Analysis of Violence in Rio de Janeiro

Ah, Rio de Janeiro, what a marvelous city! The beautiful hills that surround it, the huge Christ the Redeemer statue, not to mention the beaches and museums spread all over the city. But not everything is rosy, right?

Due to "n" delicate historical factors, which I won't get into, violence in the city has reached alarming levels, as we see in every newspaper and TV show out there. However, Agência Brasil, a federal government agency, reports a reduction in crime, such as intentional homicides, which fell by 21% in 2019. The agency relies on data published by the State of Rio de Janeiro, more specifically by the Instituto de Segurança Pública (ISP, the state's Public Security Institute), the body responsible for recording incidents reported to Civil Police stations.

So I decided to explore the Institute's databases. Knowing full well that these issues are not black and white, I ran an exploratory analysis of some of the variables to extract a few insights, as you'll see. The full project is in my repository, which contains the notebook I used.

The data used here was obtained from the Instituto de Segurança Pública of the State of Rio de Janeiro. It has 61 variables and 6,992 records, extracted on June 11, 2020, showing the categories of crimes recorded in the municipalities where they occurred, as a time series starting in 2014.

#Importar os dados                                                                         
df="http://www.ispdados.rj.gov.br/Arquivos/BaseMunicipioMensal.csv"
data=pd.read_csv(df,sep=";",encoding='latin-1')

As we can see in the first five entries below:

However, for this work I had to apply a few filters. First I set the fmun_cod column, which holds the identifier of each municipality in the state, as the index. That way, I filtered only the data for the city of Rio de Janeiro.

#Localizar ID do Rio de Janeiro
data_rj=data.loc[3304557]

Then I started to understand how the dataset behaves, first making a copy of the original data. I began by dropping the mes_ano, mes, regiao and fmun columns, then filtered out all 2020 data, since at the time of the project that year was still under way. Finally, I set the year column as the index and aggregated the values of each variable by year.

#Agrupar pelo ano
data_time = data_time.groupby(['ano']).sum()

As a result, we end up with the following data, with 6 entries and 57 variables.

With the dataset organized, I started asking it questions, beginning with: "What is the average number of vehicle robberies per year?" Vehicle robbery is a common occurrence in state capitals; national statistics indicate that a car is robbed every minute.

Its descriptive statistics show an average of 19,612.50 vehicle robberies per year. In addition, the maximum was 25,894 and the minimum 13,725, which are fairly high values.

count        6.000000
mean     19612.500000
std       4888.905061
min      13725.000000
25%      16225.000000
50%      18890.000000
75%      23427.000000
max      25894.000000
Name: roubo_veiculo, dtype: float64

When we compare the values for each year, we get the following chart, which shows fairly similar values:

Continuing with the questions: what about thefts? Their values turned out to be much lower than robberies and fairly linear over time. This data has an average of 7,025.66 per year, a maximum of 7,515 and a minimum of 6,710.

With this, I was able to calculate the percentage of recovered vehicles using the following equation:

#Porcentagem de veículos recuperados
rec_vec = (data_time.recuperacao_veiculos)/(data_time.furto_veiculos+data_time.roubo_veiculo)*100

For the percentage of recovered vehicles, I noticed fairly similar values, with an average of 53.71% per year, a maximum of 59.09% and a minimum of 49.28%. In addition, the percentage of recovered vehicles over the entire 6-year period was 54.29%.

count     6.000000
mean     53.713210
std       4.214933
min      49.284958
25%      50.816075
50%      52.290771
75%      57.316638
max      59.096325
dtype: float64

Next, I asked: "What is the temporal distribution of robbery-homicides (latrocínios)?" Latrocínio refers to robbery followed by death, an important indicator of urban violence. There were 62 cases per year on average, with a maximum of 93 and a minimum of 34.

count     6.000000
mean     62.666667
std      23.997222
min      34.000000
25%      45.750000
50%      58.500000
75%      82.500000
max      93.000000
Name: latrocinio, dtype: float64

Robbery-homicides in the city of Rio peaked in 2017, followed by a drop in 2018 and 2019.

I also looked at robberies on public transport (buses, vans, trains, subways and others), which was a common practice among criminals a few years ago. However, the data shows that this practice is on the rise, with a very high average of 7,266 cases, a maximum of 9,775 and a minimum of 4,412.

count       6.000000
mean     7266.333333
std      2367.376748
min      4412.000000
25%      5115.250000
50%      7898.500000
75%      9044.500000
max      9775.000000
Name: roubo_em_coletivo, dtype: float64

It's worth stressing that this is a preliminary study that shows the potential of using Python, basic descriptive statistics and data to extract insights and support decision-making. The project was an exercise from the Data Science na Prática course.

It showed that the average number of vehicle robberies has decreased, while thefts have remained practically unchanged. When we compare these with the number of recovered vehicles, the overall average across all years is 54%, which means that a little over 50% of stolen vehicles are returned to their owners.

I also ran a quick analysis of the number of robbery-homicides over the period, which showed a drastic drop in 2018 and 2019. Finally, I analyzed robberies on public transport, which have been growing strongly since 2016.