Daytime Population Estimation and its Effect on Risk Terrain Modeling of Crime

Daytime Population Estimation and its Effect on Risk Terrain Modeling of Crime

Now in its third year, Azavea’s Summer of Maps Program has become an important resource for non-profits and student GIS analysts alike.  Non-profits receive pro bono spatial analysis work that can enhance their business decision-making processes and programmatic activities, while students benefit from Azavea mentors’ experience and expertise.   This year, three fellows worked on projects for six organizations that spanned a variety of topics and geographic regions.  This blog series documents some of their accomplishments and challenges during their fellowship.  Our 2014 sponsors, GoogleEsri and PennDesign helped make this program possible.  For more information about the program, please fill out the form on the Summer of Maps website.


When using Census data for research or analysis, sometimes the standard total population count for a region just doesn’t suffice. Transportation planners and crime analysts, for example, must account not only for residential populations but also “daytime” or “commuter-adjusted” population data, since many people spend most of their days working or running errands in different Census tracts, different towns, or even different regions from their homes. Nowadays we’re always on the go, so shouldn’t our population data reflect that?

I encountered this daytime population issue this summer while working as a Summer of Maps fellow with DataHaven to analyze the geographies of crime risk in New Haven, Connecticut. In this project, we used the Risk Terrain Modeling Diagnostics (RTMDx) Utility, a software application that uses crime data and the spatial influences of potential “risk factors” to model where conditions are ripe for crimes to occur in the future. One of the influencing factors of crime we used in our analysis was population density. While the exact effect of population density on crime rates is the focus of ongoing criminology research, in our study, we proposed that crimes would occur where high volumes of people were located. Since our focus here was on where people are “located” and not necessarily where they “live,” we incorporated commuter-adjusted population estimates to account for New Haven’s daytime population.

To acquire daytime population estimates, some data assembly is required. The Census Bureau provides instructions on calculating daytime population estimates using American Community Survey or Census 2000 population data. My first step in calculating daytime population was to download workplace geography data from Census Transportation Planning Products (CTPP), which includes Census data particularly useful to transportation planners. I selected “2006-2010 CTPP Tract-Tract work flow data” and followed the download instructions to get tract-to-tract population flow counts from residences to workplaces. I then queried the Access database to extract all records including a residence tract OR a workplace tract located in Connecticut to account for interstate commuting. With these statewide commuter counts, I was able to hone in on New Haven Census tracts and calculate the total number of workers working and/or living in New Haven. Lastly, I used the Census Bureau’s “Method 2” for calculating daytime population:

Total resident population + Total workers working in area – Total workers living in area

With both resident and commuter-adjusted population counts available, the next stage of the analysis was to incorporate this data into the RTMDx. I created risk terrain surfaces across four crime types (robbery, burglary, simple assault, and assault with a dangerous weapon) and two population counts (resident and daytime populations), producing eight risk maps in total. Each risk terrain model (RTM) included five risk factors in addition to the population count: foreclosures, bus stops, schools, parks, and job locations related to retail, entertainment, and food service (provided by Census LODES data via DataHaven).

In the figures below, we can compare the different risk terrain surfaces created for assaults using resident population and daytime population. The risk terrain surfaces are displayed with the “relative risk score” produced by the RTMDx Utility. To interpret this map, if an area has a risk value of 50, this means that the expected rate of assault in this location is 50 times higher than an area with a score of 1. The higher the score, the greater the risk of an assault based on the model.




In comparing the geographies of crime risk between resident and daytime population counts in central New Haven, we see generally higher risk scores when resident population is modeled. The heavily residential neighborhoods surrounding downtown New Haven see greater risk scores with resident population, perhaps owing to the fact that many residents here commute to jobs in other neighborhoods or cities during the day. Alternatively many of these neighborhoods, including Fair Haven, Newhallville, and Edgewood, see sharply increased risk scores when resident population is considered. The effect of population is more difficult to gauge in downtown New Haven, which is dominated by Yale University and Yale-New Haven hospital, the city’s two largest employers. Despite a much larger daytime than resident population, assault risk scores decreased when accounting for daytime population. This could be due to the nature of assault crimes in relation to population density, the geography of assault incidents in our crime dataset, the role of uncounted university students in influencing assault patterns, or other issues. Our results demonstrate that while daytime population is an important element to consider in risk terrain modeling, crime risk analysis remains a complex and inexact science.

While some spatial analyses may not require the granularity of daytime population estimates, using commuter-adjusted population data has important implications when exploring time-sensitive phenomena like crime or transportation dynamics. The Census may not be able to account for population spikes associated with university students, tourism or shopping, CTPP data still gets closer to understanding where people spend their days outside of the home.