Feedback3

Currently, I work as a marketing researcher for the Stockton Shelter for the Homeless in California. The shelter is expanding its services to further assist the local community by transitioning to becoming a homeless navigation center. A homeless navigation center differs from a traditional homeless shelter as it is classified as “low-barrier” with the end goal of transitioning clients into permanent housing and, more importantly, has a referral process that clients must go through in order to enter the program. This means that the center’s outreach team must select, in the Stockton Center’s case, 180 individuals to refer. Those who come to the center without being referred will, unfortunately, be turned away as the center cannot feasibly assist the entire homeless community at once. My coworkers and I are faced with the problem of having to determine which individuals best qualify for and should be referred to the program based on their likelihood to successfully exit into permanent housing

This problem can be solved through the use of predictive modeling to determine which clients are the most likely to successfully exit the program. Given that I have a clear target variable that is categorical– the clients that are most likely to succeed– I would use supervised segmentation. To do this, I can develop a decision induction tree to determine which attributes are of the most important in determining who will be successful in the program and subsequently identify which specific individuals are the most likely to be successful. This is valuable as the decision induction tree can provide me with the probability of an individual succeeding in the program based on their classification within a set of attributes. Once I have this data, my coworkers and I can then use the data to determine which 180 individuals to refer to the program. The same model and process can be used to select the next set of individuals to refer to the program, adding value to the Stockton Navigation Center by minimizing the guessing and error that may occur when choosing who to refer; thus, setting the program up for long-term success. The program’s success is particularly important as it determines how much governmental funding it will receive each year to continue assisting the housing insecure.

Five attributes that will help my predictive modeling include the employment status of the individual, their highest level of education completed, the length of their homelessness, their history of substance abuse, and their risk of being harmed by others. The employment status is an important attribute that would be useful for the classification model as it would separate those who are currently employed (part-time or full-time) from those who are not. This is essential as our previous research has indicated that those who are ‘successful’ in similar programs commonly have some form of employment. The education level is an important attribute as it would separate those with a lower education level (middle school or less) from those with a higher education level (high school or higher). This is vital as our previous research on similar programs indicates that those with an education level of high school or higher are more likely to be successful in the program and achieve permanent housing. The length of homelessness would identify the individuals whether an individual is recently or chronically homeless, which is essential to know as previous research indicates that those who are successful in programs alike are typically newly homeless. The history of an individual’s substance abuse is important as it helps determines what risks an individual carries as well as what additional services they will need in order to obtain and maintain permanent housing (i.e. support groups, rehab, etc.). Results of similar programs indicate that those without a history of substance abuse typically perform better in navigation centers. Lastly, an individual’s risk of being harmed by others is an important attribute as the center desires to help individuals reach safety. Our current research indicates that those facing domestic violence or are fleeing someone who wants to hurt them are in specific need of the center’s services and are at a high risk of becoming chronically homeless. If an individual is at risk of becoming chronically homeless, it may potentially be more difficult to help transition to permanent housing in the future.

The data for these attributes can be obtained by having our outreach team survey the current homeless shelter attendees, where the results would then be represented on the classification decision tree model to determine which attributes supply the highest information gain and which group of individuals would be the most successful for the program. Using this data, my colleagues and I would be able to select specific individuals to refer to the program based on their likelihood to be successful in the program.