Methodology

View the most recent version.

Data sources
Definition of neighbourhoods
Sample size
Description of variables
Multilevel analysis
Odds ratio

Data sources

The data for this study come from two sources: the 2004 General Social Survey (GSS) and the 2001 Statistics Canada Census. In 2004, as part of the GSS program, Statistics Canada conducted a fourth survey cycle on victimization and public perceptions of crime and the justice system. The target population of the 2004 GSS included all people aged 15 and over, except full-time residents of the Yukon, Nunavut, and the Northwest Territories, and full-time residents of institutions. Data were collected each month from January to December 2004.

The 2001 Census provides the most recently available population and dwelling counts for Canada and also for smaller geographic units such as cities and areas within cities. The detailed socio-economic data used in this study are derived from the long form of the Census, which is based on a 20% sample of households. These data exclude the institutional population, which includes individuals living in hospitals, nursing homes, prisons and other institutions.

Definition of neighbourhoods

This study is concerned with the influence of neighbourhood environments on individual experiences of fear of crime. While there are numerous methods of defining the geographic boundaries that make up neighbourhoods, for the purposes of this study, Statistics Canada census tracts are used to approximate neighbourhoods. Census tracts are small, relatively stable geographic areas averaging about 2,500 to 8,000 residents. They are located in large urban centres with populations of 50,000 or more. Census tracts are reasonable representations of local perceptions of urban neighbourhoods since their boundaries are determined by committees of local specialists, for example, planners, health workers, and educators, in conjunction with Statistics Canada. An important consideration in determining census tract boundaries is that the residential population of the area is as homogeneous as possible with respect to socio-economic characteristics (Statistics Canada 2003, 249).

Sample size

The analyses presented in this study are based on respondents residing in large urban centres, with populations of 50,000 or more, in the 10 provinces. The analytic sample was composed of over 12,300 respondents representing about 15 million Canadians, living within approximately 3,900 census tracts.

Description of variables

Outcome variable

Fear of crime in the neighbourhood: Respondents were asked how safe they felt from crime while walking alone in their area after dark. Possible responses included very safe, reasonably safe, somewhat unsafe or very unsafe. A two-category outcome variable was created so that somewhat or very unsafe = 1, and very or reasonably safe = 0.

Individual-level variables

Sex of the respondent: A two-category variable where women = 1, and men = 0 (reference category).

Age: Four categories included 15 to 24, 25 to 44 (reference category), 45 to 64 and 65 years and older.

Total household income: Included an estimate of the total income, before deductions, of all household members from all sources during the past 12 months. Data were analyzed by income quartiles where the fourth (highest) income quartile was the reference category.

Education: Four categories included non-completion of secondary school, secondary school completion, some post-secondary school, and completion of a post-secondary degree or diploma (reference category).

Visible minority status: A two-category variable where visible minority = 1, and non-visible minority = 0 (reference category). The definition is based on the concept of "visible minority" in the Employment Equity Act applying to those who identified themselves as being non-Caucasian in race or non-white in colour. Under this definition non-visible minority included those who identified themselves as single origin White, single origin Aboriginal, multiple origins White/Latin American or White/Arab-West Asian.

Victimized in past year: A two-category variable where being the victim of one or more crimes in the past 12 months = 1, and not being a victim = 0 (reference category). Includes all forms of criminal victimization including cases where the spouse or ex-spouse was the offender.

Physical disorder a problem in the neighbourhood: Respondents were asked two questions about their neighbourhood physical environments: How much of a problem are (1) garbage and litter lying around? (2) vandalism, graffiti and other deliberate damage to property or vehicles? Possible responses included a very big problem, a fairly big problem, not a very big problem, or not a problem at all. A two-category variable was created such that 1 = either type of physical disorder was a fairly or very big problem, 0 = both types of physical disorder were not a very big problem or were no problem at all.

Social disorder a problem in the neighbourhood: Respondents were asked seven questions about their neighbourhood social environments: How much of a problem are (1) noisy neighbours or loud parties? (2) people hanging around on the streets? (3) people sleeping on the streets or in other public places? (4) people being attacked or harassed because of their skin colour, ethnic origin or religion? (5) people using or dealing drugs? (6) people being drunk or rowdy in public places? (7) prostitution? Possible responses included a very big problem, a fairly big problem, not a very big problem, or not a problem at all. A two-category variable was created such that 1 = any type of social disorder was a fairly or very big problem, 0 = all types of social disorder were not a very big problem or were no problem at all.

Crime is higher than other neighbourhoods: Respondents were asked if compared to other areas in Canada, they thought that their neighbourhood had a higher amount of crime, about the same or a lower amount of crime. A two-category variable was created where 1 = a high amount of crime, and 0 = about the same or a lower amount of crime.

Neighbourhood-level variables

Variables describing the neighbourhood context were based on Statistics Canada Census measures of the percentage of socio-economic, demographic and dwelling characteristics in the population. In the analyses, each Census variable was represented by a two-category variable split at the median across all neighbourhoods, indicating high or low levels of the characteristic.

Exploratory analyses using more than two categories for these neighbourhood variables showed no meaningful or statistically significant differences in fear of crime across other levels of the variables. Consequently, a median split was used for ease of interpretation in this study.

High proportion of low-income families: A two-category variable where 1 = neighbourhoods with a proportion of persons in low-income families that was above the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion below the median. Low income refers to families who spend 20% more of their disposable income than the average family on food, shelter and clothing. Statistics Canada's low-income cut-offs (LICOs) are income thresholds that vary according to family and community size. Although LICOs are often referred to as poverty lines, they have no official status as such.

High proportion of persons aged 65 years and older: A two-category variable where 1 = neighbourhoods with a proportion of older Canadians that was above the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion below the median.

High proportion of visible minorities: A two-category variable where 1 = neighbourhoods with a proportion of persons identifying themselves as visible minorities that was above the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion below the median. (See definition of visible minority person mentioned previously in this section).

High proportion of lone-parent families: A two-category variable where 1 = neighbourhoods with a proportion of lone-parent families that was above the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion below the median.  

Low proportion of longer-term residents: A two-category variable where 1 = neighbourhoods with a proportion of residents living at the same address 5 years earlier that was below the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion above the median.  

Low proportion of newer dwellings: A two-category variable where 1 = neighbourhoods with a proportion of dwellings built within the last 10 years that was below the median proportion for all neighbourhoods, and 0 = neighbourhoods with a proportion above the median.

Multilevel analysis

The aim of this study is to investigate whether at least part of the differences in fear of crime among people may be attributable to the areas in which they live. Neighbours may share similar socio-economic and demographic characteristics, resources and experiences. Consequently, it is reasonable to assume that residents of one neighbourhood may be more similar to each other with respect to their levels of fear when alone at night in the area than to residents of other neighbourhoods.

Statistically, it is necessary to use techniques that consider the possible dependence of individuals clustered in the same area. Conventional regression analysis techniques assume that individual observations are independent from one another, if this assumption is violated estimates of the regression coefficients can be biased and standard errors may be underestimated. Multilevel regression techniques make it possible to take into account the possible dependence of the outcome variable between people in the same neighbourhood (Raudenbush and Bryk 2002; Snijders and Bosker 1999).

In this study, a series of multilevel logistic regression models were estimated to investigate variation in the chances of experiencing fear of crime among individuals clustered within neighbourhoods. First, the 'empty' model (i.e., containing no explanatory variables) provided an estimate of the expected probability of fear of crime for someone with nationally average background characteristics, as well as an estimate of how much variation in fear of crime existed between neighbourhoods (see Intraclass correlation below). In the second stage of analysis, a model assessed whether neighbourhood variation in fear of crime was associated with individual characteristics. And in the final stage of analysis, two models assessed whether neighbourhood level factors were associated with the chances of experiencing fear over and above the influence of individual-level factors.

Intraclass correlation coefficients (ICC) were calculated for each model. The ICC indicates the proportion of the total variance in the outcome variable, fear of crime, that is explained by the neighbourhood-level and is equal to the variance between neighbourhoods divided by the sum of the between-neighbourhood variance and the individual-level variance (Raudenbush and Bryk 2002, 72). In multilevel logistic regression models the ICC is approximated as the between-neighbourhood variation divided by the sum of between-neighbourhood variance and π²/3 (Snijders and Bosker 1999).

Possible ICC values range from 0 to 1 where 0 would indicate that no respondents share common neighbourhood-level chances of reporting fear of crime, and 1 would indicate that 100% of the respondents in each neighbourhood share chances of reporting fear of crime. Thus, an ICC value less than 0.5 indicates that there is greater variability within neighbourhoods than between neighbourhoods, while a value greater than 0.5 indicates that there is greater variability between neighbourhoods than within them. An ICC value of 0 would indicate that single-level, rather than multilevel analysis, is justified.

Odds ratio

When an outcome variable for a regression model has two categories, for example, feeling unsafe in your area while alone after dark versus feeling safe, researchers are interested in determining the probability of the occurrence of that event under a particular set of circumstances, for example, having low income, being female, or having been victimized in the past year. In this case logistic regression is the most appropriate technique to use. An odds ratio is a statistic generated by a logistic regression and can be used to assess whether, other things being equal; people with specific characteristics are more or less likely to report experiencing fear of crime than those in another group, referred to as the reference category.

For example, if we consider the risk of experiencing fear for a woman in comparison to a man (the reference category), an odds ratio near 1.0 implies there is no difference in fear between the two groups; an odds ratio less than 1.0 implies those in the group being considered (i.e. women) are less likely to experience fear than those in the reference group (i.e. men) and an odds ratio greater than 1.0 implies those in the group being considered are more likely to experience fear than those in the reference category.