Insecurity or Perception of Insecurity? Urban Crime and Dissatisfaction with Life: Evidence from the Case of Bogotá

My paper follows this practice and will use the three concepts as synonyms.

For instance, in Bogotá in 2012 policy makers argued that in the city there was a reduction on humicides due to policies such as the interdiction of wearing guns, however there is not evidence of its impact on welfare.

Despite having information for the years 2003 and 2007 (years in which the quality of life survey was also performed), it is impossible to analyze several years simultaneously because: 1) the samples are not comparable and, 2) the life satisfaction module within each survey is located in different places. This is important since, as it is showed by Kahneman and Kruger (2004), the reports of the life satisfaction is highly dependent of the questions made before at the survey. Thus, the use of the multipurpose survey is justified not only by being the most recent survey but also because it has the greatest coverage and representativeness.

CAI or Centro de Atención Inmediata, for its abbreviation in Spanish, is a small police station that watches a specific zone of the city. It allows the police to have a faster and more effective respond.

This definition is defined in this way because the question distinguishes between positive (very good and good) and negative (so-so and bad) aspects of the satisfaction due to the way in which it was performed. Thus, this measure is capturing the life dissatisfaction.

Although the depend variable is between 0 and 1, the model is estimated by ordinary least squares (OLS). According to Angrist and Pischke (2008) no linear models like probit are highly sensible to the specification of the model. Moreover, marginal effects, that are necessary to interpret the size of the effect of the independent variables, are very close to those reported by an OLS estimation. The OLS is the best linear approximation of the data.

The crime rates were taken for the year 2010 because the survey was made in the first months of 2011.

In the last group it is excluded the households whose moving decision was motivated for security reasons, since this group can be sensible to the security changes in the new neighborhood.

This estimator can present bias because of two-way causality, since the channel (both the perception of insecurity and the victimization) are measured at the household level. This is in part because the happy people tend to act in a some way and view the world in a different way, which makes probable that they feel less insecure and have some behavior correlated with the victimization. However if one of the channels were not important in the relationship between criminality and welfare, this could be used as an instrument since this is a guarantee of the exclusion restriction. The crime rates would constitute an exogenous variation of the channel that is not correlated on other form with welfare.

There could be other mechanisms that are not being taken into account and that could mediate the relationship. In this cases the sum of these effects is captured by coefficient Φ₃ in model 5.

This spatial patterns are included in the estimation through fixed effects of localidad and allowing cluster by CAI. This controls for correlation of the households that live in the same CAI jurisdiction.

It is also important to address the point of Diener (1984, 2000). This kind of measure might have noise by the factors such as person’s mood who answered the question or the acceptable social standards in the moment of answering. All these consideration can influence the answer.

The correlation coefficient of the satisfaction with the income is 0.27 and with the stratum is 0.29.

It is asked if a household member has been victim of theft, murder, extortion or kidnapping.

In the three years, there is a positive correlation between welfare and the socioeconomic stratum (Table A2), and between welfare and the levels of education (Table A3), while at the same time, the correlation of these variables (education and economic stratum) with crime is negative. This shows that both the socioeconomic stratum and the education of the household’s head are important variables in the analysis of life satisfaction, since both variables not only have an impact in these results but also mediate in the relationship of the households’ perceptions and crime.

At least 6% of the households are classified in the middle-high and high stratum (5 and 6 respectively).

While the coverage of land-line telephone is about 70%, the internet coverage is just 43%.

Just 12% of households have an undergraduate education level.

All these variables and other household’s characteristics are included as controls (in the vector X_hkj,t) in the regressions. The descriptive statistics of all the controls are reported in the Table A4.

This section uses crime in terms of 1000 inhabitants. The main reason to do this modification is that the effects estimated will have less zeros after the comma, causing the figures shown on the tables to be bigger. However, this does not affect the core interpretation of the results.

The correlation inside the zone of the self-reported welfare is of almost 7%.

Angrist and Pischke (2008) show that these kind of errors are consistent when there is a big amount of groups. This condition is fulfilled in this case because in Bogotá in 2011 there were 141 CAIs.

With the inclusion of this kind of error it is taken into account the variation at CAI level, the same that would be taken into account a multilevel model with the advantage that this model can be interpreted in a rigorous causal sense since all the assumptions are not violated.

It could be explained by the fact that while theft affects the possessions of the households, murder affects the life and it has more traumatic effects, so it could have a bigger weight in the definition of household’s welfare.

This amount of variables are used to control by all the possible characteristics that can be related with welfare and the place of dwelling, and since the welfare is a subjective concept it can be argued that it is related with almost every characteristics of the household. This large list of controls does not produce a big problem of multicollinearity. Table A7 shows that the R² for the structural estimation are not huge to think the presence of this problem. Moreover, although the mean variance inflation factor is not tiny this factor remains relatively constant when the controls are added.

All the regressions summarized in the table include the whole set of control, equivalent to the column 6 in Table 3.

The estimations of the direct and indirect effects are calculated based on a probit model.

In this calculation were used the standardized coefficients (Ender 2010; Kenny 2008) because these variables are not continuous. Since the distribution of these estimators is unknown, the estimation of the standard errors for the tests uses the bootstrap method.

In other words, the average treatment on treated effect (ATT) is bigger than the average treatment effect (ATE).

For instance, in the case of Bogotá, crime reduction has always been in the public opinion as a measure of mayors’ quality and public reception.

According with Green (2003) (Look also Chiburis, Das, and Lokshin 2011) this solution would be solving the bias problems of feeling the environment insecure.

This assumption is crucial in order to interpret the coefficients as causal relations, the effects of the insecurity perception on the life perceptions.