Increasing Liberalization: A Time Series Analysis of the Public’s Mood toward Drugs

Abstract

Previous research suggests that American drug sentiment is becoming more liberal. However, the absence of a reliable and valid over time measure limits our understanding of changes in drug attitudes. This project utilizes the dyad ratios algorithm and 298 administrations of 66 unique survey indicators to develop a measure of public mood toward drugs from 1969 to 2021. I find that drug mood has trended more liberal since the late 2000s. I then test for the predictors and consequences of drug mood empirically using ARMAX modeling. Results suggest that the violent crime rate, presidential rhetoric on drugs, and college attendance are not significant predictors of drug mood, but punitiveness is significant and negative. Moreover, only drug mood emerges as a significant and negative predictor of punitiveness. Granger causality tests indicate that drug mood Granger causes changes in punitiveness. These results elucidate the socio-political dynamics regarding public opinion toward drug policy.

Keywords:

• Drug policy
• drug liberalization
• policy mood
• dyad ratios algorithm
• time series

Acknowledgements

The author wishes to thank Justin T. Pickett, Batya Ballard, James A. Stimson, David McDowall, and the anonymous reviewers for their helpful feedback on this project.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Only questions asked two or more times can be used since the method depends on ratio change within an indicator.

2 If a question has multiple administrations in a single year, those administrations are aggregated together. After aggregation, there are a total of 271 survey administrations.

3 I use 18–24-year-olds because drug use is most common among this age group (Schulenberg et al., 2020).

4 I conclude my independent variables are weakly exogeneous to drug mood by confirming the lagged differences of the regressors are not significantly predictive of drug mood.

5 These models are estimated using maximum likelihood with the Kalman filter and include one lag of the outcome variable (Δ drug mood). I also use robust standard errors to account for heteroskedasticity of the errors.

6 These sensitivity analyses include unit-root and cointegration tests. The results suggest no cointegration is present, and that ARMAX models in first differences are appropriate.

7 See https://www.datafiles.samhsa.gov/dataset/national-household-survey-drug-abuse-1979-nhsda-1979-ds0001

8 Examining the autocorrelation function and partial autocorrelation function from the first difference of death penalty suggests the series follows a first-order autoregressive process. I confirm the series is I(0) after first-differencing, and there is no cointegration between the series. The death penalty model is estimated using maximum likelihood with the Kalman filter and includes one lag of the outcome variable (Δ death penalty). I also use robust standard errors to account for heteroskedasticity of the errors.

Additional information

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.