AEA365 | A Tip-a-Day by and for Evaluators

CAT | Design and Analysis of Experiments

I am Melinda Davis, a Research Assistant Professor at the University of Arizona in Psychology, coordinate the Program Evaluation and Research Methods minor, and serve as Editor-in-Chief for the Journal of Methods and Measurement in the Social Sciences.  In an ideal world, evaluation studies compare two groups that differ only on the treatment assignment.  Unfortunately, there are many ways that a comparison group can differ from the intervention group.

Lesson Learned: As evaluators, we conduct experiments in order to examine the effects of potentially beneficial treatments.  We need control groups in order to evaluate the effects of treatments. Participants assigned to a control group usually receive a placebo intervention or the status quo intervention (business-as-usual). Individuals who have been assigned to a treatment-as-usual control group may refuse randomization, drop out during the course of the study, or obtain the treatment on their own.  It can be quite challenging to create a plausible placebo condition, or what evaluators call the “counterfactual” condition, particularly for a social services intervention.  Participants in a placebo condition may receive a “mock” intervention that differs in the amount of time, attention, or desirability, all of which can result in differential attrition or attitudes about the effectiveness of the treatment.  At the end of a study, evaluators may not know if an observed effect is due to time spent, attention received, participant satisfaction, group differences resulting from differential dropout rates, or the active component of treatment.  Many threats to validity can appear as problems with the control group, such as maturation, selection, differential loss of respondents across groups, and selection-maturation interactions (see Shadish, Cook and Campbell, 2002).

Cool Trick: Shadish, Clark and Steiner demonstrate an elegant approach to the control group problem. While the focus of their study was not control group issues, their doubly randomized preference trial (DRPT) included a well-designed control group.  Some participants were randomized to math or vocabulary treatment whereas the other group was randomized into their choice of instruction.

The evaluators collected math and vocabulary outcomes for all participants throughout the study.  The effects of the vocabulary intervention on the vocabulary outcome, the effects of the mathematics intervention on the mathematics outcome, and changes across the treated versus untreated condition could be compared, taking covariates into account.  This design allowed the evaluators to parse out the effects of participant bias, and the effect of treatment on the outcomes.

As evaluators, it is helpful to be aware of potential threats to validity and novel study designs that we can use to reduce such threats.

The American Evaluation Association is celebrating the Design & Analysis of Experiments TIG Week. The contributions all week come from Experiments TIG members. Do you have questions, concerns, kudos, or content to extend this aea365 contribution? Please add them in the comments section for this post on the aea365 webpage so that we may enrich our community of practice. Would you like to submit an aea365 Tip? Please send a note of interest to aea365@eval.org . aea365 is sponsored by the American Evaluation Association and provides a Tip-a-Day by and for evaluators.

· · ·

I’m Allan Porowski, a Principal Associate at Abt Associates and a fan of experiments – when they’re conducted under the right circumstances. Experiments, commonly referred to as RCTs (randomized controlled trials) go through three stages: (1) crazy start-up period, (2) normal data collection period, and (3) crazy analysis period.

Hot Tips:  Here are some tips to make that start-up period less crazy:

  • Don’t Fall in Love with the Method: Too often, evaluators try to force a given method to fit reality instead of using it to measure reality. Even though we may really want to conduct an RCT, it may not be appropriate. Experiments are not appropriate for new initiatives because they may not yet have excess demand for services, necessary data collection infrastructure, an randomization-accommodating intake process, or staff buy-in. If these criteria are not met, then the program is not ready to be tested experimentally.
  • Be Forward-Looking by Working Backwards: There’s no substitute for in-person planning sessions to hammer out evaluation details. A half day (or better yet, a full day) on-site is needed; and you’ll need a big whiteboard. It helps to start with a discussion of what the site hopes learn, and design the study to meet those goals. Starting out with the big-picture and moving into the details also gets the conversation off to a more productive start than diving into the nuances of randomization.
  • Know Your Audience, and Let Them Know You: Don’t forget that when conducting an RCT, you are asking staff to replace professional judgment with a completely random process. That’s not an easy proposition to make. It’s really important to convey your understanding that RCTs can be disruptive, and explain what can be done to minimize that disruption. Likewise, teach program staff to think like an evaluator. Get them involved in formulating research questions, identifying mediators, and developing hypotheses about the relationship between program services to outcomes. Keep in mind that nodding does not equal understanding. RCTs are not intuitive to most people, including many researchers, so take the time to explain study procedures in multiple ways.
  • Pressure-Test Your Sampling Frame: RCTs are often knocked for lacking generalizability, and unfortunately, that criticism is often warranted. Did you just recruit a bunch of sites that only serve left-handed kids in Boston? Recruitment is tough, but it’s even tougher to make a case that results are generalizable when your sampling frame doesn’t represent the program participants you’re studying.

Rad Resource:  Key Items To Get Right When Conducting a Randomized Controlled Trial in Education. Though over 10 years old, the advice is timeless.

The American Evaluation Association is celebrating the Design & Analysis of Experiments TIG Week. The contributions all week come from Experiments TIG members. Do you have questions, concerns, kudos, or content to extend this aea365 contribution? Please add them in the comments section for this post on the aea365 webpage so that we may enrich our community of practice. Would you like to submit an aea365 Tip? Please send a note of interest to aea365@eval.org . aea365 is sponsored by the American Evaluation Association and provides a Tip-a-Day by and for evaluators.

· ·

<< Latest posts

Archives

To top