Using Propensity Score Subclassification for Multiple Treatment Doses to Evaluate a National Antidrug Media Campaign

Abstract

In 1998, the U.S. Office of National Drug Control Policy launched a national media campaign in an effort to reduce and prevent drug use among young Americans. Because the campaign was implemented nationwide, there is no control group available for use in evaluating the effects of the campaign. Nevertheless, it is possible to use propensity score methods to evaluate the effects of the campaign. However, because teens receive varying degrees of exposure to the media campaign, it is necessary to apply propensity score methods that accommodate multiple treatment doses. This work extends that of previous authors to subclassification on the propensity score for observational studies with multiple treatment doses, rather than matching on the propensity score, and proposes modifications to accommodate complex survey data. This methodology is illustrated using data from a pilot study for the media campaign evaluation.

Length of stay, therapeutic change, and recidivism for incarcerated juvenile offenders

“…To conduct the propensity score subclassification used in this analysis, it was necessary that dosage be conceptualized as ordinal rather than continuous (Lu, Zanutto, Hornik, & Rosenbaum, 2001). Previous studies using dosage to create propensity score subclassification have used five levels to achieve both interpretability and feasibility; consequently, we also used five levels (Loughran et al, 2009;Zanutto, Lu, & Hornik, 2005). The distribution of length of stay in the present study ranged from 0-36 months with a negative skew towards a mean of 6 months.…”

“…For dosage, in which categories represent an ascending level of exposure, an ordinal logit model can be used instead (Joffe & Rosenbaum, 1999). Either the linear prediction portion of the logit equation, B T x (Zanutto et al, 2005) or the response probability for any one of the dosage levels (Huang, Frangakis, Dominici, Diette, & Wu, 2005) can be used as the propensity score. In this study, the response probability of falling into the longest length of stay group (12þ months) is used to classify youth into one of five propensity groups (quintiles) using an ordinal regression with the 39 covariates as independent variables.…”

“…Following the approach taken by Zanutto et al (2005) and Loughran et al (2009), we estimate the expected rate of recidivism (filed charges) as a stratum-weighted average for each dosage level. These are calculated using the following formula:…”

Empirical Likelihood Methods for Complex Surveys With Data Missing-by-Design

“…Balancing covariates through propensity scores has been studied by several authors, including Austin (2008Austin ( , 2009), Imai and Ratkovic (2014) and Li, Morgan and Zaslavsky (2015). Rosenbaum and Rubin (1984) proposed using subclassification based on propensity scores, and Zanutto, Lu and Hornik (2005) provided an application of the subclassification method. The idea was also investigated by Lunceford and Davidian (2004) and Miratrix, Sekhon and Yu (2013) under slightly different terms such as "stratification" or "post-stratification".…”

The Role of Dosage and Quality of Head Start Experiences in the Development of Self-Regulation

“…In a randomized design the probability of a child receiving the treatment (in this case attending two year of preschool) is by definition 50%, and there is expected equivalence in terms of retreatment characteristics (Austin, 2011), while in a non-randomized design children can vary in their probability to attend two years of preschool and in their re-treatment characteristics. The samples of children that attend two years may be very different to the sample of children that attend one year, in ways that may also be related to the outcome of study (Zanutto, Lu, & Hornik, 2005).…”