Data Analysis
Harm Reduction – Executive Summary
Since 2003, the AIDS Network Kootenay Outreach and Support Society (ANKORS), now partnered with Interior Health and the British Columbia Centre on Substance Use, has been offering free drug checks to attendees at music festivals like the Shambhala and Bass Coast held in British Columbia, Canada. Drug checking is a free harm reduction service that allows service users to bring drugs and have their compositions checked using technologies such as Fourier-transform infrared spectroscopy (FTIR). Service users are then shown the results, allowing them to make informed decisions regarding the drug’s use. To understand the impact of such services, ANKORS created survey sheets for service users, technicians, and advisors to record information during their transactions. Data has been collected regarding where drugs are obtained, what service users think they are, whether or not the FTIR results concord, and what the service user then determines to do with the drug.
Based on the studied dataset, some of the key findings of the data analysis are:
● It was discovered that the percentage of new service users accessing the drug checking services increased from 52% in 2018 to 70% in 2019.
● The two drugs most commonly brought by the same service user during the same visit at Shambhala are MDMA and LDS, and the most common drug additive is one of many possible forms of sugar.
● There is no statistically significant difference between what a service user believes a drug to be and the spectroscopy analysis.
Researchers also conducted unsupervised learning to look for patterns, specifically utilizing various clustering algorithms to draw inferences from the data. Additionally, various ensemble and boosting algorithms were conducted to predict whether or not a service user will discard their drug as well as to predict what service users will decide to do with a drug after getting the results of the drug check.
The results of the exploratory data analysis were clear enough, but the results of both the supervised and unsupervised learning approaches were inconclusive. Further research could involve experimenting more with variable selection, aggregating multiple datasets over many years, or asking more specific questions on the survey sheets, and suggestions are made for data collection that would facilitate this future research. An appendix is included that explains, to the general reader, the supervised and unsupervised techniques that were utilized.
To obtain a copy of the full report, email mikelabun@gmail.com.