We had some trouble starting off on this project because of our initial belief that this problem was an optimization one. Originally, we approached this project by identifying characteristics of locations that would influence attendance and then attempted to find an optimal location using the characteristics we identified. Unfortunately, attendance to a program meant to help students address mental health cannot be simplified down to characteristics of a location. Instead, we took a step back and reevaluated our problem. We decided to address the problem instead through queuing theory because of a shift in our consideration of attendance to Let’s Talk being a function of other variables to one that varies with probabilities assigned to students. Our model allows us to easily modify the characteristics of students. We can see how changing a location could affect attendance based on the changing population that is served by that location. The model also accounts for differences in mental health status caused by gender, age, and school which are often defining characteristics of mental health status. The limitations of the model and methods we utilized are fairly obvious – there are no hard and fast rules for determining utilization of a mental health resource, so any assumptions made should be considered fluid and subject to change with environmental changes. For example, if the stigma surrounding mental health were to decrease, we might see an increase in utilization of resources or we could see a decrease in the need for the resource (further survey data would be needed to help analyze this).
Another issue we encountered was when assessing location accessibility for the Center for Diversity and Inclusion (CDI). When creating our survey to assess location accessibility, we did not distinguish between the CDI and the general library. From experience, it is clear that the attendance at the CDI is very distinct from attendance to the library as a whole. When the simulation was run with the willingness to attend numbers from the survey we administered, the results were very off and not at all accurate or representative of reality. Since we could not administer another survey and because it was clear that we had made an error in assuming similar willingness probabilities, we contacted the CDI and were told that attendance ranges from 8-10 people per day. Based off this range, we can assume that there is a pool of 100 times that many people who would be willing to come to this location. We divided that pool of total students who would be willing to go to the CDI by the total student body and determined the probability that any given student would frequent the CDI given that they have a need for Let’s Talk. This pool of 1000 students who would be willing to come to the CDI also corresponded with a 75% decrease in the number of people who said they would be willing to attend a Let’s Talk session anywhere in Olin.
Because of the way that we assigned location preference as a constant for all populations to the CDI, we decided against doing a sensitivity analysis on this location because it would not reflect any important differences.
We also came across challenges when trying to identify which program to use. We began by thinking that we should use Matlab or Simulink to create a series of variables that could be changed and controlled and running them through a series of filters to get to our final pool of Let’s Talk attendees. Unfortunately, Matlab and Simulink are not intuitive programs in this situation and one of our group members identified Simscript as an alternative. We spoke with multiple different professors about potential programs that we might be able to use to simulate this situation, however many of those programs were unavailable to use or the wrong simulation type for our system.