COVID19 Impact on Child Maltreatment: Evidence from Abuse and Neglect Investigations in Texas

. COVID-19 has brought challenges to the society in various aspects, as one of the most vulnerable members of society, children’s lives have also been significantly affected by it. This study aims to address the child maltreatment impacted by the COVID-19 pandemic. Leveraging county level data sets from Texas Department of Family and Protective Services and United States Census Bureau, XGBoost method and fixed effect model was used to investigate the most important economic, demographic, and social factors. It is found that population of 16-year-old or over, the rental vacancy rate, the population of 16-year-old or over that commutes to work by walking, and population of 16-year-old or over that works in agriculture, forestry, fishing and hunting, and mining industry is positively associated with total number of child maltreatment cases. On the other hand, the total population in labor force, the female population not in labor force, the average of public cash assistance (in dollars), the average time (in minutes) commuting to work, and the population whose household contains 2 units/rooms are negatively associated with total number of child maltreatment cases. Also, a single-difference model was implemented to identify that the COVID-19 pandemic was associated with 7.6% increase in total number of child maltreatment cases.


Introduction
The outbreak of the COVID-19 pandemic has brought unprecedented challenges to societies in North America, affecting every aspect of human life.Among the various concerning issues, child maltreatment emerged as a pressing concern during this turbulent time.The confinement measures, school closures, and economic hardships brought on by the pandemic have disrupted the protective networks for many children, exacerbating their vulnerability to maltreatment.As families faced heightened stress levels and uncertainties, risk factors for child maltreatment surged, prompting a closer examination of this critical issue.This study explores the effects of COVID-19 on child mal-treatment cases through an extensive analysis of existing literature, statistical modeling, and well as causality inferences.We will examine how changes in demographic, economic, and social factors contributed to child maltreatment cases.Additionally, the pandemic impact on number of child maltreatment cases was examined.The significance of this research lies in its potential to inform policymakers, healthcare professionals, and community organizations about the urgent need for attention in child protection and maltreatment prevention during stressed time.By understanding the complex factors at play, we can develop predictive strategies to identify children under risk of maltreatment and provide protective environments for children during times of crisis.

Literature Review
Child maltreatment is a complex issue that is influenced by a variety of social, economic, and demographic factors.Research has identified several key factors that can contribute to child mal-treatment, including poverty, a high number of children per adult resident, population turnover, and a high concentration of female-headed families [1].During times of stress, such as the COVID-19 pandemic, these factors can interact in ways that lead to an even greater impact on child maltreatment.The COVID-19 pandemic has presented many challenges that had a significant impact on the mental health of children and adolescents [2].Social isolation and loneliness have been shown to have negative effects on the mental health of young people during the pandemic [3].Parenting has also been affected by the pandemic, with many parents reporting increased levels of stress [4].This increased stress can increase the risk of family violence [5].Healthcare workers have also been affected by the pandemic.Many were facing mental health challenges because of their work during this time [6].Children and adolescents w also experiencing mental health challenges because of the pandemic [7], and school closures have had a negative impact on their mental health [8].In addition to these studies, there are other reports and research articles that have documented the impact of the COVID-19 pandemic on child maltreatment cases.For example, a report by the Center for Innovation in Child Maltreatment Policy, Research and Training at Washington Univer-sity in St. Louis discussed how the pandemic has impacted child abuse and neglect.The report highlighted how the loss or reduction of job and financial stability, health, housing, and childcare can increase parental stress, which is strongly associated with increases in child abuse and neglect [9].A systematic review found that compared to the period before the pandemic, some studies identified an increase in child maltreatment while others found a decrease or no difference [10].In conclusion, child maltreatment is influenced by many factors and is a complex issue.The COVID-19 pandemic has presented new challenges and has had a significant impact on children and families.This study will investigate the social, economic, and demographic factors that affect child maltreatment and will analyse the impact of the pandemic on child maltreatment while controlling for these factors.

Data
This study leveraged the county level economic and social factor data sets from United States Cen-sus Bureau and CPI 3.8 Abuse/Neglect Investigations -Alleged and Confirmed Victims data from Texas Department of Family and Protective Services.The United States Census Bureau provides county level demographic, economic, and social related factors for Texas.The Texas Department of Family and Protective Services provides data on victims in investigations that have been concluded by county from FY2013 to FY2022.We use the total number of cases (including confirmed and unconfirmed cases) as the dependent variable.A confirmed case in a completed investigation is a child who has been identified as a victim in at least one allegation with a disposition of reason to believe.An unconfirmed case in a completed investigation is a child who was alleged to be a victim in at least one allegation with a disposition of unable to complete, unable to determine, or ruled out.The two data sets were merged by year and county.Therefore, we utilized county-level panel data with 221 counties over the time series from 2013 to 2021 in Texas.
After extracting the data, linear interpolation or extrapolation is used to handle missing values.To avoid the interference of outliers, the variables are subjected to a 1% trimming process, which removes extreme values beyond the 1% percentile and 99% percentile.The natural logarithm transformation is applied to all variables (including dependent variable) for re-scaling and lowering standard deviation.

Model
This study first leveraged the XGBoost method to select the importance of variables, that is, to score the importance of variables.The XGBoost model is an algorithm based on gradient-boosted decision trees, which is known for significantly improving the scalability and generalization of vari-able selection.In each training iteration, the algorithm creates a new decision tree model in the direction of the gradient of the previous model's loss function and prunes it during the tree con-struction.On the other hand, the XGBoost algorithm adopts a distributed forward additive model, directly fitting the residuals of the upper-level trees after each iteration of generating weak learners, thus, enhancing variable selection accuracy through ranking and weighted quantiles.
After preliminary feature selection, a Hausman test is conducted to determine whether to use a fixed effects model or a random effects model.The results indicate a P-value lower than 0.01, suggesting that a fixed effects model should be used instead of a random effects model.The final model is expressed in the following format: where c and t refers to county and year respectively, Total c,t refers to the total number of confirmed and unconfirmed child maltreatment cases in county c and year t, x c,t refers to the explanatory variable, θ c represents the individual fixed effects (county-specific fixed effects), and ϵ c,t represents the error term.
Lastly, we would use single-difference model to study the COVID-19 pandemic impact on the total number of confirmed and unconfirmed child maltreatment cases.We used Covid c,t as a time dummy variable.As an indicator for pandemic, it equals 1 if the data point is after the year of 2020 (including 2020) and equals 0 otherwise.The model is expressed in the following mathematical format: log  ,    ,   , ,   , where α 1 represents the main causal relationship that is being attempted to identify.A positive and statistically significant α 1 indicates a significant positive effect of the COVID-19 pandemic on the dependent variable.Control i,c,t is a control variable, which is statistically significant explanatory variable from the fixed effect model constructed previously.θ c represents the individual fixed effects (county-specific fixed effects), and ϵ c,t represents the error term.

Results and Empirical Analysis
Based on the feature importance ranking generated from XGBoost model, we selected control variables and choose the most relevant ones for predicting the dependent variable to fit into the fixed effect model.Table 1 summarizes the top 20 variables with the highest importance score.From table 2, it can be observed that the coefficients of variables log(population16yearsandover), log(walked), log(rentalvacancyrate),and log(agricultureforestryfishingetc) are significantly positive, which indicates that these indicators have a significant positive impact on the dependent variable.On the other hand, the coefficients of log(inlaborforce), log(notinlaborforce), log(meancashpublicassistance), log(meantraveltimetowork), and log(units_2) are significantly negative, which indicates that the abovementioned indicators have a significant negative impact on the dependent variable.Also, we calculated some descriptive statistics for all variables, which are summarized in table 4. The output for single-difference model is summarized in table 5.It can be seen that the estimated coefficient of Covid c,t is significantly positive.

Discussion and Conclusion
Based on the fixed effect model results, we can see that holding all other variables constant, when the population of 16-year-old or over, the rental vacancy rate, the population of 16-year-old or over that commutes to work by walking, and population of 16-year-old or over that works in agriculture, forestry, fishing and hunting, and mining industry increases by 1%, the total number of child maltreatment cases is estimated to increase by 2.18%, 0.04%, and 0.15% on average respectively.Also, holding all other variables constant, when the total population in labour force, the female population not in labour force, the average of public cash assistance (in dollars), the average time (in minutes) commuting to work, and the population whose household contains 2 units/rooms increases by 1%, the total number of child maltreatment cases is estimated to decrease by 1.15%, 0.75%, 0.01%, 0.25%, and 0.03% on average respectively.When controlling for those variables that have a significant impact on the dependent variable, based on the single-difference model output, we found that the COVID-19 pandemic was associated with 7.6% increase in total number of children maltreatment cases.There are literature suggesting that the COVID-19 pandemic has had a significant impact on child maltreatment cases.For example, a study found that child maltreatment often increases after natural disasters and economic downturns like the Great Recession [11].
According to a report by the International Labour Organization (ILO), the economic crisis caused by the COVID-19 pandemic is expected to contribute to global unemployment of more than 200 million people next year, with women and youth workers worst-hit [12].At the same time, it is found that unemployment, food insecurity, and financial stress can increase parents' risk of child maltreatment [13].Additionally, stress is known to increase the risk of child abuse and neglect [14].These studies provide strong evidence that the COVID-19 pandemic has had a significant impact on child maltreatment cases, with increased levels of stress among parents being a major contributing factor.Moreover, it is found that the COVID-19 pandemic has been referred to as a "shadow pandemic" of domestic violence [15].Another study found that parenting stress and the risk of child maltreatment increased during the COVID-19 pandemic [16].Nevertheless, a paper discussing the potential impact of containment measures on child wellbeing, found that some pandemic mandates, such as movement restrictions and the closure of schools, services, and businesses increased the likelihood of violence against children and women, while cutting them off from social and professional support systems [17].Finally, the patterns of parent stress across COVID-19 were found to increased [18].These studies provide further evidence of the significant impact of the COVID-19 pandemic on child maltreatment cases and highlight the need for policymakers and practitioners to take these findings into account when developing interventions and support programs to address this issue.
Given the urgent need for attention of child maltreatment during stressed time such as COVID-19 pandemic, there are some suggestions to resolve this issue.For example, in Canada, one recommendation is to strengthen the surveillance and research framework for child maltreatment by adding questions about maltreatment to national health and victimization surveys, such as the upcoming Canadian Health Survey on Children and Youth [1].This would provide valuable information on trends over time and among different subgroups, allowing for the development and evaluation of interventions to prevent child maltreatment.Similarly, based on our study, it is suggested that child protection agencies or social organizations should focus more on parents' mental health as well as the family financial situations during stressed time, which are closely related with child maltreat-ment.Lastly, it is suggested that the child protection agencies in North America, such as Child Aid Society (CAS) in Canada, Federal Agencies Associated with Preventing Child Abuse and Neglect in the United States, could cooperate more with medical institutions to aid in early identification and intervene when abuse and neglect are suspected.For example, the Canadian Paediatric Society has released a statement on "Medical neglect: Working with children, youth, and families" that outlines how health care providers can promote the best interests, well-being, and safety of children or youth at risk for medical neglect [19].The statement highlights the importance of identifying and addressing any barriers that may prevent a parent from ensuring their child receives the care they need.Similarly, in the United States, the Child Welfare Information Gateway provides information on "Medical Responses to Child Abuse and Neglect," which discusses the crucial role that medical professionals play in assessing and documenting injuries that may be due to abuse or neglect [20].
To sum up, this study leveraged a data set from county level in Texas with rich number of social, economic, and demographic variables with a wide range of time including the stressed time as comparison.It is found that the COVID-19 pandemic has a significant impact on the child maltreatment cases, which led to increasing number of cases.It is suggested that policymakers, social workers, and organizations should pay sufficient attention to parents' mental health, financial stress, employment situations as well as stress resolutions to prevent and mitigate the harm of maltreatment brought to children.At the same time, it's important to note that child maltreatment is a complex issue with many contributing factors.It's difficult to determine a direct causal relationship between these variables and child maltreatment rates without conducting further research.

Table 1 :
Importance Score Ranking

Table 5 :
Single-difference Model Output