# Evaluating the Use of Google Street View to Visually Verify the Locations of Cannabis Retailers in the United States Extracted from Websites, 2015–2018

## TL;DR

This study explores using Google Street View to verify the locations of cannabis retailers in the US from 2015 to 2018, finding it useful for current verification but not for historical data.

## Contribution

The study introduces the use of Google Street View for verifying cannabis retailer locations and highlights its limitations for historical data reconstruction.

## Key findings

- 74% of cannabis retailers were visually verified at least once between 2015 and 2018.
- Verification rates were low before 2017 due to missing Google Street View images.
- GSV is suitable for current verification but not for reconstructing historical retailer lists.

## Abstract

Our ability to advance public health and policy responses to cannabis legalization is limited by a lack of geographic data on cannabis retailers across states and over time. This study evaluated the feasibility and utility of using Google Street View (GSV) to: 1) visually verify cannabis retailers locations extracted from websites and (2) create a historical retailer list. We extracted and deduplicated cannabis retailer addresses from 6 websites advertising medical and adult use cannabis retailers across 13 US states in February 2019. We visually verified the locations of cannabis retailers in 2015, 2016, 2017 and 2018, summarizing results by state and year. We assessed proportions of verified retailers within and across the legislative period in which states first legalized medical cannabis (ballot [1996–1999], early [2000–2008], late [2009–2018]) using a Chi-square test. Of 1,739 retailers, 74% were visually verified at least one year between 2015 and 2018. Verification was ≥ 74% in all but California and Florida in 2018. Visual verification prior to 2017 was low, largely due to missing GSV images. Proportions of visually verified retailers varied between states legalizing medical cannabis in the ballot (p < 0.001), but not in the early (p = 0.236) and late legislative periods (p = 0.585). GSV virtual audit methods may be appropriate for visually verifying cannabis retailers contemporaneously, but are not appropriate for reconstructing historic retailer lists. Some states (e.g., California) may require the use of additional verification methods. Future studies should assess the feasibility of using GSV in conjunction with novel methods for processing data (e.g., machine learning).