Forecasting Crime Rates in Uganda (2011-2022
Abstract
The research aimed to predict crime rates in Uganda accurately and reliably using monthly data from the Uganda Police Force from 2011 to 2022. It employed VAR modelling methods, rated against ARIMA, to determine the optimal approach for forecasting crime rates in the country.
The study found that the ARIMA (2,1,2) model with a drift term was the best ARIMA modelling choice for accurately fitting and forecasting crime rates in Uganda. Key coefficients: AR1 (0.50), AR2 (-0.15), MA (-1.48), and MA2 (0.52), indicated the influence on current crime rates based on past values and forecast errors. A drift term with a coefficient of -49.4 addressed any systemic trends or biases in the crime data. Statistical tests showed; no compelling evidence for significant high-order serial correlation, limited first-order serial correlation, and no significant heteroscedasticity in the model's residuals.
The VAR model revealed that, variables like Percentage of Cases Taken to Court, Percentage of Cases Under Inquiry, Crime in Rural, Crime in Urban, Male Juvenile, Female Adults, and Female Juvenile suspects exhibited negative effects on reported cases, though weren't statistically significant. Conversely, Number of Police Officers, Crime on Highways, and Male Adult suspects had a positive influence, though not statistically significantly. Percentage of Cases Not Detected (PCND) had a significant association with total reported cases. Specifically, with each increase in PCND, total cases reported increased by 113. This could mean that as more cases go undetected, it might encourage criminals, leading to more crimes. It could also reflect a loss of public confidence in the police force, potentially leading to increased lawlessness. Alternatively, it might suggest that as overall crime increases, the police force becomes overwhelmed, leading to more undetected cases.
The study revealed that, on average, there were 19,960.06 monthly reported cases, with a minimum of 10,002 and a maximum of 28,959. Furthermore, it highlighted a higher frequency of crimes in Rural areas (51.66%) compared to Urban areas (45.04%), while Highways had the lowest occurrence (3.29%). This could be due to Less police presence in rural areas, Poverty and lack of economic opportunities in rural regions, Possible underreporting of crimes in urban areas. Additionally, the research found that, on average, 63.94% of reported cases remained under investigation, 22.30694% progressed through the court system, and 13.74% went undetected.
The relative root mean squared error (RRMSE) of 0.99, which is less than one, indicated that the VAR model outperformed ARIMA in forecasting crime rates in Uganda. This conclusion is reinforced by other accuracy measures such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) which were all better in VAR than ARIMA.
The study recommends using the VAR model for predicting Uganda's crime rates and underscores the need to incorporate more socio-economic variables to improve the model's explanatory capabilities, providing a deeper understanding of the factors impacting crime rates in Uganda.
The study recommended the following policy actions: Allocation of more resources to rural areas, enhancement of investigation efficiency with modern techniques, prioritization of crime detection and prevention through training and data-driven methods, implementation of data-driven policing, promotion of public awareness and reporting improvement of police form one to capture all the relevant factors that influence crime like Unemployment, Poverty, Education levels, Population, Urbanization, Income inequality, Family structure, Substance abuse and Migration.