Connecting the Underemployed with Job Opportunities
- Point of contact: Darron Fuller (darron.fuller@DAEN-Society.org)
- Twitter: @da_engineers
- As of Feb. 26, 2016, Washington, DC’s unemployment rate was 6.6%, 48th in the nation (Source: BLS).
- Idea: Connect skilled workers of high unemployment/low economic geographic areas (census tracts) to opportunities in areas of high-opportunity.
- Take-away: Unemployed workers may need to travel long distances to obtain work in high opportunity areas. This project pairs workers with jobs in order to maximize income after commuting expense. The results should inform transit decisions for DC.
- Gathered data on job locations by census tract (compiled by aggregating census block (a subset of each tract) data) located at https://celebratingcities.data.socrata.com/Census/LEHD-LODES-7-Workplace-Area-Characteristic-DC-2013/6ezg-vkmc
- Gathered unemployment data by census tract from http://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?pid=ACS_13_5YR_S2301&prodType=table
- Used this info to identify high underutilization tracts (unemployment >= 10%) and high opportunity tracts (areas where there are more jobs than people living in the area).
- We passed this info along to our Operations Research/Prescriptive Analytics team to run their optimization model built in Gurobi Optimizer to determine the optimal job paths with the goal of maximizing profit to underemployed/unemployed tracts (i.e., Profit = Expected Revenue from job less the Estimated cost for transportation and Estimated travel time * Expected wage rate). They used number of people unemployed in a tract as the source number and number of available jobs (computed using 6% (our estimate of the national unemployment rate) times the number of jobs, which is a really rough estimate).
Visualization link: http://rvisio.github.io/index.html : Green = High Opportunity (HO) Tract. Red = High Unemployment (HU) Tract. Yellow = Satisfied conditions for being both High Opportunity and High Unemployment tract (OU). Caution on use of this Visualization: The model for identifying High Opportunity Tracts and High Underemployment Tracts utilizes a rough approximation and for demonstration purposes only. For example, the model tends to identify college campuses for high unemployment while students are unemployed and not currently seeking employment. Please let us know if you have expertise in social sciences/employment policy or a discipline that could assist with refining the model for more realistic predictions.
- Another project is to find information about the number of people with given skills (by location) and the number of people working in an area with given skills (by location). This is relevant and important but not doable in the time frame.
- Other helpful links:
- DC Jobs: https://www.dcnetworks.org/vosnet/Default.aspx
- For future reference, here’s a similar project by brookings: http://www.brookings.edu/~/media/research/files/reports/2015/03/24-job-proximity/srvy_jobsproximity.pdf
- Unemployment by census tract: http://does.dc.gov/sites/default/files/dc/sites/does/page_content/attachments/2015%20Unemployment%20Rate%20by%20Ward_6.pdf
- American Community Survey Census Dataset
- General programming and Data Preparation : Python w/ Pandas, re, and NumPy
- Exploratory Data Analysis : Tableau / Excel
- Route Distance Calculations for Walk, Transit, Car : Google DistanceMatrix API
- Route Distance Calculations for Ride-sharing Service : Uber API
- Map Visualization: MapBox API
- Linear Programming Optimization Model / Prescriptive Analytics : Gurobi Optimizer
- We came close to completing the model today but ran out of time implementing this ambitious plan. Though we managed to get most of it completed in the few hours we had organize and collaborate around this idea. Reaching our goal is a matter of a few more lines of code to complete implementation of the cost matrix (i.e., cost to travel from tract-to-tract) that is used in the transportation optimization model. We will post to GitHub once completed and provide a link here on Hack-Pad. Please send an eMail request to darron.fuller@DAEN-Society.org if you would like to be contacted once the model has been posted to GitHub.
- Instead of using a simple filter to select tracts that were considered underemployed and tracts where job opportunity is high, we are planning to build a predictive model for selection of the targeted tracts of interest to feed into the transportation optimization model. This predictive model could identify areas trending toward or at-risk of underemployment that would otherwise not be identified using the standard definition for underemployment. A similar predictive model can also be used to proactively identify areas that are fertile grounds for businesses and job growth so that DC may also be proactive in their policy-making decisions.
If the data is available, we would like to improve the model to support job and opportunity matching at the skills level and have more detail of specific jobs. This approach should result in a better "fit" between tracts (or blocks) of low-employment and tracts offering opportunity and provide an enhanced decision management tool for city officials. Furthermore this profit-maximizing framework can be extended down to the automated decisioning of matching job seekers directly to employers with available jobs. Update 4/20/16: Began working with the American Time Use Survey (ATUS) which asks respondents to provide details into their daily activities as well as their employment status. The ATUS has specific questions regarding Employment and Underemployment and also could be used to better understand the skills of the respondents based upon the data of their logged activities.
- We were approached by representatives of both DC Government and a Federal agency with requests to continue discussions on further development of this model. Such a partnership will facilitate the enhancement of model assumptions and scoring functions to support a more realistic scenario that can prescribe accurate solutions to the very real problem of underemployment and shed light on the impact of transportation options on access to jobs.
- People that contributed: Darron Fuller ( MSc. GMU/VSE), Will Will Traves (USNA), Kate Riesbeck, Greg Elin (GovReady), Rob Jarvis ( MSc. GMU/VSE), Alex Lyte (MSc. GMU/VSE), Bob Kraig (MSc. GMU/VSE), Jonathan Pleban (Socrata), Mike Delgado (ATU.org), Muhammad Imran (MSc. GMU/VSE) for moral support, Kara Turner (Socrata).