**4. Research properties and methodology**

This section discusses the data properties and research methodology used in this chapter.

#### **4.1 Data properties**

This chapter's data was sourced from the Global Insight and it ranged from 1996–2019. The chapter focused on Gauteng local municipalities namely: Ekurhuleni metropolitan; City of Johannesburg metropolitan; City of Tshwane metropolitan, Emfuleni, Lesedi, Midvaal, Merafong, Mogale city and Rand west local municipality. The data used for these municipalities include: crime, economic growth, unemployment rate, trade openness, education, poverty and income inequality. **Table 1** provides a summary of the variables used in this chapter.

An increase in goods and services is expected to reduce crime since economic growth increases job opportunities. This creates more income avenues. Conversely, economic growth exerts more crime when goods and services produced only benefits few people in a region. This leaves many unemployed youths with nothing to do and this breeds a seed of crime in a society. Employment is also expected to decrease crime activities according to the Becker's theory. The theory suggests that

*An Analysis of Economic Determinants and Crime in Selected Gauteng Local Municipalities DOI: http://dx.doi.org/10.5772/intechopen.96339*


#### **Table 1.**

*Summary of variables used.*

an increase in employment plays a critical role in reducing crime [14]. Poverty is expected to increase the crime rate in this study. The rationale is that poor households commit crime in order to improve their lives [13].

Trade openness is expected to influence crime positively or negatively. There are two reasons attached to this. First, if a country or a bloc uses many trade controls, majority of traders engage in illegal trade which increases crime in a society [11]. Second, a lack of economic or trade control increases the number of goods and services circulating in the society thereby decreasing crime activities in a society [11]. Income inequality increases the number of crime activities in a society. A huge income inequality frustrates households with low income and end up devoting to crime as a solution to their problems and this is in line with the strain theory of crime [10]. On the other hand, education is expected to reduce the number of crime activities in a society. Education increases some household earnings that leads to a higher opportunity cost of crime. Education also influence a household personality trait that makes one disciplined.

### **4.2 Descriptive statistics**

The descriptive statistics summarizes the data set used in this study. It serves two purposes. First, descriptive statistics illustrates the basic information of variables employed in the study. Second, descriptive statistics shows the potential relationship among the variables used in the study. Thus, **Table 2** gives a summary of the variables used in the study focusing more on the mean of variables. The descriptive statistics illustrate that Gauteng trades with other provinces and this is shown by an average of 52 percent. The results also show the dominance of inequality and crime in the local municipalities shown by average of 41 and 40 percent respectively. The results are also in line with the national statistics [25]. In addition, poverty rate and unemployment were also found to be dominant in the province exhibiting an average rate of 30 and 20 respectively.

#### **4.3 Panel unit root test**

Panel unit root is the first estimation technique used in the estimation of variables. Garidzirai et al. [26] stipulates that it used to determine whether the variables are stationary and determine the order of integration. In achieving these two aims. Lin, Levin and Chu and the Pesaran and Shin were used to determine the stationarity of variables and order of integration. These tests have a null hypothesis of unit root test.


**Table 2.**

*Descriptive statistics.*

Thus, a null hypothesis is rejected if the p- value is less than 10 percent. The results reveal that lngdp, lnpov, lntrad and lngini are stationary at levels. Therefore, the variables are integrated at level one 1(1). Variables such as lncrime and lnunm were found not to be stationary at levels, therefore, became stationary at first difference. The panel unit root test concludes that the variables are integrated at 1(0) and 1(1). Noteworthy is that panel unit root test prescribes the research methodology to use. For instance, if the variables are stationary at levels, a panel least square is estimated while if the variables are of different level; levels and first difference then a Panel Autoregressive Distributive Lag is deemed fit [26]. In the event that the variables are integrated at first difference, a Panel – VAR or Panel – VECM is appropriate [27]. Since the panel unit root tests indicate a combination of zero and one a Panel Autoregressive Distributive Lag under the Pooled Mean Group is deemed fit (**Table 3**).

#### **4.4 Methodology**

Since the variables were found to be integrated at levels and first difference a Pooled Mean Group (PMG) was deemed fit for this chapter. A PMG allows the researcher to estimate the long-run relationship without performing cointegration tests since this model is the new cointegration test [26, 28, 29]. Another advantage of using this model is that it gives robust and accurate parameters. In addition, the model eliminates the risk of using data with a unit root and it is appropriate for all samples [30]. Furthermore, a PMG allows a researcher to analyze both short-run and long-run relationship. Noteworthy is that, a Hausman test was employed to confirm whether the PMG is the appropriate and accurate estimator to use. The PMG model is illustrated in Eq. 1:

$$
\Delta \ln \text{crime}\_{i,t} = \bigotimes\_{i} \left( \ln \text{crime}\_{i-t} - \beta\_i X\_{i,t-j} \right) + \sum\_{j=1}^{p-1} \gamma^i \Delta \left( \ln \text{crime}\_{i,t-j} + \sum\_{j=0}^{q-1} \partial \Delta \left( X\_i \right)\_{t-j} + \mu\_i + \varepsilon\_t \right) \tag{1}
$$

Where lncrime is the dependent variable and X = poverty, gini, unemployment, economic growth, education and trade openness in the Gauteng provinces. The signs δ and γ represents the short-run coefficients of dependent and independent variables respectively while I is cross sections and t for time. β is long-run coefficients while u represents fixed effect and e = error term.

*An Analysis of Economic Determinants and Crime in Selected Gauteng Local Municipalities DOI: http://dx.doi.org/10.5772/intechopen.96339*


#### **Table 3.**

*Panel root results.*


#### **Table 4.**

*Cointegration test.*

To determine whether a long-run relationship exists, cointegration tests were employed using the Kao test. The test sets the null hypothesis on no cointegration implying there is no long-run relationship among the variables under study. The rule of thumb is to reject the null hypothesis if the variables are below 0.10 and conclude that the variables under study have a long-run relationship. The cointegration results are illustrated in **Table 4**.
