The city population estimates included in the city boundary shapefiles are calculated by applying mortality and migration rates to the 2010 Census count and controlling for age, race-ethnicity, and gender proportions from the Census Bureau’s LA county population estimates for the previous year. These city boundary shapefiles are available for download and public use at the County of Los Angeles GIS Portal . To quantify the impact of multiple dispensaries being located near a school, I first calculated the association between the continuous distance between the school and the nearest dispensary in within a mile and within LA County. I wanted to know at which point a dispensary was located close enough to a school to have an influence on student marijuana use, so I also conducted sensitivity tests of distance within a mile using increments of a quarter mile. These distances are much further away from schools than the state requirement of 500 feet or and the maximum distance dispensaries are required to be located away from schools by a city ordinance in LA County, which is 1,000 feet. I then constructed a series of “buffers” using ArcMap 10.4 GIS Software and recorded counts of how many dispensaries were located within each buffer. A buffer is created by specifying the length and unit of measurement for the radius around a point of interest, such as LA County public high schools. A series of 3 buffers were created for this analysis. The first buffer was 500 feet in radius; the minimum distance a city in LA County allowed dispensaries to be located near schools in 2016,cannabis drying when the city policy data was collected. I suspected that dispensaries could have an impact at greater distances from schools than at 500 feet so I tested the impact of dispensaries being located with 1,000 feet and 2,000 feet. The dispensary count within 500, 1,000, and 2,000 feet of each school were imported into SAS and matched with the data for each school that participated in the CHKS survey by CDS code.
This allowed for information about student marijuana use to be associated with the number of dispensaries within a specific radius of each school. These buffer counts were then used as independent variables in the multilevel logistic regression analyses to determine the impact of the number of dispensaries near the schools on students’ marijuana use behavior. Student characteristics assessed include gender, ethnicity , race, grade , highest level of parent education, whether the student qualified for free or reduced-price meals, and whether the student attended their school’s after school program at least one day a week. Male gender is sometimes associated with greater likelihood of and higher rates of marijuana use , whereas female gender has been associated with lower rates of use overall, but with younger ages of initiation and faster transition to regular use . Some studies have found that rates of marijuana use among people of Latino ethnicity are higher relative to other racial/ethnic groups in early adolescence but are often overtaken by rates of use by white people in later adolescence .Therefore, the analyses presented in this dissertation use the students’ grade in school as a measure of student’s age. Older age is almost universally correlated with greater substance use among adolescents , so age is an important factor to account for in any analysis of the risk of substance use among high school age youth. The analyses presented in this dissertation are based on students in the 9th and 11th grade, per CHKS study protocol. Higher grade is logically a powerful predictor of lifetime marijuana use due to it being determined by greater age, but has also been shown to be associated with a greater likelihood of recent marijuana use , which is not necessarily dependent on greater age. Participation in after-school programs was included as a covariate because it has been shown to be a protective factor against adolescent substance use in general .
The count of days each student participated in after school programs was only used in the cross-sectional analyses, as it was not available for all of the school years between the 2005/2006 and 2016/2017 school years. Eligibility for school meals and highest parent education were included as a measure of social economic status because some studies have found higher SES to be associated with greater rates of marijuana use . Self-report of receiving free and reduced-price school meals was included as the only available proxy for low family income, based on California State eligibility criteria, e.g., annual income $ 32,630 for a family of four ,” n.d.). The school meals variable was ultimately found to have a high rate of “don’t know” responses , which were grouped with “no” responses using the logic that the student would likely be receiving free-reduced price meals if they were eligible and therefore would be aware of their eligibility. This variable was only used in the cross-sectional analyses, as it was not available for all of the school years between the 2005/2006 and 2016/2017 school years. After-school program participation was operationalized using a variable in CHKS that asked “How many days a week do you usually go to your school’s after school program?” and had ordinal response categories ranging from 0 – 5 days per school week. The ordinal form of this variable was used as a covariate to account for how many days a week the student spent time at an after-school program in the regression analysis. This variable was only used in the cross-sectional analyses, as it was not available for all of the school years between the 2005/2006 and 2016/2017 school years.An indicator variable for non-traditional schools was available in the CHKS dataset and a matched more detailed descriptions of school type from the CA Department of Education School Directory. The non-traditional school indicator variable was included in all of the cross-sectional analyses to account for the expectation that students attending non-traditional schools may be more likely to likely to use marijuana. This variable was included in the trend analysis as it was available for all of the school years between the 2005/2006 and 2016/2017 school years.
The number of dispensaries within 500 feet, 1,000 feet, and 2400 was initially used to measure the density of dispensaries near the students’ schools. I defined “near” as 2,000 feet, which was quadruple the distance of 500 feet that the State of California currently requires marijuana businesses to be located from schools . The 600-foot distance from schools set by the State may be rather arbitrary, however, as no existing research has established the distance threshold at which dispensaries no longer influence students’ marijuana use. Some the LA County cities that allowed dispensaries specified that they be located greater distances from schools, such as 1,000 feet,grow trays but it is similarly unknown whether these requirements place dispensaries sufficiently far enough away from schools to prevent them from having an impact on rates of lifetime and recent marijuana use among high school students. My preliminary analyses for the trend analysis indicated that I needed to revise the first hypothesis for Research Question 1 . While testing the parallel trends assumption for the difference in difference analysis, I compared frequencies by time for both lifetime and recent marijuana use by whether the city the school was located in allowed dispensaries. Figures 4.1 and 4.2 indicate that the intervention and control groups exhibited remarkably similar trends, where lifetime and recent marijuana use increased in both groups from baseline through the 2011-2013 combined school years and was followed by a decline that was maintained through the 2015-2017 school years. The evidence of similar trends between the intervention and control groups satisfied a key assumption of difference-in-difference analyses that trends in the outcomes under study were parallel between the intervention and control groups before an event of interest has occurred. However, the similar and non-linear nature of the trends in each group indicated a need to investigate if any events had occurred in LA County that could have influenced cities that allowed dispensaries and cities that did not in similar ways. After learning more about Proposition D and the impact it had on the medical marijuana market in the City of Los Angeles it was clear that Proposition D represented a significant event that affected the intervention group and not the control group. I felt that making any conclusions about trends in marijuana use differing between cities that allowed dispensaries and those that didn’t within LA County without accounting for the impact of Proposition D on Los Angeles students would be invalid. It was less clear whether the federal raids that occurred in 2011 and 2012 affected one of the study groups more than the other, but if it did affect both groups equally, the difference-indifference study design would account for any impact the federal raids had on the marijuana use behaviors of Los Angeles students. I therefore chose to address Research Question 1 by analyzing the impact of enacting stricter regulations on dispensaries students’ marijuana use within the City of Los Angeles, using the cities that had never allowed dispensaries as a control group. Research Question 1 was therefore revised to ask “Do city restrictions on dispensaries have an influence on trends in adolescent marijuana use time?” The revised hypothesis for this question was that cities that enacted more restrictive MMDS policies would see a trend of declining marijuana use among students attending school there . To focus on the impact of Proposition D on trends in student marijuana use in the City of Los Angeles, I excluded the 2005/2007 combined school years and used 2007/2009 as the baseline time period. The 2007/2009 time period was two time periods before Proposition D was enacted and the 2015-2017 time period concluded one time period after the enactment of Proposition D. The analysis plan for Research Question 1 was changed to focus on the impact of Proposition D within the City of Los Angeles compared to cities that did not allow dispensaries . The control group for this analysis includes the 436,834 students that attended school in the 70 LA County cities that had dispensaries bans in place throughout the study period. The cities excluded were cities that had changed dispensary policies between the 2005/2006 school year and the 2016/2017 school year, which excluded cases from the cities of Diamond Bar , Huntington Park , Long Beach , Malibu , Santa Monica , South El Monte , and West Hollywood , and students from schools that could not be matched to CA Dept of Education addresses . City of Los Angeles students were chosen as the intervention group because Los Angeles and the City of West Hollywood were the only cities in LA County that allowed storefront dispensaries to operate within their borders for the entire 12-year study period. West Hollywood schools, however, did not participate in the CHKS survey during the study period and therefore could not be included in an intervention group of cities that allowed dispensaries throughout the study period. Using students who attended school in the City of Los Angeles as the intervention group was preferable for the difference-in-difference analysis of marijuana use trends because data was available for City of Los Angeles schools for every year of the study period and the population of students within this large and diverse city mirrored the population of the County as a whole for most racial/ethnic categories. Exceptions were that City of Los Angeles students were more likely to be Hispanic and less likely to be Asian or White than the control cities . The association between policy changes and subsequent outcomes is often evaluated by pre-post assessments, where outcomes after implementation of the policy are compared with conditions and outcomes from before. This design is valid only if there are no underlying time dependent trends in outcomes unrelated to the policy change . If, for example, outcomes were already improving before the policy was enacted, using a pre-post study would lead to the erroneous conclusion that the policy was associated with better outcomes. The difference-in-difference study design addresses this problem by using a comparison group that is experiencing the same trends but is not exposed to the policy change .