	0.40	1.00	0.668
Gender	1216	1	2	1.46	1.00	.499
Age	1216	18	85	41.29	39.50	14.161
Victimization	1216	0	1	.047	.000	.213
Trust in the police	1216	1	4	3.16	3.00	1.032
Dwelling type	1216	1	4	1.74	2.00	.435
Living arrangement	1216	1	2	1.04	1.00	.189
Location	1216	1	3	1.92	2.00	.884
Household income level	1216	200	15.000	567.41	425	767.90
Perceived neighborhood disorder level	1216	0	3	1.00	1.00	.982

Table 2. Bivariate correlations between key research variables.

	1	2	3	4	5	6	7	8	9	10	11
(1)	–	0.020	−0.012	−0.006	−0.033	−0.056	.093**	−0.055	−0.018	0.017	−0.040
(2)	0.020	–	−0.033	−0.008	0.040	−0.003	0.034	−.099**	0.043	−0.006	0.020
(3)	−0.012	−0.033	–	−.177**	−.061*	.374**	−0.016	−.181**	−.107**	−0.050	.057*
(4)	−0.006	−0.008	−.177**	–	.116**	−.259**	−.115**	.166**	.175**	.063*	0.014
(5)	−0.033	0.040	−.061*	.116**	–	−0.019	−.119**	0.049	.308**	.125**	−.221**
(6)	−0.056	−0.003	.374**	−.259**	−0.019	–	−.064*	−.109**	−0.045	−0.020	0.013
(7)	.093**	0.034	−0.016	−.115**	−.119**	−.064*	–	−.060*	−.192**	−.073*	0.041
(8)	−0.055	−.099**	−.181**	.166**	0.049	−.109**	−.060*	–	.086**	0.056	0.020
(9)	−0.018	0.043	−.107**	.175**	.308**	−0.045	−.192**	.086**	–	.212**	−.157**
(10)	0.017	−0.006	−0.050	.063*	.125**	−0.020	−.073*	0.056	.212**	–	−.141**
(11)	−0.040	0.020	.057*	0.014	−.221**	0.013	0.041	0.020	−.157**	−.141**	–

(1) Living arrangement; (2) Gender; (3) Location; (4) Security level; (5) Perceived neighborhood disorder level; (6) Dwelling type; (7) Age; (8) Household income level; (9) PPV; (10) Victimization; (11) Trust in the police.

*p < 0.05, **p < 0.01.

Ordinal logistic regression

Prior to estimating the regression models, a series of diagnostics were performed to ensure that certain assumptions are met. The variables are ordered, and one or more of the independent variables are either categorical or ordinal (there is no continuous variable). Next, the output indicates that the parallel assumption holds since the probability (p values) for all variables are greater than alpha = 0.05. Test of parallel lines results for both models are greater than 0.05. Therefore, the proportional odds assumption is not violated, and the model is a valid model for this dataset.

City (b = 0.392, p < 0.05) and county (b = 0.571, p < 0.005) dwellers are more likely to hold a higher PPV level. Those aged 18–25 (b = 1,265, p < 0.05), 26–35 (b = 1,069, p < 0.05) and 36–45 (b = 0.842, p < 0.05) have greater PPV levels. A lower level of security measures at home is associated with a lower perceived probability of victimization level for acquisitive crimes (b= −0.683, p < 0.001). Respondents citing more than 3 problems in their area tended to feel more likelihood of victimization than those citing 2 or fewer problems. Lower trust in the law enforcement authorities and the court was associated with a higher perceived likelihood of victimization (b = 0.554, p < 0.005) Previous property crime victimization is linked to a greater perceived probability of victimization (b = 1,319, p < 0.001). Overall, the model explains 21.5% of the variation in PPV.

Relative to other significant variables, the significance level between PPV and security level in this model was somewhat marginal (b= −0.583, p = 0.027). City (b= −0.583, p < 0.05). and county dwellers (b= .647, p < 0.01) were more likely to feature one or more security features. Apartment dwellers were more likely to reside in a dwelling with one or more security features (b = 1,036 p < 0.001). Lower income was associated with a lower home security level (b= −0.801, p < 0.001). Overall, the model explains 16.5% of the variation in home security level. The full results are presented in Tables 3 and 4.

Table 3. Ordinal regression model for PPV.

Variables and coding	Coefficients (b-values)	Standard error	Significance (p values)
Dependent variable
Perceived probability of victimization = 0	−1.086	0.669	0.104
Perceived probability of victimization = 1	1.102	0.667	0.099
Key independent variable
Security level = 0	−0.683	0.209	0.001
Security level = 1	−0.356	0.232	0.125
Security level = 2	0
Control variables
Victimization = 0	−1.319	0.270	0.000
Victimization = 1	0
Perceived neighborhood disorder level = 0	−1.821	0.218	.000
Perceived neighborhood disorder level = 1	−1.133	0.205	.000
Perceived neighborhood disorder level = 2	−1.184	0.234	.000
Perceived neighborhood disorder level = 3	0
Age = 1	1.265	0.404	0.002
Age = 2	1.069	0.389	0.006
Age = 3	0.842	0.391	0.031
Age = 4	0.479	0.413	0.246
Age = 5	0.289	0.417	0.488
Age = 6
Dwelling type = 1	−0.058	0.165	0.724
Dwelling type = 2	0.000
Income level = 1	−0.267	0.259	0.304
Income level = 2	−0.238	0.258	0.357
Income level = 3	0.034	0.293	0.908
Income level = 4	−0.191	0.311	0.539
Income level = 5	0
Location = 1	0.392	0.170	0.021
Location = 2	0.571	0.188	0.002
Location = 3	0
Living arrangement = 1	0.108	0.379	0.776
Living arrangement = 2	0
Trust in the police = 1	0.554	0.197	0.005
Trust in the police = 2	0.528	0.301	0.080
Trust in the police = 3	0.272	0.153	0.077
Trust in the police = 4	0
Victimization = 1	−1.319	0.270	0.000
Victimization = 2	0
Gender = 1	0.023	0.034	.050
Gender = 2	0.000
Valid cases	1216
Final model fit (Chi-square)a	224.797	P < 0.000
Nagelkerke pseudo R-square	.215
Test of parallel lines (Chi-square)a	1502.668 (29.402)	P = 0.167

Table 4. Ordinal regression model for home security level.

Variables and coding	Coefficients (b-values)	Standard error	Significance (p values)
Dependent variable
Security level = 0	0.047	0.628	0.940
Security level = 1	1.499	0.630	0.017
Key independent variable
Perceived probability of victimization = 0	−0.583	0.263	0.027
Perceived probability of victimization = 1	−0.062	0.267	0.817
Perceived probability of victimization = 2	0
Control variables
Victimization = 0	−0.142	0.289	0.624
Victimization = 1	0
Perceived neighborhood disorder level = 0	−0.462	0.227	0.042
Perceived neighborhood disorder level = 1	−0.185	0.218	0.396
Perceived neighborhood disorder level = 2	−0.294	0.248	0.236
Perceived neighborhood disorder level = 3	0
Age = 1	0.440	0.344	0.201
Age = 2	0.259	0.323	0.423
Age = 3	−0.004	0.326	0.991
Age = 4	−0.166	0.347	0.632
Age = 5	−0.431	0.351	0.220
Age = 6
Dwelling type = 1	1.036	0.150	0.000
Dwelling type = 2	0.000
Income level = 1	−0.801	0.244	0.001
Income level = 2	−0.787	0.243	0.001
Income level = 3	−0.387	0.276	0.161
Income level = 4	−0.271	0.288	0.347
Income level = 5	0
Location = 1	0.457	0.174	0.009
Location = 2	0.632	0.192	0.001
Location = 3	0
Living arrangement = 1	0.049	0.360	0.892
Living arrangement = 2	0
Trust in the police = 1	−0.359	0.211	0.088
Trust in the police = 2	−0.454	0.332	0.171
Trust in the police = 3	−0.147	0.150	0.327
Trust in the police = 4	0
Victimization = 1	−0.142	0.289	0.624
Victimization = 2	0
Gender = 1	−0.035	0.133	0.790
Gender = 2	0.000
Valid cases	1216
Final model fit (Chi-square)	171,367	P < 0.000
Nagelkerke pseudo R-square	,165
Test of parallel lines (Chi-square)	1598.754 (31,488)	P = 0.111

Discussion

What is the relationship between security measures and PPV? The current findings suggest that the households with a lower level of the perceived risk of being a property crime victim live at home with no or fewer safety measures at home. One potential explanation is that lower risk translates to a lower need for safety measures, which has been reported previously. For instance, Yuan and McNeeley (2016) found a positive association between individuals’ likelihood to engage in target-hardening defensive behavior and their fear of being a victim of crime. Our results, however, contrast with those reported by Vilalta (2012) in Mexico, who found a minimal to the non-existent effect of home security measures on the fear of crime. The present results differ from those reported by Tillyer et al. (2011) in school settings as well, which found irrelevance of several access-control and target-hardening strategies in reducing students’ perceived risk of victimization, Further regression analysis (Model 2) suggests that there are factors more significant than PPV that predicts the number of security measures. Placing the significant variables in rank order based on the size of the beta weights (from Model 1 in Table 3) shows that perceived neighborhood disorder level is the relatively most important explanatory variable. The implication is that despite the explanatory power of PPV, sociodemographic variables play a more influential role in shaping the security level of a dwelling. From a policymaking point of view, the implication is that the authorities can reduce a sense of risk through interventions on neighborhood level, such as tackling street drug use and crime rates. In light of these findings, it is important to acknowledge that we do not claim that home security measures are ineffective in reducing PPV, since the present research design and data do not allow for the identification of such causality. Nonetheless, while the cross-sectional design does not allow for causality, the significance of these variables in both models somewhat confirms the statement put forward by Roth (2018, p. 710)—“precautionary behavior may be both a product and predictor of the perceived risk and fear of crime”

Our findings regarding other correlates of the perceived probability of being a victim were consistent with those presented in previous studies, though some studies cited below explored the fear of crime, not PPV as an outcome variable. Victims of acquisitive crimes in the last 3 years had higher PPV, which overlaps with what was reported elsewhere (Schafer et al., 2006; Oh & Kim, 2009). Those citing more neighborhood-level issues in their community were more likely to have higher PPV, which can be partly explained by “broken windows” theory (Wilson and Kelling, 1982), which argues that signs of neighborhood decay (e.g. vacant run-down housing, graffiti-covered walls, etc.) foster fear of crime because it denotes community indifference to delinquent and other deviant behaviors. Indeed, while Wyant (2008) reports that residents’ perception of the incidence of crime, rather than the actual incidence of reported crime, can shape their levels of fear, other scholars have found a strong effect of socio-economic problems and other uncivilized behavior on the fear of crime (Alper & Chappell, 2012; Doran & Lees, 2005; Moore & Shepherd, 2007). Age emerged as a significant correlate, which conforms to previous findings exploring the fear of crime or perceived risk correlates (Chadee & Ditton, 2003; Vilalta, 2011a; Yun et al., 2010). In line with several previous studies (Ferguson & Mindel, 2007; Vilalta, 2011a), previously having been a victim of property crime was another correlate, which can be explained by the traumatic experience suffered by these respondents. Respondents living in cities and counties are likely to have a higher perceived probability of being a victim of crime than their rural counterparts, which overlaps with studies on fear of crime (e.g. Dammert & Malone, 2003; Kristjansson, 2007).

Regarding the distribution of security systems, PPV does not predict the presence or absence of security features at home, which differs from previous studies exploring the association between fear of crime and various home protective behaviors (Cook & Fox, 2011; Giblin et al. 2012; Yuan & McNeeley 2016). Similarly, being a property victim did not predict the presence or absence of security features at home, which contrasts with some studies (May et al., 2009; Yuan & McNeeley 2016), but overlaps with a recent study by Roth (2018). Thus, our findings suggest that the presence or absence of security features at home may not necessarily reflect residents’ PPV, and not all property crime victims install security systems. This, and the findings discussed so far have implications for the security industry—residents do not necessarily install security features purely because of their victimization, and those features are not prerequisite for residents to have low PPV.

Instead, four sociodemographic correlates predict dwellings’ security level. Residents of buildings are likely to have one or multiple security items. Indeed, this variable has the highest beta value among all significant variables. Interestingly, living in an apartment or detached/semi-detached house did not correlate with PPV, something different from many previous findings (e.g. Rollwagen, 2016), but predicts dwellings’ security level. One can explain it by the fact that residents in buildings typically do not choose whether to install security systems or not. Rather, it is the building’s owner who installs security items such as controlled access, CCTV and security guards. Nonetheless, it is an issue that needs in-depth exploration. Households with higher income are more likely to possess security features at home, which could be the result of greater affordability of security systems, and a greater perceived attractiveness of the household as a target. Previous research has found a similar relationship between income and a decision to equip one’s dwelling with security features (Nilsson & Estrada, 2006; Reid et al., 1998). Finally, urban, and county-based respondents are more likely to live in dwellings with one or multiple security features. While there is no definitive explanation to this pattern, one potential explanation is related to the income variable. That is, as previously noted by some international organizations (e.g. Asian Development Bank, 2014), there is a serious gap between urban and rural areas in Azerbaijan regarding socioeconomic welfare. Thus, it is possible that some households in rural areas do not have security features simply because they cannot afford them. Alternatively, as shown by our data, respondents in villages have lower PPV, hence, may see lesser or no need for security features.

Limitations and future research

Our findings contribute to the literature on the fear of crime and PPV by ascertaining the home security level’s explanatory power in understanding the level of PPV. However, this paper has several limitations. The authors acknowledge that the present analysis falls short in identifying the causality of the relation between the perceived probability of being a victim and home security systems. While our study identified the statistical significance of home security measures—both as a dependent and independent variable—in a relationship with PPV, the existing data do not allow us to ascertain whether it is the resident’s perceived probability of being a victim of a crime that affects the dwelling’s security features, or the other way round. An experimental or longitudinal research design could resolve this issue.

Sample-wise, the vast majority of the sample lived with someone else; a sample consisting of more people living alone may have produced a different result because residents living alone are likely to have no one to rely on as a physical guardian. Similarly, we acknowledge the fact that less than one-third of the sample had at least one security measure in place, this may undermine the generalizability of the findings. Thus, a future sample covering more households with home security systems could produce a more representative picture of the relationship between the perceived probability of being a victim and the level of safety systems. The mean score of 1.37 on a scale of 3.00 means that the dominant portion of the sample mostly felt safe, which prevents us from capturing sufficient variation in PPV. The implication is that to enhance the generalizability of the results regarding the relationship between the perceived probability of being a victim of crime and the level of safety systems, further studies should be conducted either among a sample where a larger proportion of those feel unsafe. Similarly, as this study is limited to one geography, its external validity is quite limited. It would be interesting to see how home security systems shape fear of crime in societies suffering from widespread crime, or crime hot spots in any given society.

The data collection method employed in this study may have affected the validity of some respondents’ answers to certain questions. Writing on Azerbaijan, Sadigov and Guliyev (2018, p. 87) warn of the “reliability of survey data on regime-related attitudes obtained by public polling in societies with limited freedoms.” Since the telephone survey method was used, respondents’ contact numbers were known. Given that perceived victimization risk, trust in law enforcement, and perceived neighborhood disorder issues may be connected to government policy, some might have felt reluctant to express their genuine opinions. A future survey employing a face-to-face data collection method would likely minimize the impact of this problem.

One may also criticize the wording of the dependent variable. As noted, the dependent variable asked the perceived probability of property crime victimization, but it did not distinguish types of property crimes. In a similar way, asking about people’s fears of crime variations between daytime and nighttime (see e.g. Rollwagen, 2016; Vilalta, 2012) would have allowed us to understand the potential effects of time on the variable under question. From a questionnaire point of view, it would have proven useful had the original question set asked respondents about (a) the status of their dwelling, i.e. public housing, private ownership, or rental, and (b) housing context, since these factors may facilitate or inhibit informal surveillance capabilities, as well as PPV. Similarly, to capture much finer data on neighborhood-level variables, future studies on this subject could adopt the approach employed by Yuan and McNeeley (2016) who measured social integration and neighborhood incivilities through scalar items.

Finally, as the literature suggests, fear of crime and risk perception can be related within a person. The current study, however, does not control for fear of crime, which has at least one implication. Not controlling for this variable does not allow us to ascertain whether the level of home security systems correlates with the level of fear of crime and perceived probability of property crime victimization differently.

In short, while this study offers some insight into the association between home security systems and the perceived probability of property crime victimization, considerably more research is required to better understand this relationship, both in Azerbaijan and abroad.

Ethical statement

This work did not apply for ethical approval, since there is no regulatory body in Azerbaijan in this area.

Disclosure statement

The authors state that they have no conflict of interest.

Notes

1 There may be at least two reasons for this different rate – the current political regime is less democratic than others (The Economist, 2019; Transparency International, 2019), and there may be different crime measurement and recording practices.

2 Gallup’s Law and Order Index uses four questions to gauge people’s sense of personal security and their personal experiences with crime and law enforcement. It is therefore not designed specifically to gauge fear of crime or perceived risk of being a victim of crime.

3 As of 2020. the national population size was 10.000.600

4 An important note of caution has to be made here for international readers. Unfortunately, State Statistics Committee does not show crime rate for every offence. Rather, it lists the most 10 widespread offences only.

5 The latest official crime data available was for 2016.

6 An economic region in Azerbaijan is an administrative division of a country. Each economic region contains at least one large city and multiple smaller cities or districts

7 Azerbaijan has 10 economic regions in total, but 2 economic regions were excluded due to political reasons, since these excluded economic regions were under the occupation of Armenian armed forces.

8 While some studies have employed a binary yes/no forced choice, others allowed respondents to rank their fear on a continuum.

9 While 70% did not have security system, the figures for those in possession of at least one item are as follows: CCTV (7.4%), multi-point locking doors (16.7%), window security bars (13.8%), formal surveillance system either inside home or the site I reside (4.5%) and alarm system (3.7%)

10 County is a type of settlement situated between city and village, in terms of both population and area size.