Abstract
Background and Aims
Epidemiological trends show marijuana use in the U.S. to have increased in recent years. Previous research has identified cohort effects as contributing to the rising prevalence, in particular birth cohorts born after 1945. However, given recent policy efforts to regulate marijuana use at the state level, period effects could also play a contributing role. This study aims to examine whether cohort or period effects play a larger role in explaining trends in marijuana use.
Design
Using data from seven National Alcohol Surveys, we estimate age-period cohort decomposition models for marijuana use controlling for socio-demographic measures.
Setting
United States
Participants
U.S. general population ages 18 and older from 1984 to 2015.
Measurements
Any past year marijuana use
Findings
Results indicate that period effects are the main driver of rising marijuana use prevalence. Models including indicators of medical and recreational marijuana policies do not find any significant positive impacts.
Conclusions
The steep rise in marijuana use in the United States since 2005 occurred across the population and is attributable to general period effects not specifically linked to the liberalization of marijuana policies in some states.
Keywords: marijuana, trends, age-period-cohort, marijuana policy
INTRODUCTION
Marijuana use in the U.S. has risen steeply since 2005 across all age groups (1) in contrast to negative trends for tobacco use and alcohol use among underage youth (2). In particular, age-specific rates have jumped among those aged 50 and older from very low rates among earlier birth cohorts to high rates among the baby boomer birth cohort (3). Our 2007 study of marijuana age-period-cohort (APC) effects in the National Alcohol Survey (NAS) series found declining marijuana use from 1984 to 2000, mainly by men, and strong cohort effects differentiating those born before 1945, who have very low marijuana use, from those born after 1945 (4). One subsequent marijuana APC analysis from 1985 to 2009 confirmed these cohort effects and found that increasing use from 2005 to 2009 was due to period effects (5). These findings from earlier APC studies indicate the need to track groups that are at highest risk for increased marijuana use.
Changing perceptions of marijuana in regard to riskiness for health and social problems and social acceptability have played a role in marijuana use (6, 7). One study examining youth use and social norms in a birth cohort framework found a connection between cohort disapproval of marijuana use and lower marijuana use (8). In another APC study, period effects were found such that perceived harmfulness of marijuana among youth has been decreasing since 1991, and in particular, in states that passed medical marijuana laws (7). Public support for marijuana legalization has been increasing over time, and especially among younger generations (9, 10). In one APC study on marijuana legalization in the U.S. from 1968 to 2015, support for legalization was linked to perceptions of marijuana safety and that trends in support were primarily the result of period effects (11).
Passage of medical marijuana legislation, in particular allowing medical marijuana dispensaries or home growing, under varying policy regimes since 1996 are potentially relevant to marijuana use trends (12). As of early 2017, 29 states and Washington D.C. have laws allowing medical marijuana use, with some states having particular provisions on regulating cultivation and distribution (13, 14). Beginning in 2012, legalization of recreational marijuana has become an important state-level policy, despite the continued prohibition at the federal level. Eight states and Washington D.C. (Washington and Colorado (2012); Oregon, Alaska, and D.C. (2014), and California, Nevada, Maine, and Massachusetts (2016)) passed policies to legalize the possession and recreational use of marijuana (13). Evaluations of the impacts of medical marijuana legalization and policy details have had mixed results with findings of increased use among those ages 26+ (15, 16) but not for younger age groups (17, 18). An analysis of the Monitoring the Future (MTF) surveys similarly found no policy effect among adolescents in the U.S. (19). However, a recent MTF analysis of recreational legalization in Washington found increased use among 8th and 10th graders in Washington but no change among 12th graders nor for any grade level in Colorado (20).
Few general adult population surveys have tracked marijuana use over time and no information on marijuana sales is available outside of states with recent legal retail sales. This study updates marijuana trends and APC decomposition analyses of the NAS series through 2015 utilizing seven surveys over 31 years. With so many policy changes in regulating use and changes in perceptions of marijuana, it is necessary to continue monitoring marijuana use to assess whether period or cohort effects are driving increased use. Furthermore, we examine how state-level marijuana policy measures, both medical and recreational legalization, have influenced marijuana use in the general population.
METHODS
Data
The NAS is a population-based survey of randomly selected U.S. adults aged 18 years and over that is conducted approximately every five years since 1979. We pooled seven waves of NAS data from 1984 (n=5,221; response rate (RR)=72%), 1990 (n=2,058; RR=70%), 1995 (n=4,925; RR=77%), 2000 (n=7,612; cooperation rate (CR)=58%), 2005 (n=6,919; CR=56%), 2010 (n=7,969; CR=52%), and 2015 (n=7,071; CR= 44%). Key changes have occurred in sampling design and survey mode with a shift from multi-stage clustered design with in-person interviews to random-digit-dialing (RDD) in 2000 and to duel-frame landline and mobile RDD in 2010, both with telephone interviews. African Americans and Hispanics were oversampled in all surveys except for 1990 (21, 22). Prior methodological studies have shown no significant differences in key alcohol measures across the shift from in-person to telephone mode (23–25). In a telephone follow-up to the 1995 NAS no significant differences in past year marijuana use found (25, 26). All surveys are weighted to the U.S. adult population, taking into account age, gender, ethnic group, and geographic area. The final analytic sample included 21,298 females and 16,061 males.
Measures
Marijuana Use
In 1984 to 2005 surveys, respondents were asked, “How often have you used marijuana, hash, THC or grass during the past 12 months?” In 2010 and 2015, the question changed slightly to reflect more contemporary terminology: “How often have you used marijuana, hash, pot, THC or “weed” during the last 12 months?” The dependent variable indicates whether the individual reported any marijuana use in the past 12 months. Starting in the 2000 survey, marijuana use was the first question asked in the illegal drugs section, while in previous surveys this question was placed in the middle of the drugs section, with questions on cocaine, heroin and amphetamine use preceding marijuana. This placement change should have no effect on participants’ responses given that in all waves the drug use questions were preceded by questions on demographics characteristics, alcohol use and alcohol-related problems.
Covariates
We examined trends by gender, age groups, race/ethnicity (African-American, Asian, Hispanic, White, and Native American and all others), and period or NAS survey year. Using Kerr’s alcohol region construct (27), U.S. states were categorized into regions corresponding to dry-to-wet environments: South, Mid-Atlantic, Pacific Coast, South Coast, New England and North Central. While these categories were developed for alcohol, they also have relevance for marijuana use with higher levels of use expected in the Pacific and New England regions (28). Other covariates include educational status (less than high school (HS), HS degree, some college, and college degree or more), inflation adjusted annual income ($0–20K, $20K–40K, $40–70K, $70K+, missing), marital status (married, widowed, divorced/separated, and never married), religion (Catholic, Jewish, no religion, and all others), and employment status (employed, unemployed, student and retired/other).
Marijuana Policy Variables
To examine policy effects, we included indicators for state-level medical and recreational marijuana use policies that were in effect by 2015. Following earlier analyses, we distinguished between states with medical marijuana laws that included provisions for dispensaries and home cultivation of medical marijuana (12).
Age-Period-Cohort
Age was grouped into eight categories starting with 18–20, 21–24, and 25–30 year olds, and then ten-year age groups thereafter until the oldest age group of 71+. We used the 41–50 age group as the reference in the APC models. Period is represented by seven NAS years with 2015 as the reference. Birth cohort was categorized into 15 groups starting with 1900–1920, followed by five-year groupings from 1921–1925 to 1986–1990, and ending with a 1991–1997 cohort. The 1956–1960 birth cohort is the reference group.
Analysis
We first conducted descriptive analyses of any marijuana use by survey year. These analyses were gender-stratified, and examined trends by age, race/ethnicity, and U.S. region. Statistical tests for linear trends were conducted by fitting logistic regression predicting any marijuana use with survey year coded from 0 to 6 and its estimated coefficients were used to identify significant changes over the period of observation.
Consistent with prior APC studies using NAS data, we used a fixed-effects (FE) approach for the marijuana APC models (21, 22, 29). While a hierarchical or cross-classified random effects approach is more common in recent APC studies (30), we were able to better account for NAS’s different survey modes, sampling designs and oversampling in a FE approach using Stata’s survey design feature (–svy– command). Principal components analyses of the age, period and cohort measures found condition numbers of 22.5 for men and 20.9 for women (31). These are both greater than 15, indicating concern regarding multicollinearity and identification, but below 30, the informal cutoff for serious concern. These analyses indicate the need for caution in determining the final model specification and support the use of outside information in this decision. We conducted a series of gender-specific logistic regression models of any marijuana use in the past year unadjusted and adjusted with covariates (21, 22, 29). Due to low marijuana prevalence in the oldest age groups and earliest cohort groups, some of the 5-year age groups had empty cells. Female birth cohorts from 1901–1920, and from 1921–1930 were combined to create more stable groups. For the male APC models, we initially obtained age coefficients that implausibly increased with older ages, and included extreme values for period and cohort effects (See Supplemental Table 1). We tried an alternative strategy by constraining age to a linear effect with odds ratio of 0.96 per year, based on estimates from the 2010 and 2015 NAS surveys where post-1945 cohorts had reached older age groups, in order to insure a declining age effect on marijuana use.
We also ran an alternative APC model using an intrinsic estimator (IE) method as a sensitivity analysis. The IE method estimates a “unique estimable function based on the linear and nonlinear components of the parameter vector” of the APC model (32). Since the IE method requires the APC design matrix to be singular, age was coded in 5-year groups, period into five-year intervals and cohort equaled to period minus age. This analysis was restricted to those ages 67 or younger to avoid empty cells in earlier cohorts, and accounted for survey weights.
After selection of the final APC models for women and men, we then estimated a series of models including marijuana policy measures, both individually and in combinations. Analyses were conducted in Stata/SE version 14 and accounted for survey weights and changes in sampling design across the pooled data files (33).
RESULTS
Descriptive Trends
Trends in past-year marijuana use prevalence are presented in Table 1. Overall, there was a J-shaped trend in use with declines amounting to about 30% 1984–2005 followed by a near doubling to 12.9% in 2015. For men, there was a U-shaped trend where use decreased from 1984 to 2005 and then sharply increased from 2010 to 2015, returning to 1984’s level. However, when examined by age, these trends show some differing patterns. For men under 40, rates in 2015 were similar to 1984 rates, while for those 50 and older, rates increased in 2010 and 2015 compared to earlier years. For women, significant increases are found in all age groups across the period. For women under 40, there was a J-shaped trend, while for women 40 and older, use was rare in 1984 and generally increased over time. Among men, the U-shaped trend is evident among all groups except for Hispanics who show an especially steep rise in 2015. For women, varying patterns are seen across groups but an upward trend is significant for all but Hispanics. Regional rates of use also vary although the U-shaped trend for men and increasing trend for women is seen in most areas. The region with the highest estimated rates of use vary across time with the Pacific region, which includes several states with legalized recreational use since 2010 and the highest rates for both men and women in 2015.
Table 1.
Percentage of population or sub-population reporting any marijuana use in the past year, National Alcohol Survey (NAS) 1984–2015
| Year | NAS7 1984 |
NAS8 1990 |
NAS9 1995 |
NAS10 2000 |
NAS11 2004–05 |
NAS12 2009–10 |
NAS13 2014–15 |
P trend |
|---|---|---|---|---|---|---|---|---|
| ALL | 9.9% | 9.1% | 7.5% | 7.2% | 6.7% | 10.2% | 12.9% | 0.003 |
| ALL MALE | 14.9% | 12.7% | 10.7% | 8.8% | 9.1% | 13.3% | 14.7% | 0.659 |
| ALL FEMALES | 5.5% | 5.7% | 4.6% | 5.7% | 4.4% | 7.3% | 10.6% | <0.001 |
| BY AGE | Males | |||||||
| 18–29 | 29.9% | 26.2% | 21.0% | 19.6% | 17.7% | 23.2% | 29.2% | 0.437 |
| 30–39 | 18.1% | 13.3% | 14.5% | 8.8% | 11.9% | 16.2% | 14.8% | 0.650 |
| 40–49 | 9.6% | 9.0% | 10.3% | 7.7% | 7.6% | 12.9% | 11.7% | 0.296 |
| 50–59 | 0.5% | 4.6% | 0.6% | 3.6% | 6.3% | 11.2% | 11.6% | <0.001 |
| 60+ | 0.6% | 0.8% | 0.1% | 0.5% | 1.5% | 1.8% | 7.0% | <0.001 |
| Females | ||||||||
| 18–29 | 13.3% | 11.5% | 11.5% | 14.6% | 12.3% | 16.9% | 23.7% | 0.002 |
| 30–39 | 7.5% | 9.1% | 6.0% | 6.9% | 3.9% | 10.1% | 15.0% | 0.029 |
| 40–49 | 0.2% | 4.4% | 3.0% | 4.7% | 4.5% | 6.3% | 8.7% | <0.001 |
| 50–59 | 0.1% | 0.4% | 1.3% | 2.1% | 1.8% | 4.0% | 7.3% | <0.001 |
| 60+ | 0.0% | 0.2% | 0.5% | 0.0% | 0.5% | 1.0% | 1.9% | 0.001 |
| BY RACE/ETHNICITY | Males | |||||||
| White | 14.4% | 12.5% | 10.6% | 7.6% | 8.8% | 13.4% | 13.4% | 0.861 |
| Black | 17.4% | 17.4% | 12.5% | 11.2% | 16.8% | 18.1% | 16.4% | 0.587 |
| Hispanic | 11.9% | 12.3% | 9.6% | 12.9% | 4.7% | 10.1% | 19.1% | 0.079 |
| Others | 27.4% | 3.7% | 10.0% | 12.7% | 9.9% | 11.9% | 15.8% | 0.870 |
| Females | ||||||||
| White | 5.6% | 5.6% | 4.7% | 5.7% | 4.5% | 8.2% | 10.1% | <0.001 |
| Black | 7.2% | 6.8% | 5.3% | 4.7% | 4.2% | 7.6% | 13.4% | 0.028 |
| Hispanic | 2.8% | 7.6% | 3.6% | 5.9% | 2.1% | 2.8% | 9.3% | 0.091 |
| Others | 0.0% | 0.0% | 1.8% | 6.5% | 8.8% | 5.8% | 12.8% | 0.002 |
| BY REGION | Males | |||||||
| Mid-Atlantic | 14.8% | 11.9% | 8.4% | 9.1% | 9.2% | 10.5% | 11.8% | 0.514 |
| North-Central | 14.9% | 12.2% | 10.8% | 7.9% | 8.7% | 17.6% | 13.7% | 0.476 |
| New England | 13.6% | 14.9% | 20.8% | 16.6% | 11.6% | 17.4% | 10.5% | 0.644 |
| Pacific | 21.4% | 16.2% | 15.2% | 11.3% | 10.9% | 17.0% | 21.4% | 0.745 |
| South Coast | 11.6% | 16.6% | 10.7% | 7.5% | 7.8% | 12.1% | 18.3% | 0.125 |
| South | 13.1% | 8.1% | 7.5% | 7.0% | 9.0% | 8.1% | 10.2% | 0.549 |
| Females | ||||||||
| Mid-Atlantic | 7.4% | 10.7% | 1.5% | 4.7% | 4.0% | 6.6% | 13.1% | 0.144 |
| North-Central | 4.6% | 5.3% | 3.7% | 5.1% | 4.4% | 8.5% | 7.9% | 0.034 |
| New England | 13.6% | 3.3% | 6.7% | 8.8% | 4.2% | 4.7% | 13.1% | 0.912 |
| Pacific | 5.9% | 3.8% | 7.6% | 8.1% | 7.6% | 9.2% | 16.4% | 0.001 |
| South Coast | 4.7% | 4.4% | 4.3% | 5.1% | 1.4% | 6.7% | 8.9% | 0.080 |
| South | 4.1% | 4.8% | 5.5% | 5.0% | 5.0% | 6.2% | 8.6% | 0.024 |
Age-Period-Cohort Trends
Trends in marijuana use prevalence were decomposed into age, period, cohort and sociodemographic characteristics in APC models (See Table 2 and Figure 1). For women, the estimated age pattern shows a peak in the early 20’s with declining odds of use to the 60’s. For men, there was an increasing, and implausible, age effect in the initial model (See Supplemental Table 1). Despite growing prevalence of marijuana use among adults ages 50+ (3), we still expect the overall age profile of marijuana use to follow that of alcohol and other drug use in that marijuana use declines with age. However, age patterns in the earlier NAS surveys were clearly influenced by the strong negative pre-1945 birth cohort effects. Thus, within the contrasting trends of very low prevalence among pre-1945 birth cohorts and higher prevalence among post-1945 cohorts, the shift in the age pattern of use over time (See Table 1) presents a difficult situation for the standard APC model to best fit the data and produce credible estimates of age, period and cohort coefficients. Therefore, we impose a declining age effect in the APC-constraint model for men (See Figure 1 and Supplemental Table 1).
Table 2.
Estimates of sociodemographic characteristics on marijuana use by gender from age-period-cohort logistic regression models in the National Alcohol Surveys 1984–2015
| Women (n= 21,298)
|
Men (n=16,061)
|
|||||
|---|---|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | |||
| Less than High School | 1.00 | 1.00 | ||||
| High School | 0.78 | 0.570 | 1.062 | 0.88 | 0.695 | 1.126 |
| Some College | 0.94 | 0.684 | 1.302 | 0.88 | 0.682 | 1.129 |
| College + | 0.77 | 0.538 | 1.102 | 0.77 | 0.590 | 1.009 |
|
| ||||||
| Income $0–20K | 1.00 | 1.00 | ||||
| $20–40K | 0.79 | 0.616 | 1.009 | 0.95 | 0.768 | 1.180 |
| $40K–70K | 0.72 | 0.550 | 0.950 | 1.02 | 0.806 | 1.284 |
| $70K+ | 0.67 | 0.478 | 0.927 | 0.89 | 0.689 | 1.136 |
| Income missing | 0.44 | 0.302 | 0.645 | 0.52 | 0.379 | 0.727 |
|
| ||||||
| White | 1.00 | 1.00 | ||||
| Asian | 0.40 | 0.198 | 0.802 | 0.41 | 0.228 | 0.736 |
| Black | 0.77 | 0.591 | 0.997 | 1.17 | 0.945 | 1.456 |
| Hispanic | 0.40 | 0.285 | 0.554 | 0.54 | 0.420 | 0.691 |
| American Indian | 0.85 | 0.419 | 1.716 | 2.00 | 1.196 | 3.335 |
| Other | 1.59 | 0.629 | 4.022 | 0.44 | 0.152 | 1.299 |
|
| ||||||
| Married | 1.00 | 1.00 | ||||
| Divorced-Separated | 1.66 | 1.236 | 2.222 | 1.87 | 1.449 | 2.411 |
| Widowed | 0.87 | 0.507 | 1.480 | 1.93 | 0.935 | 3.985 |
| Never Married | 1.62 | 1.260 | 2.088 | 1.56 | 1.291 | 1.881 |
|
| ||||||
| Region-Mid-Atlantic | 1.00 | 1.00 | ||||
| North Central | 0.86 | 0.639 | 1.161 | 1.22 | 0.948 | 1.559 |
| New England | 1.56 | 0.922 | 2.633 | 1.78 | 1.258 | 2.510 |
| Pacific | 1.54 | 1.138 | 2.094 | 1.74 | 1.341 | 2.257 |
| South Coast | 0.94 | 0.675 | 1.296 | 1.13 | 0.860 | 1.476 |
| South | 0.91 | 0.667 | 1.232 | 0.83 | 0.625 | 1.102 |
|
| ||||||
| Other religion | 1.00 | 1.00 | ||||
| Jewish | 4.78 | 2.619 | 8.732 | 1.70 | 1.005 | 2.874 |
| Catholic | 1.12 | 0.859 | 1.464 | 1.41 | 1.161 | 1.722 |
| No Religion | 2.13 | 1.702 | 2.663 | 2.00 | 1.663 | 2.415 |
|
| ||||||
| Employed | 1.00 | 1.00 | ||||
| Unemployed | 1.47 | 1.072 | 2.020 | 1.49 | 1.149 | 1.935 |
| Student | 0.88 | 0.576 | 1.347 | 0.94 | 0.642 | 1.370 |
| Retired/other | 1.01 | 0.786 | 1.299 | 1.64 | 1.251 | 2.140 |
| _constant | 0.11 | 0.051 | 0.214 | 0.69 | 0.440 | 1.087 |
Notes: OR=incidence rate ratios; CI=Confidence Intervals. Models control for age, period, and cohort.
Figure 1.
Age, period and cohort effects for women and men displayed as odds ratios (OR) from logistic regression models of any marijuana use from 7 National Alcohol Surveys. Reference groups are the 41–50 age group, the 2015 survey and the 1956–1960 birth cohort. Dotted lines represent the estimated 95% confidence intervals. Male age effects are estimated from the age-constraint model, and thus no confidence intervals are not presented.
Overall, these APC models mainly attribute trends in marijuana use to period effects for both men and women (See Figure 1). Previous findings of lower use among cohorts born before 1945 for men and 1950 for women were confirmed. Positive effects among recent cohorts for women are suggested, though no significant differences were found among post-1950 cohorts for either gender. Significant period effects demonstrate generally rising odds of use for men from 2000 and for women from 2005 to 2015. For men, there was also a steep decline in the odds of use from 1984 to 2000, resulting in a U-shaped trend over the 30-year period consistent with the rates presented in Table 1.
In sensitivity analysis using the IE method, the results are not consistent with prior APC trends of marijuana use, in particular for women where the age and period effects appear to be exaggerated and the cohort effects decline with birth year, the opposite of previous findings (See Supplemental Table 2). The IE’s steeply declining age effects for men seem more plausible than the increasing age effects in our unconstrained fixed effects model, however, the estimated cohort effects for some of the pre-1945 cohorts were implausibly high, similar to the most recent cohorts. Our selection of categorical models with a constrained age effect for men and a moderate age profile for women were based on informed choices taking into account the strong pre-1945 cohort effects found in previous APC studies and retrospective measures of lifetime marijuana use.
Sociodemographic variables were found to have significant influences on marijuana use during this 30-year period as shown in Table 2. Low-income women were found to be significantly more likely to use marijuana. White men and women had the highest risk of use with the exception of American Indian men, who had double the odds of use compared to White men. Both Asian and Hispanic men and women had substantially lower odds of use compared to their White counterparts. Black women also had lower odds than White women, but Black men did not differ from White men. Having no religion was strongly positively related to the odds of marijuana use as was Jewish religion, especially for women. Catholic religion was also positively related for men only.
Within the APC framework, results for the state policy indicators recreational legalization, medical marijuana dispensaries and home growing are shown in Table 3. We examined each policy individually and in combination to allow consideration of how the policy effects may influence marijuana use after controlling for age, period, and cohort. No significant effects were found for any of the measures entered individually. For men, having dispensaries selling medical marijuana was found to significantly reduce marijuana use in models including the other policies. The addition of policy measures did not substantially change the results for other variables presented in the main APC Models in Table 2.
Table 3.
Estimates of policy effects on marijuana use by gender from APC logistic regression models, National Alcohol Survey 1984–2015
| Women (n= 21,298)
|
Men (n=16,061)
|
|||||
|---|---|---|---|---|---|---|
| OR | 95% CI | OR | 95% CI | |||
| Model estimates with single policies | ||||||
| 1. Legalizing recreational MJ | 1.36 | 0.878 | 2.111 | 0.93 | 0.623 | 1.398 |
| 2. Medical MJ grown at Home | 1.05 | 0.755 | 1.456 | 0.98 | 0.754 | 1.279 |
| 3. Medical MJ sold at dispensaries | 1.27 | 0.898 | 1.799 | 0.75 | 0.556 | 1.001 |
|
| ||||||
| Model estimates with combination of policies | ||||||
| Policies 1 + 2 | ||||||
| 1. Legalizing recreational MJ | 1.39 | 0.901 | 2.142 | 0.93 | 0.623 | 1.381 |
| 2. Medical MJ grown at Home | 1.09 | 0.794 | 1.496 | 0.97 | 0.750 | 1.264 |
| Policies 1 + 3 | ||||||
| 1. Legalizing recreational MJ | 1.38 | 0.894 | 2.127 | 0.90 | 0.601 | 1.357 |
| 3. Medical MJ sold at dispensaries | 1.28 | 0.912 | 1.799 | 0.74 | 0.552 | 0.993 |
| Policies 1 + 2 + 3 | ||||||
| 1. Legalizing recreational MJ | 1.36 | 0.878 | 2.097 | 0.94 | 0.630 | 1.416 |
| 2. Medical MJ grown at Home | 0.91 | 0.614 | 1.353 | 1.24 | 0.901 | 1.702 |
| 3. Medical MJ sold at dispensaries | 1.36 | 0.887 | 2.073 | 0.65 | 0.454 | 0.930 |
Notes: OR=incidence rate ratios; CI=Confidence Intervals. Models control for age, period, cohort, demographic characteristics, religion, and region.
DISCUSSION
Estimates of past-year marijuana use prevalence from the NAS series illustrate a decline in use during the 1980’s and 1990’s, and a steep rise in marijuana use from 2005 to 2015. Differential patterns across gender, age, race/ethnicity and U.S. region show variations in these trends, especially between younger and older age groups. Marijuana use among those born before 1945 was dramatically lower than those born after and confirms findings from earlier surveys of a cohort effect among older groups (4). In this study, our APC analyses of repeated cross-sectional surveys over the 30-year period from 1984 to 2015 find that these trends are primarily explained by period effects, consistent with a recent APC analysis showing period effects of support for marijuana legalization (11).
Modeling marijuana use in the U.S. through APC techniques is complicated due to the dramatic differences in marijuana use between earlier and later birth cohorts. Because lifetime use was so low in the pre-1945 cohorts, there was no opportunity to observe declines in marijuana use with aging, which are needed to inform the estimation of age effects. Marijuana use was more prevalent among the post-1945 birth cohorts, but due to the lack of available data from these cohorts at older ages until more recent survey years, there were difficulties in capturing these age trends. Specifically, those born in 1950 did not reach 50 years old until 2000 and 60 years old until 2010 so that age effects utilizing earlier survey data would likely exaggerate or have difficulties with estimation as we found in our analyses of U.S. men. The varying APC estimates from different models demonstrate the sensitive nature of APC modeling. Models presented in Supplementary Tables 1 and 2 provide examples of how a naïve specification can go wrong. We found that utilizing age patterns of marijuana use from more recent surveys to set parameters for age effects resulted in more plausible estimates of period and cohort effects. Thus, our APC model selection of a constrained age effect for men and a moderate age profile for women were based on informed choices taking into account the strong pre-1945 cohort effects found in previous APC studies and retrospective measures of lifetime marijuana use.
Medical and recreational marijuana policies did not have any significant association with increased marijuana use in the NAS data. This does not preclude the possibility that these policies have differential effects for different sub-groups or that alternative policy definitions could play a stronger role on marijuana use. While these analyses are unique in considering policy effects in an APC framework, they do not utilize policy-focused causal methods to explicitly control for state-specific trends (20). Marijuana policy liberalization over the past 20 years has certainly been associated with increased marijuana use; however, policy changes appear to have occurred in response to changing attitudes within states and to have effects on attitudes and behaviors more generally in the U.S. Legalization has been shown to increase the perceived availability of marijuana for older adults (17), but not for the heaviest-using younger groups who appear to have better access to illicit markets.
Our estimated period effects are substantial, implying a doubling of marijuana use rates from 2000 for men and 2005 for women. Importantly, marijuana use prevalence by men and women in their forties and fifties has reached rates above those seen for the thirties in 2000 and even for men in their sixties or older, the rate has reached 7%. The steep increases in use among older adults correspond with aging baby-boomers but also reflect the dwindling influence on pre-1945 birth cohorts on attitudes and public policy. The attribution of marijuana use trends primarily to period effects implies changing society-wide factors. Other studies have shown that disapproval of use has decreased (8) and support for marijuana legalization and perceptions of marijuana safety have increased (11) which have led to or changed along with increased marijuana use prevalence.
Limitations of these analyses include self-report of marijuana use, which may be strongly affected by illegality of use and social desirability bias, with the magnitude of such effects differing by state and changing over time as well as potentially being related to socio-demographic characteristics. Changing policies including decriminalization, medical marijuana legalization and recreational legalization could also be important factors, potentially exaggerating recent increases in marijuana use in states where policies favoring marijuana have passed. Differences in in survey mode, sampling frame, sampling method and response or cooperation rates may have affected characteristics of the respondents and their responses to sensitive questions. The outcome measure of any past year marijuana use does not capture the frequency or intensity of use or other aspects of use such as simultaneous use with alcohol or other drugs, which may be relevant to marijuana-related harms.
APC models offer a unique and important perspective on behavioral trends such as marijuana use. While we have identified some specific factors associated with marijuana use risk including unemployment, low income, having no religion or belonging to certain religious groups, divorce or separation and, for men only, retirement, these factors do not explain the 30-year trends or recent increases in marijuana prevalence. Our estimates, consistent with previous studies (5, 11), point to general period effects influencing the whole population toward a greater likelihood of past year marijuana use and generally more positive attitudes toward marijuana use and legality (34). Future studies should aim at understanding this seemingly broad phenomenon in the U.S. and relevance to changes in other countries.
Supplementary Material
Acknowledgments
This work was supported by the U.S. National Institute on Alcohol Abuse and Alcoholism (NIAAA) (grant number P50AA005595). Opinions are those of the authors, and do not necessarily reflect those of NIAAA or the National Institutes of Health.
Footnotes
Conflict of interest and financial disclosure: no conflicts to disclose
References
- 1.Wu L-T, Zhu H, Swartz M. Trends in cannabis use disorders among racial/ethnic population groups in the United States. Drug Alcohol Depend. 2016;165:181–90. doi: 10.1016/j.drugalcdep.2016.06.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Lanza ST, Vasilenko SA, Dziak JJ, Butera NM. Trends among U.S. high school seniors in recent marijuana use and associations with other substances: 1976–2013. J Adolesc Health. 2015;57(2):198–204. doi: 10.1016/j.jadohealth.2015.04.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Han BH, Sherman S, Mauro PM, Martins SS, Rotenberg J, Palamar JJ. Demographic trends among older cannabis users in the United States, 2006–2013. Addiction. 2017;112(3):516–25. doi: 10.1111/add.13670. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Kerr WC, Greenfield TK, Bond J, Ye Y, Rehm J. Age-period-cohort influences on trends in past year marijuana use in the U.S. from the 1984, 1990, 1995, and 2000 National Alcohol Surveys. Drug Alcohol Depend. 2007;86(2–3):132–8. doi: 10.1016/j.drugalcdep.2006.05.022. [DOI] [PubMed] [Google Scholar]
- 5.Miech R, Koester S. Trends in U.S., past-year marijuana use from 1985–2009: an age-period-cohort analysis. Drug Alcohol Depend. 2012;124:259–67. doi: 10.1016/j.drugalcdep.2012.01.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Compton WM, Han B, Jones CM, Blanco C, Hughes A. Marijuana use and use disorders in adults in the USA, 2002–14: analysis of annual cross-sectional surveys. Lancet Psychiatry. 2016;3(10):954–64. doi: 10.1016/S2215-0366(16)30208-5. [DOI] [PubMed] [Google Scholar]
- 7.Keyes KM, Wall M, Cerdá M, Schulenberg J, O’Malley PM, Galea S, et al. How does state marijuana policy affect US youth? Medical marijuana laws, marijuana use and perceived harmfulness: 1991–2014. Addiction. 2016;111(12):2187–95. doi: 10.1111/add.13523. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Keyes KM, Schulenberg JE, O’Malley PM, Johnston LD, Bachman JG, Li G, et al. The social norms of birth cohorts and adolescent marijuana use in the United States, 1976–2007. Addiction. 2011;106(10):1790–800. doi: 10.1111/j.1360-0443.2011.03485.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kilmer B, MacCoun RJ. How medical marijuana smoothed the transition to marijuana legalization in the United States. Annu Rev Law Soc Sci. doi: 10.1146/annurev-lawsocsci-110615-084851. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Salas-Wright CP, Vaughn MG. Marijuana use among young people in an era of policy change: what does recent evidence tell us? Am J Drug Alcohol Abuse. 2017;43(3):231–3. doi: 10.1080/00952990.2016.1226319. [DOI] [PubMed] [Google Scholar]
- 11.Campbell W, Twenge J, Carter N. Support for marijuana (cannabis) legalization: untangling age, period, and cohort effects. Collabra: Psychology. 2017;3(1):2. [Google Scholar]
- 12.Pacula RL, Powell D, Heaton P, Sevigny EL. Assessing the effects of medical marijuana laws on marijuana use: the devil is in the details. J Policy Anal Manage. 2015;34(1):7–31. doi: 10.1002/pam.21804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Cambron C, Guttmannova K, Fleming CB. State and national contexts in evaluating cannabis laws: a case study of Washington State. J Drug Issues. 2017;47(1):74–90. doi: 10.1177/0022042616678607. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Pacula RL, Smart R. Medical marijuana and marijuana legalization. Annu Rev Clin Psychol. 2017;13:397–419. doi: 10.1146/annurev-clinpsy-032816-045128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Williams AR, Santaella-Tenorio J, Mauro CM, Levin FR, Martins SS. Loose regulation of medical marijuana programs associated with higher rates of adult marijuana use but not cannabis use disorder. Addiction. doi: 10.1111/add.13904. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Hasin DS, Sarvet AL, Cerdá M, Keyes KM, Stohl M, Galea S, et al. US adult illicit cannabis use, cannabis use disorder, and medical marijuana laws 1991–1992 to 2012–2013. JAMA Psychiatry. 2017;74(6):579–88. doi: 10.1001/jamapsychiatry.2017.0724. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Martins SS, Mauro CM, Santaella-Tenorio J, Kim JH, Cerda M, Keyes KM, et al. State-level medical marijuana laws, marijuana use and perceived availability of marijuana among the general U.S. population. Drug Alcohol Depend. 2016;169:26–32. doi: 10.1016/j.drugalcdep.2016.10.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Johnson J, Hodgekin D, Harris SK. The design of medical marijuana laws and adolescent use and heavy use of marijuana: analysis of 45 states from 1991–2011. Drug Alcohol Depend. 2017;170:1–8. doi: 10.1016/j.drugalcdep.2016.10.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Hasin DS, Wall M, Keyes KM, Cerdá M, Schulenberg J, O’Malley PM, et al. Medical marijuana laws and adolescent marijuana use in the USA from 1991 to 2014: results from annual, repeated cross-sectional surveys. Lancet Psychiatry. 2015;2(7):601–308. doi: 10.1016/S2215-0366(15)00217-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Cerdá M, Wall M, Feng T, Keyes KM, Sarvet A, Schulenberg J, et al. Association of state recreational marijuana laws with adolescent marijuana use. JAMA Pediatr. 2017;171(2):142–9. doi: 10.1001/jamapediatrics.2016.3624. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Kerr WC, Greenfield TK, Bond J, Ye Y, Rehm J. Age, period and cohort influences on beer, wine and spirits consumption trends in the US National Surveys. Addiction. 2004;99(9):1111–20. doi: 10.1111/j.1360-0443.2004.00820.x. [DOI] [PubMed] [Google Scholar]
- 22.Kerr WC, Greenfield TK, Ye Y, Bond J, Rehm J. Are the 1976–1985 birth cohorts heavier drinkers? Age-period-cohort analyses of the National Alcohol Surveys 1979–2010. Addiction. 2013;108(6):1038–48. doi: 10.1111/j.1360-0443.2012.04055.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Greenfield TK, Midanik LT, Rogers JD. Effects of telephone versus face-to-face interview modes on reports of alcohol consumption. Addiction. 2000;95(2):227–84. doi: 10.1046/j.1360-0443.2000.95227714.x. [DOI] [PubMed] [Google Scholar]
- 24.Midanik LT, Greenfield TK. Telephone versus in-person interviews for alcohol use: results of the 2000 National Alcohol Survey. Drug Alcohol Depend. 2003;72(3):209–14. doi: 10.1016/s0376-8716(03)00204-7. [DOI] [PubMed] [Google Scholar]
- 25.Midanik LT, Greenfield TK, Rogers JD. Reports of alcohol-related harm: telephone versus face-to-face interviews. J Stud Alcohol. 2001;62(1):74–8. doi: 10.15288/jsa.2001.62.74. [DOI] [PubMed] [Google Scholar]
- 26.Midanik LT, Rogers JD, Greenfield TK. Mode differences in reports of alcohol consumption and alcohol-related harm. In: Cynamon ML, Kulka RA, editors. Seventh Conference on Health Survey Research Methods [DHHS Publication No (PHS) 01-1013]; Hyattsville, MD: National Center for Health Statistics, Centers for Disease Control and Prevention; 2001. pp. 129–33. [Google Scholar]
- 27.Kerr WC. Categorizing US state drinking practices and consumption trends. Int J Environ Res Public Health. 2010;7(1):269–83. doi: 10.3390/ijerph7010269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Hughes A, Lipari R, Williams MR. The CBHSQ Report. Rockville, MD: Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental Health Services Administration; Jul 26, 2016. [Accessed: 2016-07-26]. Marijuana use and perceived risk of harm from use varies within and across states. http://webcache.googleusercontent.com/search?q=cache:InFfh7l2bbIJ:www.samhsa.gov/data/sites/default/files/report_2404/ShortReport-2404.html+&cd=1&hl=en&ct=clnk&gl=us. [PubMed] [Google Scholar]
- 29.Kerr WC, Greenfield TK, Bond J, Ye Y, Rehm J. Age-period-cohort modeling of alcohol volume and heavy drinking days in the US National Alcohol Surveys: divergence in younger and older adult trends. Addiction. 2009;104(1):27–37. doi: 10.1111/j.1360-0443.2008.02391.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Yang Y, Land KC. Age–Period–Cohort Analysis: New models, methods, and empirical applications. New York: Taylor & Francis Group; 2013. p. 338. [Google Scholar]
- 31.Golub GH, Van Loan CF. Matrix Computations. 3. Baltimore, MD: John Hopkins University Press; 1996. [Google Scholar]
- 32.Yang Y, Schulhofer-Wohl S, Fu WJ, Land KC. The intrinsic estimator for age-period-cohort analysis: what it is and how to use it. Am J Sociol. 2008;113(6):1697–736. [Google Scholar]
- 33.Stata Corp. Stata Statistical Software: Release 14.0. College Station, TX: Stata Corporation; 2015. [Google Scholar]
- 34.Subbaraman MS, Kerr WC. Marijuana policy opinions in Washington state since legalization: would voters vote the same way? Contemp Drug Probl. 2016;43(4):369–80. doi: 10.1177/0091450916667081. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.