Note: This report presents results for the Urban Spatial/Predict Align Prevent analysis predicting geospatial child maltreatment risk in the District of Columbia. The work presented here is based on the PAP research protocol. Some additions will be made over the next week, which will be shared before the final presentation the week of March 9th.

This document is designed to be read alongside the Urban Spatial/PAP Technical Appendix. Links are provided throughout the document to the corresponding sections in the Technical Appendix to provide an in-depth explanation of the methods used in the workflow.

For readers interested in the policy implications of this study, an Executive Summary is provided in Section 1 and the strategic planning tools are demonstrated in Section 2. Sections 3 and 4 present more details on the workflow and predictive model. Section 5 concludes.

1. Executive Summary

Washington D.C., our nation’s capital, has a population of 672,000 spread out across ~68 square miles. D.C. has experienced 677, 722, 892, 923, 948 substantiated reports of child maltreatment in 2015, 2016, 2017, 2018, and 2019 respectively. As Figure 1.2 below suggests, maltreatment is on the rise.

While the number of maltreatment events may be increasing, the spatial distribution has remained consistent over time. Figure 1.3 visualizes maltreatment events as hotspot or kernel density maps and is clearly clustered in a handful of communities, most notably, the Anacostia neighborhood in the southeast.

This spatial pattern is most pronounced when maltreatment events for 2017, 2018, and 2019 are summed by neighborhood and ward, as in Figure 1.4.

Child maltreatment is relatively rare, and by nature, likely under-reported. There is likely more maltreatment activity than what is reported by community stakeholders or substantiated by the City. Thus, the observed count of substantiated maltreatment likely discounts the actual risk of maltreatment.

We characterize this phenomena as latent risk and present a set of analytics below that predict latent maltreatment risk throughout D.C. to help stakeholders better ‘align’ limited child welfare resources with demand for those services. Washington D.C. deploys at least two different community-level interventions to mitigate maltreatment risk. Our goal is to provide a robust tool that allows them to better target these limited resources in the communities that need them most. A geospatial risk predictive model is estimated to predict maltreatment risk for every location in Washington D.C. The intuition is to borrow the observed maltreatment ‘experience’ in D.C. and test whether those experiences generalize to places where maltreatment may be occurring but is not directly observed. If this experience is generalizable, one can confidently forecast maltreatment risk across space.

The basic hypothesis of the approach is that maltreatment risk is a function of exposure to a series of geospatial risk and protective factors. Examples of the former might be blight, while examples of the latter may include community centers, libraries, and churches. Exposure between maltreatment and risk/protective factors is measured by relating one to the other across a lattice grid of polygons, the ‘fishnet’, that comprise the entire city (Figure 1.5). The size of each grid cell is 1000 by 1000 ft2 - a size comparable to roughly five city blocks. Empirically, this scale helps satisfy both the assumptions of the statistical model used to predict and the need to target community interventions at a scale smaller than the neighborhood but larger than a single city block.

Figure 1.5 visualizes the study area of this analysis. Approximately 20 square miles (30% of the city’s total area) of national park land and hydrology is masked from the analysis. Maltreatment events for 2017, 2018, and 2019 are summed for each grid cell, and in Figure 1.6 below, the maltreatment rate per 100 people is visualized. A clearer picture of the maltreatment hotspots in Washington D.C. now emerge.

Additional hypothesis testing reveals hotspots exhibit greater local clustering than we might otherwise expect due to random chance alone. Figure 1.7 visualizes both maltreatment counts for each fishnet grid cell and those local maltreatment clusters significant at the p = 0.05 level. These significant clusters appear in Anacostia, Columbia Heights, and near Benning.

Once maltreatment is associated with each fishnet grid cell. Several ‘feature engineering’ strategies, explained in Section 3 below, are developed measuring exposure from each grid cell to nearby risk and protective factors. Figure 1.8 visualizes the three such feature engineering strategies. Figure 1.8a visualizes the locations of homeless shelters. Figure 1.8b, 1.8c, and 1.8d measure exposure by: counting the number of shelters for each grid cell; measuring the euclidean or straight-line distance from each cell to its nearest shelter; and measuring the average distance from each grid cell to its n nearest shelters.

Exploratory Data Analysis (Section 3) estimates correlations for each of the three feature engineering strategies across the 29 risk and protective factors gathered for this analysis. A set of final features are selected for inclusion in the predictive modeling phase. These features are then used to predict maltreatment count across D.C., and do so with an error rate of less than half of one maltreatment event on average. Far more methodological context for the predictive model is provided in Section 4 below as well as Section 6 and Section 7 in the Technical Appendix.

Figure 1.9 visualizes the predicted count of maltreatment throughout the city as well as prediction errors. These predictions are a measure of maltreatment risk and not surprisingly, cluster in the aforementioned neighborhoods. Section 4 provides a host of analytics that test predictive accuracy and generalizability across different D.C. communities.

Finally, Figure 1.10 provides a measure of how well the predictive model helps to target limited child welfare resources relative to the most common method of hotspot analysis - kernel density. Figure 1.10a shows the kernel density calculated on maltreatment from 2017 and 2018. The overlaid points are maltreatment events from 2019 (offset slightly in space to protect individual privacy). Figure 1.10b visualizes the maltreatment risk predictions, trained from 2017 and 2018 maltreatment events with 2019 events overlaid.

Validating predictions generated from 2017-2018 on 2019 maltreatment events provides a critical test for the model’s ability to predict into the near future. Visually, the risk predictions provide a better tool for targeting community interventions. There are also a host of locations where few 2019 events actually occurred, highlighting places with latent risk for maltreatment. Section 4 provides an additional test of the potential for the predictive model to target community interventions.

The geospatial risk prediction model created for this project provides an objective tool child welfare practitioners in Washington D.C. can use to allocate limited resources in the communities most at need.

The next section describes how these risk predictions can be adopted into a strategic planning process to ‘align’ the supply of child welfare resources with the demand for those services in space. Section 3 discusses the Exploratory Data Analysis; Section 4 presents the predictive model; and Section 5 concludes.

2. Align

The risk prediction methodology is most useful when deployed in a strategic planning framework. Such a framework is introduced here. The section begins by relating maltreatment risk to a host of other social phenomena. Predicted risk is then associated with protective factors throughout D.C. - places where stakeholders may wish to deploy community interventions. Finally, a simple gap analysis is performed to show neighborhoods where there are more protective factors than maltreatment and conversely, areas where there is more risk than protection.

2.1 How many Washingtonians live in high risk areas?

How many people live in areas with high predicted maltreatment risk? Figure 2.1 shows that approximately 105,000 people live in the highest risk category, an area that comprises only about 10% of the District. Another 160,000 people are located in the second highest risk category, indicating nearly half of Washington, D.C. residents live in areas of high demand for child welfare resources.1

2.2 What is the relationship between poverty and maltreatment risk?

Figure 2.2 maps the area weighted poverty rate by grid cell for Washington, D.C. To what extent is poverty and maltreatment risk associated? Figure 2.2 also plots predicted maltreatment risk as a function of weighted poverty rate. The scatter plot shows the correlation between poverty and predicted risk is moderate and is confirmed by a correlation coefficient of 0.46. For more information on how the weighted poverty rate is calculated, please refer to Section 8.2 in the Technical Appendix.

2.3 Maltreatment risk and removals of a child from the household

How does maltreatment risk relate to child removals? In 2019, there were 309 cases where a child was removed from the home. Half of these removal cases occurred in the highest risk category, with another 33% of removals taking place in the second highest risk category (Figure 2.3).

2.4 Do alternative response cases occur in high risk places?

There is a growing recognition that not all child protective services cases require a traditional investigation. In these instances, alternative response provides an assessment of family needs that helps community agencies and other organizations support lower risk households. Alternative response frees up limited resources that child welfare agencies can then focus on higher risk cases in which maltreatment has been confirmed. Figure 2.4 shows that in Washington, D.C. the majority of alternative response cases are located in the highest and second highest risk categories. This outcome suggests that alternative response resources are currently well aligned with maltreatment risk.

2.5 Maltreatment risk and foster homes

Are foster homes located in high-risk areas? There is limited research on the question of whether foster home prevalence reduces environmental maltreatment risk. Nevertheless, if foster homes are a protective factor, it is reasonable to analyze their distribution relative to risk.

Figure 2.5a maps foster homes in D.C. For privacy reasons, foster homes in grid cells with one point are masked and all other locations are offset slightly at random. Figure 2.5b shows that the largest number of foster homes (approximately 38%) are located in the second highest risk category. About 30% are in the highest risk category.

2.6 Assigning risk scores to protective land uses

Protective locations are those where child welfare stakeholders may wish to deploy education, outreach, and prevention resources in the community. An ideal location is one in a high risk area. Figure 2.6 maps protective locations. Figure 2.7 extends a quarter mile radius or ‘buffer’ around each protective location and calculates the average risk in the local area.

To communicate which specific resources are optimally located relative to maltreatment risk, Table 2.1 lists the top 20 locations citywide. These 20 locations are visualized in Figure 2.8 below. Child welfare stakeholders can use this table to more effectively target community resources.

Protective_Use Address Mean_Predicted_Count
Food Bank 896 Southern AVE SE Washington DC 20032 9.681583
Food Bank 840 Chesapeake St SE Washington DC 20032 9.308504
SNAP 601 Chesapeake St Se 8.866987
SNAP 607 Chesapeake St Se 8.866987
Grocers 601 CHESAPEAKE ST SE 8.866987
Child Development Centers 908 SOUTHERN AVE, SE 8.820443
Schools 1700 Q STREET SE 8.683494
Religious Organizations 1606 17TH PL SE 8.674025
Child Development Centers 4228 4TH ST, SE 8.496206
Food Bank 1345 Savannah St SE Washington DC 20032 8.446890
Parks Minnesota Ave.,36th St.,Croffut Pl.,SE 8.376615
Schools 885 BARNABY STREET SE 8.336277
Schools 3650 ELY PLACE SE 8.155387
Schools 425 CHESAPEAKE STREET SE 8.127060
Food Bank 4301 9th ST SE Washington DC 20032 8.110176
Schools 4301 9TH STREET SE 8.110176
Religious Organizations 4025 9TH ST SE 7.986401
Child Development Centers 4224 6TH ST, SE 7.972891
Religious Organizations 5228 CALL PL SE 7.950444
Child Development Centers 5111 E ST, SE 7.945320
Child Development Centers 5100 E ST, SE 7.945320
Religious Organizations 5100 E ST SE 7.945320
Table 2.1


The Child and Family Services Agency in Washington, D.C. currently manages two community intervention programs - Collaboratives and Families First D.C. centers.2 Collaboratives have been a community resource for years, however, Families First D.C. centers are new interventions and what services will be offered at each location are currently being discussed internally. Are these community programs located in the optimal locations? Figure 2.9 highlights the measure of risk exposure for these resources. Table 2.2 sorts these interventions by average predicted count.

Protective_Use Address Mean_Predicted_Count
Families First DC 2357 Ainger Place SE, Washington, DC 6.9029881
Families First DC 5035 C Street SE, Washington, DC 6.5557347
Families First DC 260 54th Street NE, Washington, DC 6.2850045
Families First DC 1351 Alabama Ave SE, Washington, DC 5.6979627
Collaborative 3333 14th St NW, Washington, DC 5.5751504
Families First DC 4 Atlantic St SW, Washington, DC 5.2355430
Families First DC 400 Atlantic Street SE, Washington, DC 4.7798401
Families First DC 2375 Elvans Rd SE, Washington, DC 4.3558766
Collaborative 3917 Minnesota Ave NE, Washington, DC 2.9986454
Families First DC 3917 Minnesota Avenue NE, Washington DC 2.9986454
Families First DC 3744 Hayes Street NE, Washington, DC 2.4977619
Collaborative 601 Edgewood St SE, Washington, DC 2.4595652
Collaborative 4420 Georgia Ave NW, Washington, DC 0.5140122
Table 2.2

2.7 Average risk by ward

To analyze average maltreatment risk by ward, Figure 2.10 maps the average predicted count of maltreatment events, illustrating greatest risk in Ward 7, Ward 8, and Ward 1.

2.8 Does risk correlate with school quality?

Figure 2.11 visualizes the average maltreatment risk and the STAR Framework Rating for Elementary school attendance boundaries. The STAR Framework Rating is a categorical value ranging from 1 to 5, where 5 is the best rating. The figure shows that schools located in Southeast DC have a greater number of students that could be at risk of child maltreatment. These schools also have lower STAR Framework Ratings as well.

Figure 2.12 further explores the relationship between maltreatment risk and school quality, plotting average risk by attendance boundary as a function of STAR Framework Scores. Unlike the STAR Framework Rating, the STAR Framework Scores are raw scores ranging from 0 to 100, where 100 is highest quality. This plot suggests that as school quality increases, the average predicted maltreatment count decreases. This relationship is confirmed with a correlation coefficient of -0.58.

2.9 Where should CPTED interventions locate?

Crime Prevention Through Environmental Design (CPTED) is the concept that appropriate environmental design can deter crime incidents by increasing visibility and reinforcing ownership of spaces. To understand where CPTED interventions should occur, Table 2.3 identifies risk features in the highest risk category.

Risk_Location Address Predicted_Count
Bus Stop 13TH ST SE + CONGRESS ST SE 13
Bus Stop CONGRESS ST SE + 13TH ST SE 13
Bus Stop CONGRESS ST SE + #1326 13
Bus Stop CONGRESS ST SE + 13TH PL SE 13
Bus Stop 13TH ST SE + SAVANNAH ST SE 13
Hair/Nail Salons 212 36TH ST SE 12
Public Housing 155 Ridge Road SE 12
Bus Stop 37TH ST SE + 37TH PL SE 12
Bus Stop 37TH ST SE + RIDGE RD SE 12
Bus Stop 6TH ST SE + GALVESTON PL SE 11
Bus Stop 6TH ST SE + DARRINGTON ST SE 11
Bus Stop 4TH ST SE + BRANDYWINE ST SE 11
Bus Stop 4TH ST SE + CHESAPEAKE ST SE 11
Bus Stop SAVANNAH ST SE + 18TH ST SE 11
Bus Stop TRENTON PL SE + 19TH ST SE 11
Bus Stop MISSISSIPPI AVE + 19TH ST 11
Bus Stop TRENTON PL SE + STANTON RD SE 11
Bus Stop 19TH ST SE + SAVANNAH ST SE 11
Public Housing 2311 Ainger Place SE 11
Bus Stop BRUCE PL SE + FORT PL SE 11
Bus Stop BRUCE PL SE + RAYNOLDS PL SE 11
Bus Stop LANGSTON PL SE + RAYNOLDS PL SE 11
Bus Stop MARYLAND AVE NE + 21ST ST NE 11
Public Housing 1501 Half Street SW 11
Bus Stop O ST SW + 1ST ST SW 11
Bus Stop P ST SW + CANAL ST SW 11
Bus Stop DELAWARE AVE SW + CANAL ST SW 11
Bus Stop P ST SW + 1ST ST SW 11
Liquor Retailer 1399 HALF STREET SW 11
Public Housing 4450 G Street SE 11
Bus Stop ALABAMA AVE SE + F ST SE 11
Bus Stop G ST SE + ALABAMA AVE SE 11
Bus Stop ALABAMA AVE SE + G ST SE 11
Convenience Stores 3012 14TH ST NW 11
NonDepository Banks 3018 14TH STREET NW 11
Public Housing 1475 Columbia Road NW 11
Bus Stop 14TH ST NW + IRVING ST NW 11
Bus Stop 14TH ST NW + HARVARD ST NW 11
Bus Stop COLUMBIA RD NW + 14TH ST NW 11
Bus Stop 14TH ST NW + COLUMBIA RD NW 11
Bus Stop IRVING ST NW + 14TH ST NW 11
Restaurants w/ Licenses (Bars) 1400 IRVING STREET NW 11
Bus Stop CHESAPEAKE ST SE + BARNABY RD SE 10
Bus Stop BARNABY ST SE + CHESAPEAKE ST SE 10
Bus Stop BARNABY ST SE + ATLANTIC ST SE 10
Bus Stop BARNABY ST SE + 9TH ST SE 10
Bus Stop SOUTHERN AVE SE + WHEELER RD SE 10
Bus Stop SOUTHERN AVE SE + WHEELER RD 10
Bus Stop WHEELER RD SE + BELLEVUE ST SE 10
Bus Stop SOUTHERN AVE SE + 9TH ST SE 10
Convenience Stores 2922 MARTIN LUTHER KING JR AVE 10
Convenience Stores 2921 MARTIN LUTHER KING JR AVE 10
Hair/Nail Salons 3012 MARTIN LUTHER KING JR AVE 10
Hair/Nail Salons 3025 MLK JR AVE SE 10
Check_Cashing 2922 MARTIN LUTHER KING JR AVENUE SE 10
Check_Cashing 2931 MARTIN LUTHER KING JR AVENUE SE 10
Bus Stop MELLON ST SE + 5TH ST SE 10
Bus Stop MALCOLM X AVE SE + NEWCOMB ST SE 10
Bus Stop MELLON ST SE + MARTIN LUTHER KING JR AVE SE 10
Bus Stop MALCOLM X AVE SE + OAKWOOD ST SE 10
Bus Stop MALCOLM X AVE SE + MARTIN LUTHER KING JR AVE SE 10
Bus Stop MARTIN LUTHER KING JR AVE SE + MALCOLM X AVE SE 10
Liquor Retailer 2922 1/2 MARTIN LUTHER KING JR AVENUE SE 10
Bus Stop NAYLOR RD SE + ERIE ST SE 10
Bus Stop NAYLOR RD SE + 28TH ST SE 10
Bus Stop NAYLOR RD SE + GAINESVILLE ST SE 10
Bus Stop MINNESOTA AVE SE + C ST SE 10
Bus Stop MINNESOTA AVE SE + B ST SE 10
Bus Stop MINNESOTA AVE SE + 36TH ST SE 10
Convenience Stores 3542 E CAPITOL ST NE 10
Check_Cashing 1237 MOUNT OLIVET ROAD NE 10
Bus Stop MONTELLO AVE NE + SIMMS PL NE 10
Bus Stop TRINIDAD AVE NE + MEIGS PL NE 10
Bus Stop 51ST ST SE + G ST SE 10
Bus Stop 51ST ST SE + E ST SE 10
Bus Stop FITCH ST SE + 53RD ST SE 10
Bus Stop 53RD ST SE + E ST SE 10
Bus Stop 51ST ST SE + D ST SE 10
Convenience Stores 2600 14TH ST NW 10
Bus Stop 14TH ST NW + FAIRMONT ST NW 10
Bus Stop 14TH ST NW + GIRARD ST NW 10
Bus Stop 14TH ST NW + EUCLID ST NW 10
Liquor Retailer 2655 15TH STREET NW 10
Check_Cashing 4682 MARTIN LUTHER KING JR AVENUE SW 9
Bus Stop JOLIET ST SW + #159-161 9
Bus Stop IRVINGTON ST SW + IVANHOE ST SW 9
Bus Stop IRVINGTON ST SW + MARTIN LUTHER KING JR AVE SW 9
Bus Stop JOLIET ST SW + MARTIN LUTHER KING JR AVE SW 9
Bus Stop JOLIET ST SW + S CAPITOL TER SW 9
Liquor Retailer 4686 MARTIN LUTHER KING JR AVENUE SW 9
Bus Stop SOUTHERN AVE SE + CHESAPEAKE ST SE 9
Check_Cashing 4009 SOUTH CAPITOL STREET SW 9
Bus Stop ATLANTIC ST SW + S CAPITOL ST SW 9
Bus Stop MARTIN LUTHER KING JR AVE SW + ATLANTIC ST SW 9
Bus Stop MARTIN LUTHER KING JR AVE SW + XENIA ST SW 9
Bus Stop S CAPITOL ST SW + #3921 9
Bus Stop MISSISSIPPI AVE SE + ATLANTIC ST SE 9
Bus Stop S CAPITOL ST SW + ATLANTIC ST SW 9
Bus Stop ATLANTIC ST SE + SOUTH CAPITOL ST SE 9
Liquor Retailer 3916 SOUTH CAPITOL STREET SE 9
Liquor Retailer 3919 SOUTH CAPITOL STREET SW 9
Public Housing 2440 Elvans Road SE 9
Bus Stop ELVANS RD SE + #2458 1/2 9
Bus Stop ELVANS RD SE + STANTON RD SE 9
Public Housing 5231 Cloud Place NE 9
Public Housing 4500 Quarles Street NE 9
Bus Stop 17TH ST NE + BLADENSBURG RD NE 9
Bus Stop 17TH ST NE + M ST NE 9
Bus Stop G ST SE + 46TH ST SE 9
Bus Stop G ST SE + BENNING RD SE 9
Laundromats 1603 GOOD HOPE RD SE 9
NonDepository Banks 1736 GOOD HOPE ROAD SE 9
Bus Stop 16TH ST SE + GOOD HOPE RD SE 9
Bus Stop GOOD HOPE RD SE + 17TH ST SE 9
Bus Stop MINNESOTA AVE SE + S ST SE 9
Bus Stop 16TH ST SE + T ST SE 9
Bus Stop GOOD HOPE RD SE + 16TH ST SE 9
Liquor Retailer 1537 GOOD HOPE ROAD SE 9
Bus Stop STANTON RD SE + WASHINGTON VIEW APARTMENT #94 9
Bus Stop STANTON RD SE + SHERIDAN RD SE 9
Bus Stop MINNESOTA AVE SE + 18TH ST SE 9
Bus Stop 16TH ST SE + R ST SE 9
Bus Stop MINNESOTA AVE SE + 19TH ST SE 9
Bus Stop ATLANTIC ST SE + 1ST ST SE 8
Laundromats 3700 MARTIN LUTHER KING JR AVE 8
Hair/Nail Salons 3707 MARTIN LUTHER KING JR AVE 8
Bus Stop MARTIN LUTHER KING JR AVE SE + UPSAL ST SE 8
Bus Stop MARTIN LUTHER KING JR AVE SE + #3726 8
Liquor Retailer 3705 MARTIN LUTHER KING JR AVENUE SE 8
Bus Stop 6TH ST SE + #3500 8
Bus Stop 6TH ST SE + #3633 8
Bus Stop MALCOLM X AVE SE + 2ND ST SE 8
Bus Stop ORANGE ST SE + OAKWOOD ST SE 8
Bus Stop ALABAMA AVE SE + AINGER PL SE 8
Bus Stop AINGER PL SE + LANGSTON PL SE 8
Bus Stop ALABAMA AVE SE + 25TH ST SE 8
Bus Stop MINNESOTA AVE NE + E CAPITOL ST NE 8
Bus Stop MINNESOTA AVE NE + BLAINE ST NE 8
Bus Stop MINNESOTA AVE NE + BLAINE ST 8
Liquor Retailer 3728 MINNESOTA AVENUE NE 8
Public Housing 1140 North Capitol Street NW 8
Bus Stop N CAPITOL ST NE + M ST NE 8
Bus Stop N CAPITOL ST NW + PIERCE ST NE 8
Bus Stop M ST NW + 1ST PL NW 8
Convenience Stores 950 EASTERN AVE NE 8
Bus Stop SHERIFF RD NE + 51ST ST NE 8
Bus Stop DIVISION AVE NE + EASTERN AVE NE 8
Bus Stop DIVISION AVE NE + JAY ST NE 8
Bus Stop 51ST ST NE + SHERIFF RD NE 8
Bus Stop SHERIFF RD NE + DIVISION AVE NE 8
Bus Stop DIVISION AVE NE + JUST ST NE 8
Liquor Retailer 1102 EASTERN AVENUE NE 8
Convenience Stores 1317 9TH ST NW 8
Hair/Nail Salons 1326 8TH ST NW # A 8
Hair/Nail Salons 700 O ST NW 8
Bus Stop 7TH ST NW + P ST NW 8
Bus Stop P ST NW + 9TH ST NW 8
Bus Stop 7TH ST NW + O ST NW 8
Bus Stop 7TH ST NW + N ST NW 8
Bus Stop P ST NW + 7TH ST NW 8
Bus Stop 9TH ST + N ST 8
Bus Stop 9TH ST NW + P ST NW 8
Restaurants w/ Licenses (Bars) 801 O STREET NW 8
Restaurants w/ Licenses (Bars) 1314 9TH STREET NW 8
Liquor Retailer 1400 7TH STREET NW 8
Liquor Retailer 1428 9TH STREET NW 8
Restaurants w/ Licenses (Bars) 1336 9TH STREET NW 8
Restaurants w/ Licenses (Bars) 922 N STREET NW 8
Liquor Retailer 1251 9TH STREET NW 8
Restaurants w/ Licenses (Bars) 1420 8TH STREET NW 8
Restaurants w/ Licenses (Bars) 1414 9TH STREET NW 8
Restaurants w/ Licenses (Bars) 1316 9TH STREET NW 8
Restaurants w/ Licenses (Bars) 1334 9TH STREET NW 8
Bus Stop MARYLAND AVE NE + 18TH ST NE 8
Bus Stop 17TH ST NE + MARYLAND AVE NE 8
Bus Stop MARYLAND AVE NE + 19TH ST NE 8
Bus Stop 17TH ST NE + K ST NE 8
Liquor Retailer 901 17TH STREET NE 8
Public Housing 2101 4th Street NW 8
Public Housing 2125 4th Street NW 8
Bus Stop BRYANT ST NW + #301 8
Bus Stop W ST NW + 2ND ST NW 8
Bus Stop 4TH ST NW + V ST NW 8
Bus Stop 2ND ST NW + BRYANT ST NW 8
Bus Stop W ST NW + 4TH ST NW 8
Convenience Stores 5026 BENNING RD SE 8
Hair/Nail Salons 5004 BENNING RD SE 8
NonDepository Banks 5026 BENNING ROAD SE 8
Bus Stop BENNING RD SE + HILLSIDE RD SE 8
Bus Stop BENNING RD SE + H ST SE 8
Bus Stop H ST SE + BENNING RD SE 8
Liquor Retailer 5000 BENNING ROAD SE 8
Public Housing 501 51st Street SE 8
Bus Stop BENNING RD SE + HANNA PL SE 8
Bus Stop HANNA PL SE + BENNING RD SE 8
Bus Stop BENNING RD SE + F ST SE 8
Bus Stop 53RD ST SE + C ST SE 8
Bus Stop 16TH ST SE + V ST SE 8
Bus Stop 16TH ST SE + U ST SE 8
Liquor Retailer 1535 U STREET SE 8
Liquor Retailer 2200 16TH STREET SE 8
Bus Stop C ST NE + 17TH ST NE 8
Bus Stop D ST NE + 17TH ST NE 8
Bus Stop D ST NE + 16TH ST NE 8
Bus Stop BRENTWOOD RD NE + BRYANT ST NE 8
Bus Stop BRENTWOOD RD NE + 13TH ST NE 8
Bus Stop SARATOGA AVE NE + 12TH ST NE 8
Hair/Nail Salons 1420 SARATOGA AVE NE 8
Bus Stop CHESAPEAKE ST SE + 6TH ST SE 7
Bus Stop 6TH ST SE + CHESAPEAKE ST SE 7
Liquor Retailer 601 CHESAPEAKE STREET SE 7
Public Housing 400 Atlantic Street SE 7
Bus Stop BARNABY ST SE + HR DR SE 7
Bus Stop BARNABY ST SE + BRANDYWINE ST SE 7
Bus Stop 8TH ST SE + ATLANTIC ST SE 7
Bus Stop 8TH ST SE + YUMA ST SE 7
Bus Stop ATLANTIC ST SE + 8TH ST SE 7
Bus Stop 15TH ST SE + SAVANNAH ST SE 7
Bus Stop ALABAMA AVE SE + CONGRESS ST SE 7
Bus Stop ALABAMA AVE SE + 15TH ST SE 7
Bus Stop CONGRESS ST SE + SAVANNAH ST SE 7
Bus Stop ALABAMA AVE SE + 15TH PL SE 7
Bus Stop ALABAMA AVE SE + 18TH PL SE 7
Bus Stop STANTON RD SE + ELVANS RD SE 7
Bus Stop STANTON RD SE + SUITLAND PKY SE 7
Hair/Nail Salons 1832 WOODMONT PL SE 7
Bus Stop GOOD HOPE RD SE + ALTAMONT PL SE 7
Convenience Stores 1830 BENNING RD NE 7
Hair/Nail Salons 1803 BENNING RD NE 7
Hair/Nail Salons 1801 BENNING RD NE 7
Hair/Nail Salons 1820 BENNING RD NE 7
Bus Stop BENNING RD NE + 18TH ST NE 7
Bus Stop 17TH ST NE + BENNING RD NE 7
Bus Stop BENNING RD NE + 19TH ST NE 7
Liquor Retailer 1818 BENNING ROAD NE 7
Liquor Retailer 1830 BENNING ROAD NE 7
Restaurants w/ Licenses (Bars) 1831 BENNING ROAD NE 7
Check_Cashing 830 BLADENSBURG ROAD NE 7
Bus Stop TRINIDAD AVE NE + NEAL ST NE 7
Bus Stop 53RD ST SE + BASS PL SE 7
Bus Stop 53RD ST SE + CENTRAL AVE SE 7
Laundromats 5007 NEW HAMPSHIRE AVE NW 7
Check_Cashing 5008 1ST STREET NW 7
Bus Stop NEW HAMPSHIRE AVE NW + HAMILTON ST NW 7
Bus Stop NEW HAMPSHIRE AVE NW + 2ND ST NW 7
Bus Stop NEW HAMPSHIRE AVE NW + FARRAGUT ST NW 7
Liquor Retailer 5010 NEW HAMPSHIRE AVENUE NW 7
Bus Stop W ST SE + 16TH ST SE 7
Bus Stop W ST SE + 14TH ST SE 7
Bus Stop SHERIDAN RD SE + #2646 7
Bus Stop SHERIDAN RD SE + #2633 7
Bus Stop MARTIN LUTHER KING JR AVE SE + ST ELIZABETHS GATE 1 7
Bus Stop MARTIN LUTHER KING JR AVE SE + POMEROY RD SE 7
Bus Stop MARTIN LUTHER KING JR AVE SE + #2652 7
Bus Stop POMEROY RD SE + SHERIDAN RD SE 7
Bus Stop POMEROY RD SE + #2551-2576 7
Bus Stop POMEROY RD SE + #2500 7
Bus Stop POMEROY RD SE + #2575 7
Hair/Nail Salons 2643 BIRNEY PL SE 7
Public Housing 1249 Eaton Road SE 7
Public Housing 1230 Sumner Road SE 7
Hair/Nail Salons 2644 BIRNEY PL SE 7
Bus Stop SUMNER RD SE + WADE RD SE 7
Bus Stop SHERIDAN RD SE + SUITLAND PKY SE 7
Bus Stop SHERIDAN RD SE + BOWEN RD SE 7
Bus Stop MARTIN LUTHER KING JR AVE SE + SUMNER RD SE 7
Bus Stop MARTIN LUTHER KING JR AVE SE + EATON RD SE 7
Bus Stop MARTIN LUTHER KING JR AVE SE + STANTON RD SE 7
Liquor Retailer 2600 WADE ROAD SE 7
Bus Stop CONDON TER SE + YUMA ST SE 6
Bus Stop CONDON TER SE + 4TH ST SE 6
Bus Stop CONDON TER SE + PEPCO POLE 6
Bus Stop CONGRESS ST SE + SAVANNAH PL SE 6
Bus Stop MISSISSIPPI AVE + 21ST ST 6
Bus Stop 22ND ST SE + SOUTHERN AVE SE 6
Bus Stop SOUTHERN AVE SE + MISSISSIPPI AVE SE 6
Bus Stop 22ND ST SE + SAVANNAH ST SE 6
Bus Stop GOOD HOPE RD SE + WOODMONT PL SE 6
Bus Stop GOOD HOPE RD SE + MARBURY PLAZA APARTMENTS 6
Public Housing 400 50th Street NE 6
Bus Stop 50TH ST NE + BANKS PL NE 6
Bus Stop BANKS PL NE + 50TH ST NE 6
Bus Stop 50TH ST NE + #321 6
Bus Stop 51ST ST NE + BANKS PL NE 6
Bus Stop DIVISION AVE NE + CLOUD PL NE 6
Bus Stop FITCH PL NE + 51ST ST NE 6
Laundromats 1653 BENNING RD NE 6
Hair/Nail Salons 1567 MARYLAND AVE NE 6
Check_Cashing 1512 BENNING ROAD NE 6
Check_Cashing 1601 MARYLAND AVENUE NE 6
NonDepository Banks 1653 BENNING ROAD NE 6
NonDepository Banks 1528 BENNING ROAD NE 6
Bus Stop MARYLAND AVE NE + BLADENSBURG RD NE 6
Bus Stop MARYLAND AVE NE + NEAL ST NE 6
Bus Stop BENNING RD NE + 17TH ST NE 6
Bus Stop MARYLAND AVE NE + MORSE ST NE 6
Bus Stop BENNING RD NE + 16TH ST NE 6
Bus Stop BENNING RD NE + 15TH ST NE 6
Liquor Retailer 1601 BENNING ROAD NE 6
Bus Stop 51ST ST SE + B ST SE 6
Bus Stop 51ST ST SE + C ST SE 6
Bus Stop 51ST ST SE + ASTOR PL SE 6
Bus Stop ASPEN ST NW + 13TH PL NW 6
Bus Stop GEORGIA AVE NW + VAN BUREN ST NW 6
Bus Stop ASPEN ST NW + GEORGIA AVE NW 6
Bus Stop GEORGIA AVE NW + ASPEN ST 6
Bus Stop FRANKLIN ST NE + 4TH ST NE 6
Bus Stop EDGEWOOD ST NE + 6TH ST NE 6
Bus Stop FRANKLIN ST NE + 6TH ST NE 6
Bus Stop FRANKLIN ST NE + MONTANA AVE NE 6
Laundromats 3127 MOUNT PLEASANT ST NW 6
Convenience Stores 3146 MOUNT PLEASANT ST NW 6
Hair/Nail Salons 3112 MOUNT PLEASANT ST NW 6
Hair/Nail Salons 3161 MOUNT PLEASANT ST NW 6
Hair/Nail Salons 3121 MOUNT PLEASANT ST NW 6
Hair/Nail Salons 3171 MOUNT PLEASANT ST NW 6
Check_Cashing 3112 MOUNT PLEASANT STREET NW 6
Bus Stop 16TH ST NW + LAMONT ST NW 6
Bus Stop MT PLEASANT ST NW + LAMONT ST NW 6
Bus Stop MT PLEASANT ST NW + KENYON ST NW 6
Bus Stop LAMONT ST NW + MT PLEASANT ST NW 6
Bus Stop 16TH ST NW + PARK RD NW 6
Restaurants w/ Licenses (Bars) 3155 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3107 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3125 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 1660 LAMONT STREET NW 6
Restaurants w/ Licenses (Bars) 3209 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3102 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3203 MOUNT PLEASANT STREET NW 6
Liquor Retailer 3170 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3201 MOUNT PLEASANT STREET NW 6
Liquor Retailer 3158 MOUNT PLEASANT STREET NW 6
Restaurants w/ Licenses (Bars) 3162 MOUNT PLEASANT STREET NW 6
Liquor Retailer 3178 MOUNT PLEASANT STREET NW 6
Hair/Nail Salons 1460 PARK RD NW 6
Hair/Nail Salons 1377 KENYON ST NW 6
Hair/Nail Salons 1403 PARK RD NW 6
Hair/Nail Salons 3328 14TH ST NW 6
Hair/Nail Salons 1452 PARK RD NW 6
NonDepository Banks 3100 14TH STREET NW 6
Bus Stop 14TH ST NW + PARK RD NW 6
Bus Stop 14TH ST NW + MONROE ST NW 6
Restaurants w/ Licenses (Bars) 3345 14TH STREET NW 6
Restaurants w/ Licenses (Bars) 1424 PARK ROAD NW 6
Liquor Retailer 3103 14TH STREET NW 6
Restaurants w/ Licenses (Bars) 1413 PARK ROAD NW 6
Restaurants w/ Licenses (Bars) 1436 PARK ROAD NW 6
Restaurants w/ Licenses (Bars) 3115 14TH STREET NW 6
Restaurants w/ Licenses (Bars) 3310 14TH STREET NW 6
Liquor Retailer 3401 14TH STREET NW 6
Liquor Retailer 1345 PARK ROAD NW 6
Restaurants w/ Licenses (Bars) 3365 14TH STREET NW 6
Hair/Nail Salons 6 ELMIRA ST SW 5
Liquor Retailer 4401 SOUTH CAPITOL STREET SW 5
Bus Stop MISSISSIPPI AVE SE + 1ST ST SE 5
Bus Stop MISSISSIPPI AVE SE + WAYNE PL SE 5
Convenience Stores 4133 WHEELER RD SE 5
Check_Cashing 4137 WHEELER ROAD SE 5
Bus Stop WHEELER RD SE + BARNABY TER SE 5
Bus Stop WHEELER RD SE + VARNEY ST SE 5
Bus Stop WHEELER RD SE + WAHLER PL SE 5
Bus Stop BARNABY ST SE + WHEELER RD SE 5
Liquor Retailer 4133 WHEELER ROAD SE 5
Convenience Stores 3509 WHEELER RD SE 5
Check_Cashing 3505 WHEELER ROAD SE 5
Bus Stop MISSISSIPPI AVE SE + WHEELER RD SE 5
Bus Stop WHEELER RD SE + MISSISSIPPI AVE SE 5
Bus Stop WHEELER RD SE + CONGRESS ST SE 5
Hair/Nail Salons 3130 MARTIN LUTHER KING JR AVE 5
Bus Stop SAVANNAH ST SE + 4TH ST SE 5
Bus Stop 4TH ST SE + SAVANNAH ST SE 5
Bus Stop MARTIN LUTHER KING JR AVE SE + WACLARK PL SE 5
Bus Stop MARTIN LUTHER KING JR AVE SE + 4TH ST SE 5
Bus Stop MARTIN LUTHER KING JR AVE SE + #3353 5
Bus Stop MARTIN LUTHER KING JR AVE SE + HIGHVIEW PL SE 5
Bus Stop 4TH ST SE + MARTIN LUTHER KING JR AVE SE 5
Liquor Retailer 3333 MARTIN LUTHER KING JR AVENUE SE 5
Bus Stop ALABAMA AVE SE + 5TH ST SE 5
Bus Stop MARTIN LUTHER KING JR AVE SE + 5TH ST SE 5
Bus Stop ALABAMA AVE SE + RANDLE PL SE 5
Bus Stop 6TH ST SE + SAVANNAH ST SE 5
Bus Stop RANDLE PL SE + ALABAMA AVE SE 5
Liquor Retailer 3113 MARTIN LUTHER KING JR AVENUE SE 5
Liquor Retailer 600 ALABAMA AVENUE SE 5
Hair/Nail Salons 1559 ALABAMA AVE SE 5
Bus Stop ALABAMA AVE SE + 18TH ST SE 5
Bus Stop ALABAMA AVE SE + STANTON RD SE 5
Bus Stop STANTON RD SE + ANITA J TURNER ELEMENTARY SCHOOL 5
Bus Stop STANTON RD SE + ALABAMA AVE SE 5
Bus Stop STANTON RD SE + TUBMAN RD 5
Hair/Nail Salons 2275 SAVANNAH ST SE 5
Bus Stop SAVANNAH ST SE + 24TH ST SE 5
Bus Stop 25TH ST SE + #3417 5
Bus Stop SAVANNAH ST SE + 23RD ST SE 5
Liquor Retailer 2281 SAVANNAH STREET SE 5
Liquor Retailer 2283 SAVANNAH STREET SE 5
Public Housing 2700 Jasper Street SE 5
Bus Stop ALABAMA AVE SE + JASPER ST SE 5
Bus Stop ALABAMA AVE SE + IRVING PL SE 5
Bus Stop IRVING ST SE + 24TH PL SE 5
Bus Stop IRVING ST SE + #2433 5
Bus Stop ALABAMA AVE SE + HARTFORD ST SE 5
Convenience Stores 3029 NAYLOR RD SE 5
Hair/Nail Salons 3025 NAYLOR RD SE 5
Bus Stop 30TH ST SE + NAYLOR RD SE 5
Bus Stop 30TH ST SE + BUENA VISTA TER SE 5
Hair/Nail Salons 326 RIDGE RD SE 5
Bus Stop RIDGE RD SE + C ST SE 5
Bus Stop RIDGE RD SE + BURNS ST SE 5
Bus Stop RIDGE RD SE + D ST SE 5
Hair/Nail Salons 4248 BENNING RD NE 5
Bus Stop BENNING RD NE + 42ND ST NE 5
Liquor Retailer 4202 BENNING ROAD NE 5
Check_Cashing 4347 HUNT PLACE NE 5
Bus Stop NANNIE HELEN BURROUGHS AVE NE + 44TH ST NE 5
Bus Stop NANNIE HELEN BURROUGHS AVE NE + HAYES ST NE 5
Liquor Retailer 4401 NANNIE HELEN BURROUGHS AVENUE NE 5
Bus Stop MAYFAIR TER NE + #3675-3691 5
Bus Stop JAY ST NE + #3733-3741 5
Bus Stop MAYFAIR TER NE + #3531-3537 5
Convenience Stores 1612 KENILWORTH AVE NE 5
Bus Stop KENILWORTH AVE NE + QUARLES ST NE 5
Liquor Retailer 1612 KENILWORTH AVENUE NE 5
Bus Stop MONTELLO AVE NE + OWEN PL NE 5
Bus Stop MONTELLO AVE NE + NEAL ST NE 5
Bus Stop TRINIDAD AVE NE + LEVIS ST NE 5
Bus Stop TRINIDAD AVE NE + QUEEN ST NE 5
Public Housing 203 N Street SW 5
Public Housing 1265 Half Street SW 5
Public Housing 1200 Delaware Avenue SW 5
Bus Stop M ST + DELAWARE AVE 5
Bus Stop M ST SW + DELAWARE AVE SW 5
Bus Stop M ST SW + 1ST ST SW 5
Bus Stop BENNING RD SE + E ST SE 5
Bus Stop E ST SE + BENNING RD SE 5
Hair/Nail Salons 3434 OAKWOOD TER NW 5
Bus Stop 16TH ST NW + OAK ST NW 5
Bus Stop 16TH ST NW + #3636 5
Restaurants w/ Licenses (Bars) 3636 16TH STREET NW 5
Convenience Stores 3540 14TH ST NW 5
Convenience Stores 3500 14TH ST NW # A 5
Convenience Stores 3500 14TH ST NW 5
Hair/Nail Salons 3610 14TH ST NW 5
Hair/Nail Salons 3539 14TH ST NW 5
Hair/Nail Salons 3602 14TH ST NW 5
Check_Cashing 3609 14TH STREET NW 5
Bus Stop 14TH ST NW + PARKWOOD PL NW 5
Bus Stop 14TH ST NW + SPRING RD NW 5
Restaurants w/ Licenses (Bars) 3521 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3605 14TH STREET NW 5
Liquor Retailer 3620 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3612 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3566 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3614 14TH STREET NW 5
Liquor Retailer 1500 OGDEN STREET NW 5
Restaurants w/ Licenses (Bars) 3568 14TH STREET NW 5
Liquor Retailer 3515 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3548 14TH STREET NW 5
Bus Stop MORRIS RD SE + POMEROY RD SE 5
Bus Stop MORRIS RD SE + HUNTER PL SE 5
Bus Stop HUNTER PL SE + MORRIS RD SE 5
Bus Stop HUNTER PL SE + BANGOR ST SE 5
Bus Stop MORRIS RD SE + #1340 5
Bus Stop MORRIS RD SE + HIGH ST SE 5
Bus Stop MORRIS RD SE + #1349 5
Bus Stop STANTON RD SE + DOUGLAS RD SE 5
Bus Stop STANTON RD SE + POMEROY RD SE 5
Bus Stop POMEROY RD SE + STANTON RD 5
Hair/Nail Salons 2906 MINNESOTA AVE SE 5
Hair/Nail Salons 2920 MINNESOTA AVE SE 5
Hair/Nail Salons 2901 N ST SE 5
Check_Cashing 2900 MINNESOTA AVENUE SE 5
Bus Stop MINNESOTA AVE SE + NASH PL SE 5
Bus Stop MINNESOTA AVE SE + 28TH ST SE 5
Liquor Retailer 2924 MINNESOTA AVENUE SE 5
Laundromats 415 RHODE ISLAND AVE NE 5
Bus Stop EDGEWOOD ST NE + DOUGLAS ST NE 5
Bus Stop RHODE ISLAND AVE NE + #610 5
Bus Stop RHODE ISLAND AVE NE + #617 5
Liquor Retailer 528 RHODE ISLAND AVENUE NE 5
Convenience Stores 3001 SHERMAN AVE NW 5
Bus Stop 11TH ST NW + COLUMBIA RD NW 5
Bus Stop IRVING ST NW + 11TH ST NW 5
Bus Stop SHERMAN AVE NW + COLUMBIA RD NW 5
Bus Stop 11TH ST NW + IRVING ST NW 5
Bus Stop SHERMAN AVE NW + IRVING ST NW 5
Bus Stop COLUMBIA RD NW + SHERMAN AVE NW 5
Bus Stop IRVING ST NW + SHERMAN AVE NW 5
Bus Stop 11TH ST NW + HARVARD ST NW 5
Bus Stop COLUMBIA RD NW + 11TH ST NW 5
Liquor Retailer 2833 11TH STREET NW 5
Restaurants w/ Licenses (Bars) 2827 SHERMAN AVENUE NW 5
Liquor Retailer 3001 SHERMAN AVENUE NW 5
Liquor Retailer 2901 SHERMAN AVENUE NW 5
Hair/Nail Salons 1402 OGDEN ST NW 5
Hair/Nail Salons 3443 14TH ST NW # 2 5
Hair/Nail Salons 3453 14TH ST NW 5
Hair/Nail Salons 3453 14TH ST NW # B 5
Hair/Nail Salons 3409 14TH ST NW 5
Hair/Nail Salons 1400 OGDEN ST NW 5
Check_Cashing 3433 14TH STREET NW 5
NonDepository Banks 3443 14TH STREET NW 5
Public Housing 1456 Oak Street NW 5
Bus Stop 14TH ST NW + NEWTON ST NW 5
Bus Stop 14TH ST NW + OAK ST NW 5
Restaurants w/ Licenses (Bars) 3460 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3475 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 1400 MERIDIAN PLACE NW 5
Restaurants w/ Licenses (Bars) 3411 14TH STREET NW 5
Restaurants w/ Licenses (Bars) 3423 14TH STREET NW 5
Convenience Stores 3426 GEORGIA AVE NW 5
Hair/Nail Salons 3553 GEORGIA AVE NW 5
Hair/Nail Salons 701 NEWTON PL NW 5
Hair/Nail Salons 3335 GEORGIA AVE NW 5
Hair/Nail Salons 3505 GEORGIA AVE NW 5
Hair/Nail Salons 3506 GEORGIA AVE NW 5
Check_Cashing 3420 GEORGIA AVENUE NW 5
Public Housing 617 Morton Street NW 5
Bus Stop GEORGIA AVE NW + PARK RD NW 5
Bus Stop GEORGIA AVE NW + NEWTON PL NW 5
Restaurants w/ Licenses (Bars) 3333 GEORGIA AVENUE NW 5
Restaurants w/ Licenses (Bars) 3400 GEORGIA AVENUE NW 5
Restaurants w/ Licenses (Bars) 3541 GEORGIA AVENUE NW 5
Liquor Retailer 3504 GEORGIA AVENUE NW 5
Restaurants w/ Licenses (Bars) 3539 GEORGIA AVENUE NW 5
Restaurants w/ Licenses (Bars) 3549 GEORGIA AVENUE NW 5
Bus Stop 3RD ST SE + #4501 4
Bus Stop LIVINGSTON TER SE + #304 4
Bus Stop MELLON ST SE + #431 4
Bus Stop NEWCOMB ST SE + 4TH ST SE 4
Hair/Nail Salons 3050 STANTON RD SE 4
Public Housing 1804 Alabama Avenue SE 4
Bus Stop STANTON RD SE + JASPER PL SE 4
Bus Stop BRUCE PL SE + STANTON RD SE 4
Bus Stop BRUCE ST SE + 15TH PL SE 4
Bus Stop JASPER PL SE + #1350 4
Bus Stop STANTON RD SE + BRUCE PL SE 4
Bus Stop SOUTHERN AVE SE + #2912 4
Bus Stop SOUTHERN AVE SE + #2924 4
Bus Stop 30TH ST SE + SOUTHERN AVE SE 4
Bus Stop SOUTHERN AVE SE + 30TH ST SE 4
Bus Stop HAYES ST NE + ANACOSTIA AVE NE 4
Bus Stop NANNIE HELEN BURROUGHS AVE NE + DIVISION AVE NE 4
Bus Stop 50TH ST NE + FITCH PL NE 4
Bus Stop NANNIE HELEN BURROUGHS AVE NE + 50TH ST NE 4
Hair/Nail Salons 1379 H ST NE 4
Hair/Nail Salons 1421 H ST NE 4
NonDepository Banks 1396 FLORIDA AVENUE NE 4
NonDepository Banks 814 BLADENSBURG ROAD NE 4
Bus Stop 15TH ST NE + BENNING RD NE 4
Bus Stop BLADENSBURG RD NE + MARYLAND AVE NE 4
Bus Stop FLORIDA AVE NE + HOLBROOK ST NE 4
Bus Stop MARYLAND AVE NE + 14TH ST NE 4
Bus Stop BLADENSBURG RD NE + H ST NE 4
Bus Stop FLORIDA AVE NE + 14TH ST NE 4
Bus Stop 14TH ST NE + H ST NE 4
Bus Stop H ST NE + 14TH ST NE 4
Restaurants w/ Licenses (Bars) 1427 H STREET NE 4
Restaurants w/ Licenses (Bars) 1423 H STREET NE 4
Restaurants w/ Licenses (Bars) 1413 H STREET NE 4
Restaurants w/ Licenses (Bars) 1378 H STREET NE 4
Restaurants w/ Licenses (Bars) 1451 MARYLAND AVENUE NE 4
Restaurants w/ Licenses (Bars) 1421 H STREET NE 4
Hair/Nail Salons 1922 BENNING RD NE # 200 4
Public Housing 2101 G Street NE 4
Bus Stop MT OLIVET RD NE + CORCORAN ST NE 4
Bus Stop MT OLIVET RD NE + CAPITOL AVE NE 4
Bus Stop WEST VIRGINIA AVE NE + 16TH ST NE 4
Liquor Retailer 1625 NEW YORK AVENUE NE 4
Liquor Retailer 2000 WALT LINCOLN WAY NE 4
Check_Cashing 4520 BENNING ROAD SE 4
Bus Stop E CAPITOL ST NE + 47TH ST NE 4
Bus Stop E CAPITOL ST SE + 46TH ST SE 4
Bus Stop BENNING RD SE + B ST SE 4
Bus Stop E CAPITOL ST NE + BENNING RD NE 4
Bus Stop E CAPITOL ST SE + 47TH ST SE 4
Liquor Retailer 5900 GEORGIA AVENUE NW 4
Bus Stop GEORGIA AVE NW + MAIN DR 4
Convenience Stores 1449 HOWARD RD SE 4
Bus Stop STANTON RD SE + BRYAN PL SE 4
Bus Stop MORRIS RD SE + BANGOR ST SE 4
Bus Stop MORRIS RD SE + WEST ST SE 4
Bus Stop MORRIS RD SE + BRYAN PL SE 4
Liquor Retailer 1447 HOWARD ROAD SE 4
Bus Stop MARTIN LUTHER KING JR AVE SE + MORRIS RD SE 4
Bus Stop MARTIN LUTHER KING JR AVE SE + W ST SE 4
Bus Stop MARTIN LUTHER KING JR AVE SE + PLEASANT ST SE 4
Bus Stop MARTIN LUTHER KING JR AVE SE + CHICAGO ST SE 4
Bus Stop GOOD HOPE RD SE + 22ND ST SE 4
Bus Stop GOOD HOPE RD SE + 18TH ST SE 4
Hair/Nail Salons 1313 U ST SE 4
Hair/Nail Salons 1308 GOOD HOPE RD SE 4
Check_Cashing 1245 GOOD HOPE ROAD SE 4
Bus Stop GOOD HOPE RD SE + 13TH ST SE 4
Bus Stop GOOD HOPE RD SE + MARTIN LUTHER KING JR AVE SE 4
Bus Stop ML KING AVE SE + GOOD HOPE RD SE 4
Bus Stop MARTIN LUTHER KING JR AVE SE + U ST SE 4
Restaurants w/ Licenses (Bars) 2004 MARTIN LUTHER KING JR AVENUE SE 4
Restaurants w/ Licenses (Bars) 1922 MARTIN LUTHER KING JR AVENUE SE 4
Liquor Retailer 1303 GOOD HOPE ROAD SE 4
Bus Stop PENNSYLVANIA AVE SE + FORT DAVIS ST SE 4
Bus Stop PENNSYLVANIA AVE SE + 40TH ST SE 4
Liquor Retailer 3851 PENNSYLVANIA AVENUE SE 4
Bus Stop MINNESOTA AVE SE + 30TH ST SE 4
Bus Stop MINNESOTA AVE SE + M ST SE 4
Check_Cashing 2318 4TH STREET NE 4
Bus Stop 4TH ST NE + ADAMS ST NE 4
Bus Stop 4TH ST NE + BRYANT ST NE 4
Bus Stop WASHINGTON HOSPITAL CENTER + OUTPATIENT CLINIC 3
Bus Stop MICHIGAN AVE NW + 1ST ST NW 3
Bus Stop 1ST ST NW + MICHIGAN AVE NW 3
Bus Stop NAYLOR RD SE + PARK NAYLOR APARTMENTS 3
Table 2.3


The 20 locations with the highest predicted maltreatment count are chosen and the total count of abandoned vehicle and illegal dumping 311 calls within a quarter mile is calculated (Figure 2.13). 311 calls for abandoned vehicles and illegal dumping were chosen based on the closeness tests presented below in Section 3.4. These 20 locations are listed by count of 311 calls in Table 2.4.


Risk_Location Address Sum_311_incidents
Bus Stop 14TH ST NW + HARVARD ST NW 270
Bus Stop 14TH ST NW + COLUMBIA RD NW 270
Convenience Stores 3012 14TH ST NW 249
Bus Stop IRVING ST NW + 14TH ST NW 247
Bus Stop COLUMBIA RD NW + 14TH ST NW 244
NonDepository Banks 3018 14TH STREET NW 232
Bus Stop 14TH ST NW + IRVING ST NW 232
Hair/Nail Salons 212 36TH ST SE 225
Restaurants w/ Licenses (Bars) 1400 IRVING STREET NW 211
Bus Stop 37TH ST SE + RIDGE RD SE 194
Public Housing 1475 Columbia Road NW 193
Bus Stop 4TH ST SE + BRANDYWINE ST SE 166
Bus Stop 37TH ST SE + 37TH PL SE 163
Bus Stop 4TH ST SE + CHESAPEAKE ST SE 160
Public Housing 155 Ridge Road SE 151
Bus Stop CONGRESS ST SE + #1326 132
Bus Stop CONGRESS ST SE + 13TH PL SE 118
Bus Stop 13TH ST SE + SAVANNAH ST SE 98
Bus Stop 13TH ST SE + CONGRESS ST SE 95
Bus Stop CONGRESS ST SE + 13TH ST SE 95
Table 2.4

2.10 Gap Analysis

The final component of the Align phase is to relate predicted risk and relative protection for each neighborhood throughout D.C. The gap analysis below does just this, visualizing in shades of red, those neighborhoods where there is more risk than protection. Conversely, shades of green show areas where protection outweighs risk. On the protective side, the gap analysis considers protective locations collected for the modeling process (i.e. police stations, community centers, homeless shelters, etc) as well as Collaborative and Families First DC locations. Future iterations of this map may include more relevant community interventions such as foster homes.

The remainder of this report presents detailed results from the child welfare risk prediction model. Section 3 discusses the Exploratory Data Analysis. Section 4 presents Modeling and Validation. Section 5 concludes. The Technical Appendix provides additional depth on motivation and methods for the framework.

3. Data Preparation and Exploratory Analysis

The geospatial risk prediction framework predicts maltreatment risk using a host of datasets provided by the Washington D.C. Child and Family Services Agency. In this section, approaches for data wrangling and feature engineering are discussed. In addition, Exploratory Data Analysis (EDA) is performed to provide additional context that is useful for understanding child maltreatment in Washington D.C. EDA brings additional hypothesis testing that helps inform the model and provide insight that may be more interpretable to a non-technical policy maker.

There are three types of data collected for this analysis. The first includes the locations of child maltreatment events between 2015 and 2019. The second includes a series of risk and protective factors that may help explain the prevalence of child maltreatment in space. Examples include indicators of blight, childcare centers, and homeless shelters. The third set includes additional information to inform the strategic planning or Align phase discussed above. All data is georeferenced, meaning x and y coordinates locate each event on the surface of Earth.

3.1 Maltreatment patterns across space and time

Beginning with the child maltreatment location data, Figure 3.1 visualizes the count and spatial distribution of maltreatment across D.C. between 2015 and 2019. Although the count of maltreatment has grown slightly during the study period, the spatial distribution has remained consistent.

The underlying hypothesis of this work is that exposure to various risk and protective factors predicts maltreatment risk in space. To test this hypothesis, we develop several approaches to relate maltreatment events and risk/protective factors in space. The first step is to overlay atop Washington D.C. 2,031 polygon grid cells, each with an area of 1000 ft2. The ‘fishnet’, as these grid cells are called, provides a standardized unit of prediction that both conforms to the assumptions of the statistical models and can provide targeted community engagement at a relevant spatial scale. Figure 3.2 visualizes the count of maltreatment events across the fishnet for each year. Figure 3.3 aggregates maltreatment counts for 2017, 2018, and 2019 and illustrates the maltreatment rate per 100 people.3 For more detail on how fishnet grid cell size is determined, please reference the Urban Spatial/PAP Technical Appendix Section 4.2.

Maltreatment events are also summarized by two aggregate areas, neighborhoods and political wards. These are visualized in Figure 3.4 below.

3.2 Risk and protective factors

Twenty-nine different datasets are gathered as risk and protective features. These datasets are collected for 2017, but when 2017 data wasn’t available, 2019 data is used. Crime and 311 call features are created from 2017 and 2018 incidents. The density of risk/protective features are mapped in Figures 3.5 and 3.6 below.

Like maltreatment, each risk/protective factor is joined to the fishnet to test the exposure hypothesis. Risk and protective factors are aggregated to the fishnet using three different approaches:

  1. Sum of risk/protective events per grid cell
  2. The Euclidean distance from the center of each grid cell to the nearest risk/protective event
  3. The average Euclidean distance from the center of each grid cell to the five nearest risk/protective neighbors.

‘Feature engineering’ is the process of converting a variable like ‘homeless shelters’ to statistical information that can more directly help test the exposure hypothesis. Although a given risk/protective factor is engineered into the three different features, only one will be chosen for the final predictive model. Below is an example of the feature engineering approach. Additional context about this process can be found in Section 4.2 of the Urban Spatial/PAP Technical Appendix.

3.3 Do maltreatment events cluster?

The plots of count of maltreatment events by grid cell, neighborhood, and per 100 people all show a higher number of maltreatment events in Southeast DC (Figures 3.2, 3.3, and 3.4) . We further explore the spatial relationship of maltreatment events by testing whether statistically significant clustering of maltreatment exists.

Maltreatment events that ‘cluster’ are those that are closer to one another than what might otherwise be expected due to random chance alone. The Global Moran’s I test indicates clustering of maltreatment events does exist, with a statistic of 0.31 and a p-value of 0.001. The Local Moran’s I analysis provides a hypothesis test of geographic clustering. Figure 3.8 below visualizes statistically significant maltreatment clusters (p <= 0.05) using 2017, 2018, and 2019 maltreatment data. More detail on the Local Moran’s I test can be found in the Urban Spatial/PAP Technical Appendix Section 5.3.

While the maps below provide an engaging indicator of maltreatment activity, there is an important conceptual reason why clustering is important for this analysis. It could be that maltreatment clusters in space because environmental risk and protective factors enable this behavior. It could also be that individuals predisposed to engage in maltreatment sort into communities with these risk and protective features. As discussed in Technical Appendix Section 1, this dynamic referred to as ‘neighborhood effects’ plays an important role in the generation of maltreatment. The models explained in Section 4.2 below include a feature to account for this clustered neighborhood effect.

3.4 Are risk and protective factors ‘close’ to maltreatment?

One way to re-conceive of the notion that exposure to a risk factor is related to maltreatment, is to consider their relative closeness. In other words, are certain risk/protective factors closer to maltreatment than what might otherwise be expected due to random chance alone? Typically, statistical correlation is measured to understand the relationship between risk/protective factors and maltreatment, as we do in Section 3.5 below. However, because the hypothesis relates these phenomena spatially, it is perhaps more reasonable to derive a measure of spatial correlation.

One way to reconceive the notion that exposure to a risk factor is related to maltreatment, is to consider their relative closeness. In other words, are certain risk/protective factors closer to maltreatment than what might otherwise be expected due to random chance alone.

Here is how it works. Imagine a large assembly hall full of primary school children and a set of teachers whose job it is to keep the kids quiet and focused. Assuming the locations of students and teachers were plotted on a map, are the teachers close enough to the students to provide supervision or randomly spread out across the room? To answer this question a test is developed as below.

  1. The observed distance from each student to his or her nearest teacher is measured and the average nearest neighbor distance is recorded for all student/teacher pairs.
  2. The teachers are then randomly relocated across the map, and again the distance from each kid to the nearest teacher is recorded and the average taken across pairs.
  3. Step 2 is repeated 999 times to generate a distribution of randomly generated average nearest neighbor distances.
  4. If the observed nearest neighbor distance is less (i.e. closer) than say, 95% of the randomly generated distances, it is possible to conclude with a p-value of 0.05, that teachers are close to students.

This same procedure is used to test the spatial relationship between risk/protective factors and maltreatment. For more detail, please reference Section 5.4 of the Urban Spatial/PAP Technical Appendix. Results are plotted for protective, crime-related, and other risk factors in Figures 3.9, 3.10, and 3.11 respectively. The vertical black bar measures the observed nearest neighbor distance between a given factor and maltreatment. The colored histogram represents the randomly distributed nearest neighbor distances. The farther the former is from the latter, the more significant the closeness test is.

Homeless shelters, police stations, and food banks are the closest protective factors to maltreatment. Homicide, assault with a deadly weapon, and robbery are the most significant crime-related risk factors. Public housing, abandoned vehicles, and illegal dumping are the three closest non-crime risk factors. The conclusion that emerges from this analysis is that blight is a powerful predictor of maltreatment risk in Washington D.C.

3.5 Pairwise correlations

More than 100 total features are created from this analysis, including 29 risk and protective factors engineered into 87 features, 17 census control features, and a feature controlling for the clustered, neighborhood effects phenomena discussed in Section 3.3. To help determine which features will be the best predictors of maltreatment, pairwise correlations are calculated between maltreatment count for each grid cell (netmal_1718) and a given feature. Figures 3.12 and 3.13 visualize the pairwise correlations for the top 15 most correlative protective and risk factors. Information on pairwise correlations can be found in the Urban Spatial/PAP Technical Appendix Section 5.2.

There are three different prefixes associated with each type of feature (see Section 3.1 above for each further description): NN refers to features calculated by taking the average distance between a fishnet grid cell and its n nearest risk/protective factor neighbor; ed refers to the Euclidean distance between a fishnet grid cell and its one nearest risk/protective factor neighbor; agg refers to the count of risk/protective factor events in a given fishnet grid cell.

A correlation of ‘1’ means a perfect positive correlation between maltreatment and the risk/protective feature. The correlation for the agg_Vio.1718_Abandoned.Vehicle.On.Public.Property feature suggests that as the count of abandoned vehicles increases, there is a noticeable increase in maltreatment count. Conversely, a correlation of ‘-1’ means perfect negative correlation. The correlation for NN_SNAP suggests when distance to SNAP retailers increases (gets farther away) maltreatment count decreases.

4. Predictive Model Estimation and Validation

The following section presents the final risk model and illustrates several validation metrics. For details on the modeling process and the validation tests, please reference the Urban Spatial/PAP Technical Appendix Section 6 and Section 7 respectively.

Exploratory Data Analysis explored both spatial and statistical correlations between maltreatment and individual risk/protective factors. In this section, those features with the greatest pairwise correlations are entered into a a series of predictive regression models to predict maltreatment risk across Washington D.C.

4.1 Accuracy and generalizability

The goal of the predictive modelling process is to create a ‘useful’ tool to help child welfare stakeholders better allocate limited community intervention resources in space. Two indicators are used to judge usefulness - accuracy and generalizability. Accuracy is a measure of error - that there is little difference between the observed count of maltreatment and the predicted count. Generalizability is a measure of consistency - that the accuracy varies little across different areas of the city. While the Technical Appendix provides lengthy context on the accuracy/generalizability trade-off, there are two key concepts to consider.

First, a predictive model that is 100% accurate is not useful. Recall from Section 1 above, that the goal of this analysis is to predict latent risk of maltreatment - that is, places where maltreatment may be occurring, but because it’s a rare event, it goes unreported. Thus, some model error correctly assumes predicted risk is greater than observed maltreatment.

Secondly, a predictive model that does not generalize well to different urban contexts is also not useful. A model that predicts accurately in one type of neighborhood but not another is not generalizable and thus, not a fair tool to guide resource allocation across the entire city. Recall from Section 1, that the intuition of this geospatial risk prediction approach is to borrow the observed maltreatment ‘experience’ in D.C. and test whether that experience generalizes to places where maltreatment may be occurring but is not directly observed. If this experience is generalizable, it is reasonable to confidently forecast maltreatment risk across space. The entire modeling framework, explained below, is based around this idea of a ‘generalizable maltreatment experience.’

4.2 The modeling framework

Maltreatment counts and associated risk/protective features for 2017 and 2018 are joined to the fishnet and used to predict maltreatment in 2019. ‘Training’ the predictive model on two years and ‘testing’ predictions for a subsequent year provides a convenient generalizability test for how useful the resource allocation tool is to the near future.

There are three different predictive algorithms employed - Poisson regression, Spatial Durbin and Random Forest, explained in more depth in Section 6.2 of the Technical Appendix. These three models each pick up on different patterns in the empirical relationship between maltreatment and risk/protective factors. Poisson is a simple Generalized Linear Model; Spatial Durbin is a algorithm uniquely suited for observing spatial relationships; and Random Forest is a powerful, contemporary machine learning algorithm. To capitalize on the nuances of each, an ‘ensemble’ or meta-model approach blends together predictions from all three into a final set of predictions.

In order to test for spatial generalizability - model accuracy across space - further blending is done by partitioning the fishnet into 46 neighborhood groups. Leave-one-group-out cross-validation (LOGOCV) is used to then train the model on n - 1 groups and predict for the hold out. This ensures that predictions for any one place are informed by a general set of experiences from across the city. For more information on LOGOCV, please refer to Section 6.3 of the Urban Spatial/PAP Technical Appendix.

Three outlier grid cells were assigned the mean observed maltreatment count for that cell’s adjacent 8 grid cell neighbors upon consultation with D.C. Child and Family Services Agency. These areas with very high, outlying maltreatment counts contained homeless shelters that have all since closed. As they are no longer operating, the maltreatment experience in those places no longer generalizes.

4.3 Model accuracy

Accuracy is the difference between the observed maltreatment counts across the fishnet and the predicted maltreatment counts. This section presents several measures of model accuracy. Figure 4.1 below plots the predicted count on the y-axis as a function of the observed count on the x-axis for each of the three models used.

In general all models systematically underpredict areas that have very high observed maltreatment counts. There are only a handful of places that have more than a dozen maltreatment events in 2017 and 2018. Given how rare these specific experiences are, it is not surprising that the LOGOCV process yields higher errors.

Figure 4.2 plots the predicted and observed maltreatment counts for the ensembled meta-model or blend of all three submodels. In general, the model fits well for most of the data, again with exception to the underprediction for grid cells with very high counts. Note that places with moderate counts of observed maltreatment events - perhaps 5-10, are actually overpredicted slightly. This is the intent of the model, as these are areas predicting latent risk - predicted maltreatment that is otherwise unreported.

Table 4.1 provides a number of goodness of fit metrics (for more information refer to Technical Appendix Section 7.1). There are two metrics worth interpreting here. First the MAE or Mean Absolute Error is a measure of accuracy and is interpreted as the difference (or error) between the observed and predicted maltreatment count; averaged across all holdout neighborhoods and on an absolute basis (meaning an indifference for over versus underprediction). The MAE for the meta-model suggests that on average, the model errs less than one half of one maltreatment event.

The second metric suitable for interpretation here is the MAE_sd or the standard deviation of the Mean Absolute Error across each neighborhood holdout. A relatively high value suggests that the model errs differently across different communities. A low value suggests errors are consistent across different neighborhoods. This is a useful measure of generalizability and one quite robust for this analysis.

Model Name R2_mean R2_sd MAE_mean MAE_sd RMSE_mean RMSE_sd logdev_mean logdev_sd
GLM - Poisson 0.534 0.312 0.526 0.536 0.937 0.878 0.688 0.230
Meta-Model 0.549 0.324 0.475 0.468 0.824 0.741 0.699 0.221
Random Forest 0.523 0.330 0.548 0.502 0.902 0.826 0.650 0.225
Spatial Durbin - sqrt 0.728 NaN 0.454 NaN 1.166 NaN 0.728 NaN
Table 4.1

Figure 4.3 below maps geospatial risk predictions and MAEs for each model type and the meta-model. These maps make it clear that again, each of the individual models are picking up different relationships in the data.

Figure 4.4 below visualizes the risk predictions and MAE for only the meta-model with national parks and hydrology masked. Several key neighborhoods are predicted to be at high risk including Anacostia and Columbia Heights. In general, higher errors appear where there are higher maltreatment counts, but errors are mostly diffused throughout the city.

Figure 4.5 takes the mean error by neighborhood.

4.4 Generalizability

Generalizability is the ability for the model to predict with consistent accuracy across different urban contexts. In this section, several tests for generalizability are provided. To begin, Washington D.C. is split into high and low areas with respect to non-white and poverty. These typologies are visualized in Figure 4.6 below.

Recall from Table 4.1 above that the overall MAE for the Meta-Model is 0.475. Figure 4.7 calculates average MAE by neighborhoods for the aforementioned typologies. While areas with larger minority populations and higher poverty levels have higher prediction errors, they also happen to have far higher observed maltreatment counts. Areas with fewer non-white residents and less poverty have observed maltreatment counts of 16 and 17, respectively. Areas with more non-white population and more poverty have observed maltreatment counts of 65 and 64, respectively. We therefore feel the model is generalizable, meaning it is useful as a resource allocation tool. Further, the difference in MAE between these typologies is driven by the large difference in maltreatment occurance.

4.5 Feature importance

To better understand which features had the greatest contribution to predicting child maltreatment, Variable Importance analytics from the random forest sub-model is plotted in Figure 4.8.

The most important predictor is ‘NN_ScreenedIn’ which is a feature created to account for the local clustering of maltreatment events, also referred to as ‘neighborhood effects’ in Section 3.3. Other features found important in the modeling process include a fixed effect (i.e. dummy variable) to control for each neighborhood, crime variables (assaults with dangerous weapons and sexual abuse crimes), as well as variables that indicate neighborhood blight (311 calls for abandoned vehicles and vacant homes).

4.6 Does the risk prediction model outperform standard hot spot maps?

Kernel density is the traditional approach for creating hot spots maps used for targeting interventions in space. In this section, the usefulness of the meta-model as a resource allocation tool is tested relative to kernel density. For more information, please refer to to Section 7.3 in the Technical Appendix. As a further test of temporal generalizability, the meta-model, trained on 2017 and 2018 data, is tested against 2019 maltreatment data.

To begin, the kernel density ‘predictions’ and meta-model predictions are both grouped into five risk categories, in ascending order of risk. Figure 4.9 below visualize the five predicted risk categories with 2019 maltreatment points overlaid, perturbed slightly to maintain privacy. The maps make it clear first, that the 2017/2018 predictions generalize to the 2019 maltreatment experience and second, that the meta-model provides a much more accurate resource allocation tool. It is also worth noting that other areas show up as high risk areas despite containing few or no 2019 maltreatment events. These are areas identified by the model as having latent risk for maltreatment.

If the meta-model predictions are more useful than the kernel density, the former should capture a greater proportion of 2019 maltreatment events in the highest risk category relative to the latter. The bar plot in Figure 4.10 suggests this is true, reaffirming the message conveyed in Figure 4.9. The meta-model provides a useful tool for allocating community interventions in space.

5. Conclusion

Between 2017 and 2019, Washington, D.C. had 2,763 confirmed cases of child maltreatment, with many incidents clustering in Southeast D.C. The goal of the geospatial risk prediction model developed in this report is to help Washington D.C. optimally allocate limited child welfare resources where they are most needed. This is done by hypothesizing that maltreatment risk is a function exposure to geospatial risk and protective factors. Many of the important risk factors responsible for predicting this behavior include blight, crime, and the neighborhood effect, which helps account for the fact that maltreatment tends to cluster in space.

Section 4 presents the predictive model, which not only accurately predicts areas with known maltreatment as high risk, but predicts the latent risk - places where maltreatment may be occurring but is not reported. The model provides more useful predictions relative to the more traditional kernel density approach.

The Align phase converts risk predictions into actionable intelligence by providing a simple to use strategic planning tool that stakeholders can use to allocate community intervention resources. Given the open source nature of this tool, it can be updated for free annually to analyze change in maltreatment risk citywide.




  1. The Align Phase (Section 2) was created using the masked meta-model predictions. As a result of the mask, a marginal amount of the total population was removed.↩

  2. The Child and Family Services Agency sent the addresses of the Collaborative and Families First DC locations to Urban Spatial. From this data, one address was missing, one center had multiple addresses listed from which the first address was used in geocoding, and one address could not be geocoded (returned no x/y coordinates).↩

  3. This table presents summary statistics of maltreatment events for 2017, 2018, 2019.
    year total_count max_count_byGridCell min_count_byGridCell avg_count_byGridCell
    2017 892 26 0 0.44
    2018 923 23 0 0.45
    2019 948 14 0 0.44
    ↩