Additional Experimental Designs

Equivalent Groups Posttest-Only Design

Pretests are not always possible, and in those instances, an equivalent-groups posttest-only design might be the only option. We can randomly assign a group of opioid-addicted individuals who are entering jail into a medication-assisted treatment program or the treatment-as-usual (counseling, Narcotics Anonymous) programs and then track participants for a period of time after release from custody and measure possible outcomes, such as recidivism, engagement in drug treatment after release from jail, or self-reported drug use after jail. The obvious drawback here is that we have no baseline or pretest data to gauge how the two groups compared on severity of drug addiction and offending history prior to the start of the treatment.


A quasi-experiment differs from the classical experimental design in that it is missing one of the three elements that characterize the classical experiment (Cook & Campbell, 1979). Random assignment of participants into treatment or control groups is not always possible, especially when researchers are dependent on an external agency to provide access to research participants. There are times, particularly during evaluation research, where it is necessary to work within the confines of agency rules or policies. When groups are selected non-randomly, we will still have an experimental group, but the control group will be referred to as a comparison group instead. The inability to randomly assign participants to the treatment and comparison groups has the potential to introduce a number of threats to internal validity here, including selection bias and selection-maturation interaction.

While less than ideal, quasi-experiments can be very valuable, especially when researchers take steps to match the treatment and comparison groups or use statistical controls. Matching involves either manually, or with statistical procedures, pairing experimental participants with comparison group members based on theoretically relevant variables. Getting back to the TC example, the criminal justice, mental health, or drug treatment agency associated with the intervention may not be in a position to allow the researchers to randomly assign people to either the TC or treatment as usual. Strah, Frost, Stowell, and Taheri (2018) faced this exact challenge when studying an inprison cognitive-behavioral treatment (CBT) program. The corrections department’s policy was that joining CBT had to be voluntary', so random assignment to treatment modality was prohibited. The researchers responded by matching program volunteers to individuals who either started the program and later dropped out or those who did not volunteer. Individuals in both groups were matched on several demographic and institutional misconduct variables to produce similar treatment and comparison groups.

An alternative to matching is to use statistical techniques to account for potential differences between the treatment and comparison groups, with the goal of isolating the effects of the treatment. A mental health agency near my university was partnering with the county jail and some police departments to offer diversion and reentry services to individuals with mental illness who were committing low-level crimes. The agency felt an ethical obligation to provide services to all who were eligible and willing to participate, so it was impossible to randomize half of those eligible into a control group. Instead, I constructed a non-equivalent comparison group by identifying people who were eligible for services but did not receive them because they either lived outside of the agency’s treatment area or staff were unable to contact them. I then used multivariate statistical models to learn if program participation remained a predictor of recidivism after controlling for the demographic and legal characteristics of the two groups (Tartaro, 2015).

Time-Series Designs

Figure 4.5 Time-Series Designs

The other type of quasi-experiment is a time-series (Jennings & Reingle Gonzalez, 2019; Maxfield & Babbie, 2012) (Figure 4.5). Time-series designs involve multiple data collection points. This would not be appropriate for use with the TC or CBT program example, as there would not be enough data collection points to make a time-series possible. This can be a very useful design, however, when you want to look at changes in crime patterns, such as monthly crime statistics over a period of a few years or yearly statistics over several decades. For example, May- hew, Clarke, and Elliott (1989) conducted an interrupted time-series. The interruption was the introduction of the independent variable, which in this case was the helmet law with enforcement. The dependent variable was yearly rates of motorcycle theft in Germany in the 1970s and 1980s. The new law was phased in gradually. Theft rates remained stable during the phase-in period but fell sharply once police began writing tickets. Mayhew and colleagues concluded that the reduction in thefts was due to fear of being caught stealing. Most motorcycle thefts are crimes of opportunity, with someone looking for transportation committing the theft at a time and place when he or she is unlikely to be caught. Once police started to pull over and ticket motorists without helmets, stealing a bike became very risky unless one happened to be carrying around a helmet in anticipation of stealing a bike.

There are multiple variations of time-series designs, including the interrupted non-equivalent comparison group time-series (Figure 4.5). This would allow researchers to assess the impact of a new policing tactic on calls for service data over several months while also collecting the same type of police data from a neighboring city that is not introducing any new police initiatives (Braga, Kennedy, Waring, &. Piehl, 2001). Another possibility is interrupted time-series with treatment removal, including several observation periods, introduction of a stimulus, more observation periods, removal of the stimulus, and then additional observation points. This design is appropriate for observing the impact of placement and then removal of crime prevention initiatives, such as such as road barriers to deter drug dealing and prostitution.

Strengths and Weaknesses of Quasi-Experimental Designs

Quasi-experimental designs are desirable in that they do provide safeguards against some threats to validity. If non-equivalent comparison groups are included, testing and instrumentation would likely impact both groups equally, as would history and maturation. Selection bias and selection-maturation remain possibilities, given the lack of equivalence between the experimental and comparison groups. As was noted earlier, it is possible to minimize the differences between the groups with matching techniques or to use statistical controls to isolate the impact of the independent variable on the dependent variable. As with classical experiments, diffusion of treatment, compensatory equalization, demoralization, and compensatory rivalry remain potential threats to validity.

< Prev   CONTENTS   Source   Next >