key: cord-0907286-v0dxiyhi authors: Hou, Miaomiao; Zeng, Zhaolong; Hu, Xiaofeng; Hu, Jinming title: Investigating the Impact of the COVID-19 Pandemic on Crime Incidents Number in Different Cities date: 2022-02-17 journal: Journal of Safety Science and Resilience DOI: 10.1016/j.jnlssr.2021.10.008 sha: c4d30bab2abd20de85d0f65af2f571644ad5a81c doc_id: 907286 cord_uid: v0dxiyhi The COVID-19 pandemic is strongly affecting many aspects of human life and society around the world. To investigate whether this pandemic also influences crime, the differences of crime incidents numbers before and during the pandemic in four large cities (namely Washington DC, Chicago, New York City and Los Angeles) are investigated. Moreover, the Granger causal relationships between crime incidents numbers and new cases of COVID-19 are also examined. Based on that, new cases of COVID-19 with significant Granger causal correlations are used to improve the crime prediction performance. The results show that crime is generally impacted by the COVID-19 pandemic, but it varies in different cities and with different crime types. Most types of crimes have seen fewer incidents numbers during the pandemic than before. Several Granger causal correlations are found between the COVID-19 cases and crime incidents in these cities. More specifically, crime incidents numbers of theft in DC, Chicago and New York City, fraud in DC and Los Angeles, assault in Chicago and New York City, and robbery in Los Angeles and New York City, are significantly Granger caused by the new case of COVID-19. These results may be partially explained by the Routine Activity theory and Opportunity theory that people may prefer to stay at home to avoid being infected with COVID-19 during the pandemic, giving fewer chances for crimes. In addition, involving new cases of COVID-19 as a variable can slightly improve the performance of crime prediction in terms of some specific types of crime. This study is expected to obtain deeper insights to the relationships between the pandemic and crime in different cities, and to provide new attempts for crime prediction during the pandemic. The COVID-19 pandemic is one of the most serious global public health events in recent years. The onset and spread of COVID-19 have affected nearly every continent. People's daily lives and the whole society have been drastically influenced around the world [1] [2] [3] . For example, in many cities, traffic is completely restricted [4, 5] ; non-essential businesses have closed for a very long time; travel became more and more difficult; and social gatherings are limited [6, 7] . Moreover, the COVID-19 pandemic is a huge challenge to education activities [8, 9] , many courses are moved online. At the same time, unemployment among many groups of workers increased sharply [10, 11] . What's more serious is that the pandemic has led to a dramatic loss of human life, economic losses and social disruption worldwide, presenting an unprecedented challenge to public health, food systems, and public safety [12] . This also raises attention to other questions related to our lives and security. Does the COVID-19 pandemic have an impact on crime? If so, is this impact strong or weak? If the COVID-19 pandemic has a strong impact on crime, will the pandemic be a factor for analyzing and predicting crimes? These questions motivated this study. During the COVID-19 pandemic, a few studies have investigated the impact of the pandemic on crime in different regions. For instance, Shayegh and Malpede [13] identified an overall drop by about 40% across crime types in San Francisco and Oakland from March 16, 2020 to March 28, 2020 . Campedelli, Aziani and Favarin [14] conducted Bayesian structural time-series and focused on nine crime categories, identifying that overall crime has significantly decreased in Los Angeles, as well as robbery, shoplifting, theft, and battery. Felson, Jiang and Xu [15] examined burglary in Detroit during three periods which are related to government suggestions, their findings indicated an overall 32% decline in burglary, with the most substantial change in the third period. De la Miyar, Hoehn-Velasco and Silverio-Murillo [16] used an event study for the intertemporal variation across the 16 districts' eight common crimes in Mexico City for 2019 and 2020, and indicated a decline in conventional crime during the COVID-19 pandemic, while organized crime remains steady. Ashby [17] found that burglary only declined in Austin, Los Angeles, Memphis, and Scan Francisco, serious assaults declined in Austin, Los Angeles, and Louisville, but not other cities. In these previous studies, investigation of impact on fraud was not reported. And what's the difference of the impacts on Chinese cities was seldom studied. Furthermore, in terms of the time series of the daily crime incidents and COVID-19 cases, are there significant correlations in different cities? This is still an open question. As is known to all, crimes are affected by many factors, such as economic variables [18] [19] [20] [21] , spatial and temporal autocorrelation factors [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] , environmental conditions [32] [33] [34] [35] [36] [37] , and current politics [38, 39] . These variables are often used to predict crimes [22, 26, 40] , providing support for crime risk prevention and control. Thus, another noteworthy question is that whether COVID-19 pandemic could be considered as a new factor to predict crime? Previous studies gave few ideas about that. In this paper, we first investigate the differences of four common crime incidents numbers (theft, fraud, assault, robbery and burglary) in four large cities (namely Washington DC, Chicago, New York City and Los Angeles) before and during the pandemic. Then the Granger causal relationships between crime incidents numbers and new cases of COVID-19 are studied. Finally, based on the results of Granger causality test between crime and COVID-19 pandemic, new cases of COVID-19 with significant Granger causal correlations are conducted to crime prediction to examine whether the new cases can improve the prediction performance of the daily crime incidents numbers. The paper is organized as follows: In "Introduction", the research background is introduced. "Materials and Methods" describes the data sets used in this study including the new cases of COVID-19 and crime incidents number in different cities, and focuses on the theory and steps of the Granger causality test and crime prediction. The results are presented and discussed in "Results and discussion". Finally, "Conclusion" draws a conclusion and points to the future research. Four large cities, namely Washington DC, Chicago, New York City and Los Angeles are selected as the research areas. These cities are typical large cities in the United States with adequate crime and COVID-19 data, and they have similar Economic, cultural and social background. Thus, it is reasonable to compare the impacts of COVID-19 cases (as well as people's activities influenced by them) on crime patterns among the above four cities. As for crime types, theft, fraud, assault, robbery and burglary are all the most common kinds of crimes through the world, specifically, in DC before the pandemic. Here, M is the mean value of the crime incidents number and SD indicates the standard deviation. In addition, some crime incidents numbers during the pandemic are much fewer than those before the pandemic in these cities, such as theft. To know whether the COVID-19 pandemic influence crimes, Granger causality test is applied [45] . First, time series stationarity are examined with the Augmented Dickey-Fuller (ADF) unit root test to avoid spurious regression [46] . If the calculated ADF statistic is lower than 1% and the p values of the significance level is lower than 0.05, the null hypothesis that assumes the presence of unit root is rejected, inferring that the time series is stationary. In contrast, if the null hypothesis is not rejected, the time series should be non-stationary. In this study, the first-order difference is applied to non-stationary time series to make all the time series stationary. Then, the Granger causality test is performed, and the optimal lags of the Granger causality test are obtained by the vector autoregressive models through the minimum Akaike information criterion (AIC) value [47] . LSTM is an improved multilayer perceptron network based on Recurrent Neural Network (RNN) which is widely used for time series prediction [48] . LSTM adds a memory unit to each hidden layer neural unit to realize controllable memory information in time series. When the time series data is transferred between the units of the hidden layer, it will pass through the input gate, forget gate, output gate and other interactive controllable gates to control the memory of previous data and current data, and the degree of forgetting, so that the neural network has a long-term memory function. In this way, LSTM effectively overcomes shortcomings such as the traditional RNN gradient disappearance, defects in effectively retaining long-term memory information [49] . In this study, new cases of COVID-19 with significant Granger causal correlations are applied in the LSTM models to improve the crime prediction performance. The time series are divided as shown in Table 2 . The test set is the last two weeks of the time series (14 days). The rest of the time series is the train set. The statistical law of stationary time series data changes little over time, and can usually be used for time series prediction. Therefore, stationarity of the time series is examined by ADF test and can apply the first-order difference to make all the time series stationary firstly. Some features including "month", "weekend", "holiday", "weekday_avg", "weekend_avg" and "month_avg" are extracted in this study (see Table 3 ). Next, the number of lagging observations is set to one. In other words, the crime incidents numbers at the previous moment are used to predict the crime incidents numbers at the current moment. Finally, all the time series are normalized for LSTM model training. [31] , and percentage root mean square error (PRMSE) [50] , which are defined as follows: where O t represents the observed value and P t represents the predicted value. n is the total number of predicted days, and the value of n in this study is 14. To examine the impact of COVID-19 on crime incidents numbers in large cities, the differences of crime incidents numbers before and during the COVID-19 pandemic are investigated. Fig. 3 shows the distributions of daily theft incidents numbers before and during the pandemic in four large cities of the U.S (Washington DC, Chicago, New York City and Los Angeles). As shown in Fig. 3 , the theft incidents numbers during the pandemic are much fewer than those before the pandemic in all the four cities. The stories of fraud, assault and robbery are quite like that of theft which means that all the four selected cities in the US witness significant decreases of crime incidents number (as shown in Fig. A1, Fig. A2 and Fig. A3, respectively) . An exception is burglary in New York City, since obvious increase of incidents number is witnessed in Fig. A4 , while in DC, Chicago and Los Angeles the trend is decreasing. As reported by the New York Post (see https://nypost.com/2020/11/14/new-stats-reveal-massive-nyc-exodus-amid-coronavirus-crime/), more than 300,000 New Yorkers moved away from the city during the pandemic, so large numbers of unoccupied houses and departments may provide many chances for burglaries. To answer the question whether COVID-19 pandemic influences crimes, Granger causality test is conducted. First, stationarity of the time series is examined and ensured to avoid spurious regression. Then, the optimal lags are selected by vector autoregressive models and the results are shown in Table 4 . After that, the Granger causality test between crime incidents numbers and new confirmed cases of COVID-19 in different cities is implemented, and the results are also shown in Table 4 . As shown in Table 4 , several Granger causal relationships are found in these US cities. For example, new confirmed cases of COVID-19 Granger cause the theft incidents numbers in DC since the p value is lower than 0.05, which means that the relationship is significant. In the US cities, both the number of crime incidents and the new confirmed cases of COVID-19 changed considerably during the periods studied in this paper. So, it is relatively easier to study their statistical laws. Based Table A1 , and the parameters of LSTM models are shown in Table A2 . In order to quantitatively evaluate and compare the prediction results, the indices RMSE and PRMSE are calculated and their values are shown in Table 5 . Indicated by them, models conducting COVID-19 cases performs slightly better than those without the cases. This demonstrates that involving new cases of COVID-19 as a variable can improve the performance of crime prediction in terms of some specific types of crime. ARIMA and NAR based prediction model for time series analysis of COVID-19 cases in India Projecting the Criticality Transmission in India Using GIS and Machine Learning Methods Forecasting of COVID-19: spread with dynamic transmission rate Analysis of mobility trends during the COVID-19 coronavirus pandemic: Exploring the impacts on global aviation and travel in selected cities COVID-19 impact on urban mobility The impact of travel and timing in eliminating COVID-19 Impacts of COVID-19: A research agenda to support people in their fight Education and the COVID-19 pandemic The impact of COVID-19 on changes in community mobility and variation in transport modes The unemployment cost of COVID-19: How high and how long? Deciphering the impact of COVID-19 pandemic on food security, agriculture, and livelihoods: A review of the evidence from developing countries Staying home saves lives, really! Exploring the immediate effects of COVID-19 containment policies on crime: an empirical analysis of the short-term aftermath in Los Angeles Routine activity effects of the Covid-19 pandemic on burglary in Detroit Druglords don't stay at home: COVID-19 pandemic and crime patterns in Mexico City Initial evidence on the relationship between the coronavirus pandemic and crime in the United States The impact of economic conditions on robbery and property crime: The role of consumer sentiment Modelling and predicting recorded property crime trends in England and Wales-a retrospective Revisiting property crime and economic conditions: An exploratory study to identify predictive indicators beyond unemployment rates Socio-economic, built environment, and mobility conditions associated with crime: a study of multiple cities Crime event prediction with dynamic features Crime prediction through urban metrics and statistical learning Crime risk analysis through big data algorithm with urban metrics Discovering spatial interaction patterns of near repeat crime by spatial association rules mining Factors influencing temporal patterns in crime in a large American city: A predictive analytics perspective The prediction of police calls for service: The influence of weather and temporal variables on rape and domestic violence Risk Prediction of Theft Crimes in Urban Communities: An Integrated Model of LSTM and ST-GCN Seasonal variation in violent victimization: Opportunity and the annual rhythm of the school calendar Spatial correlations, clustering and percolation-like transitions in homicide crimes Study on Time Series Prediction of Theft Crime Based on LSTM Network The association of ambient temperature and violent crime Crime, weather, and climate change The effect of temperature on crime Impact of climate variability and change on crime rates in Tangshan, China Impacts of climate variations on crime rates in Weather and crime The geography of crime: A political critique Policy tolerance of economic crime? An empirical analysis of the effect of counterfeiting on Italian trade Analyzing the impact of foursquare and streetlight data with human demographics on future crime prediction Impact of COVID-19 pandemic on injury prevalence and pattern in the Region: a multicenter study by the American College of Surgeons Committee on Trauma Air quality changes in New York City during the COVID-19 pandemic Investigating causal relations by econometric models and cross-spectral methods Likelihood ratio statistics for autoregressive time series with a unit root A new look at the statistical model identification Long short-term memory Long short-term memory neural network for traffic speed prediction using remote microwave sensor data Modeling and sensitivity analysis of transport and deposition of radionuclides from the Fukushima Dai-ichi accident A global analysis of the impact of COVID-19 stay-at-home restrictions on crime ☐The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: