Crime patterns in Los Angeles
Understanding the crime landscape in LA, with data.
The city of Los Angeles has made available its crime reports database. I’m interested in learning how to analyse crime data in order to derive policy recommendations. In the following excerscise, I try to use all this data to answer three questions:
- What are most recurring crimes in LA?
- Is the police department’s response a match for the criminal activity?
- Does the LA police department seems to follow a strategy based on data to mitigate crime?
Let’s start with question 1: What are most recurring crimes in LA?
There are eight types of crimes that account for more than 96% of all the criminal activity: larceny, other assaults (which include battery and domestic violence), burglary, miscellaneous violations (related to violations of parole and other court orders), aggravated assault, vehicle theft, robbery and crime related to sex.
To answer question 1 (What are most recurring crimes in LA?): The most recurrent crime in LA is larceny, with more than 28k cases every six months. Along with assaults, and burglaries, they account for more than 60% of the criminal activity. As we can see, these crimes share in common that they show an even distribution during the weekend days, and an increase toward the weekend. Activities related to stealing property, like burglary and larceny, reach a peak on Fridays; assaults, aggravated and other types, occur more often on Saturdays and Sundays.
Moving on to question 2: Is the police department’s response a match for the criminal activity?
The ratio of arrests to crimes reported is 0.39. So, about for every two reported crimes, there is approximately one arrest. Initially, it can be assumed that multiple reports could be linked to one case, so this number might not be necessary labeled as ‘low’. Let’s turn to see if what are the most types of arrest.
If we analyse the arrest activity during the week, we get a different picture than the crime reports. Most of the arrests happen in the miscellaneous other violations, and narcotic drug laws categories. Actually, they happen mostly during working weekdays, peaking on Wednesdays.
There is one category that does not appear in the top crimes reported, that account for a significant chunk of arrests: driving under influence. Going a bit deeper, we can see that this type of arrests spike on weekends, consistent with the days of the weeks most likely to have higher alcohol consumption.
The first type of arrest that is consistent with the crime reports is related to aggravated assault, even in the fact that the busiest days are Saturdays and Sundays. For the category “other assaults”, while it is reassuring to see it in the top, at least from a graphic point of view, it does not mirror the pattern observed in the crime reports, busier on weekends.
At first glance, we can see that the police focuses on activities that seem to be led by the judiciary system (e.g. warrants of arrest), and a certain degree of flagrancy (attacks). While larceny appears high in the types of arrest, it comes only in sixth place. On Friday, the day of the week with the highest crime rate, when larceny hits its peak, is also one of the days when the least number of arrests for this type of crime are made.
Aggravated assault and other miscellaneous violations have the largest ratio of arrest to crime reports, both higher than 0.5 (or 1 arrest for every 2 reports). The crimes with the highest ratio, are also associated with violence or flagrancy. Larceny, the most frequent crimes, has one of the lowest ratios, with less than 1 arrest for every 10 reports. This is consistent with analysis by day of week.
To answer question 2, the police seems to consistently address crimes that appear to be urgent or violent. However, for two the top three most frequent types of crimes, the ratio is lower than 1 arrest for every 10 reports. The police seems to be more of a match for urgent type of crimes, than for crimes related to theft.
Last question: 3. Does the LA police department seems to follow a strategy based on data to mitigate crime?
At this point, we introduce a new dimension: area where crimes are reported. Across neighbourhoods, larceny remains the most reported crime, except in 77th street, where larceny comes in second place. The second most frequently reported crimes are either other assaults and burglaries.
When we contrast the previous results, with the police activity (i.e. arrests), it could be argued that there is a disconnection. The main focus for arrests are miscellaneous other violations in almost all of the neighbourhoods, with a significant emphasis in Central. While this neighbourhood seems to have most of the police department’s attention, with 10% of the arrests performed there, it only comes in third place in crime reports, after 77th St, and Southwest. The second most frequent type of arrest if for both types of assault (aggravated, and other). The only places where arrests for larceny comes first are Topanga, Van Nuys and Devonshire, all neighbourhoods with relatively lower levels of crimes reported.
So far, it seems that not all of the police arrests is driven by the crime reports. There seems to be a trade off between violent/court order and going for theft cases. To test this hypothesis, we should try to understand the variation in the crime activity. If we observe that certain types of crimes are correlated to certain areas, day of the week or other types of crime, we could asses the polices’ strategy to target efforts.
Since we have data for crime reports about the time when it happened, some characteristics about the victim, and the location, we will implement an algorithm to reduce dimensionality and describe the factors that explain for most of the variance in crime. For this we use a principal component analysis (PCA), and analyse the first eigenvectors to see how the variables correlate with each other. Thus, we could be able to see if crimes go hand in hand with other types of crime or with certain areas.
Usually, we could expect that the first couple of components would explain a big chunk of the variance, allowing to target on specific factors. From the previous plot, we can see that the variance is spread out across components, and even the first principal component explains less than 5% of the variance.
At first glance, we can see that criminal activity is widely scattered across the city, and across time.
When we map the weighs of the the first component, we observe that both types of assault are positively correlated. Also positively correlated with assaults are being a black victim, female, hispanic and being in the neighbourhoods 77th street and Southeast. Incidentally, the prevalence of assault is negatively correlated to larceny and car theft, and having a victim of white ethnicity. Here I present the top ten weights for the first component:
larceny -0.473301
car_theft -0.316645
victim_white -0.206821
Southeast 0.118019
77th Street 0.129195
victim_black 0.226855
victim_female 0.248909
victim_hispanic 0.309210
aggravated_assault 0.325383
other_assault 0.460622In the second component (not shown), having a victim of white ethnicity, is positively correlated with burglary, having a female victim, and happening on a neighbourhood of ‘middle’ criminal activity. These factor, are negatively correlated with car theft, larceny, having a victim of hispanic descent and happening on a neighbourhood with higher criminality. To complete the picture, for the third component, in general, having a victim of black ethnicity is positively correlated with both types of assault, happening in the neighbourhoods with higher criminal activity, and negatively correlated with having a victim of hispanic ethnicity, being in a neighbourhood with lower crime and miscellaneous violations.
To answer question 3 it seems that the criminal activity is so scattered, that the most violent crimes are not correlated with the least violent but more prevalent offences, like larceny. The data seems to suggest that unless there is a differentiated strategy, fighting one type of crime, does not help to curve or prevent the other type.
Conclusion
Crime in LA seems to have a very heterogeneous pattern, making it difficult to maximize the impact of the police department’s action. It seems that fighting the most violent types of crimes (assaults) is uncorrelated with the more prevalent crimes, like larceny and burglary.
It also seems that the department concentrates efforts in those efforts that seem “close to home”. Namely, the most frequent type of arrests have to do with court orders (violations of parole, etc), and flagrant acts (like assaults), paying more attention to the Central neighbourhood, rather than the one where most of the crime reports happen: 77th street.
While the data analysed only considers arrests and not any other actions performed by the police, it seems that the department could benefit from a more systematic approach to its own data. Prospectively, this data could shed some light toward strategies focusing on prevention.
About the data:
- LA Police department crimes reports from 2018–12–25, until 2019–06–22, which contains a total of 103,181 crime reports.
- LA Police department arrests reports from 2018–12–25, until 2019–06–22. This table contains 40,884 cases. The LA police identifies in its database 27 types of charge groups, and 835 types of charges.
- All the analysis and files I used and produced are in github.
