发帖
雷达卡
- PURPOSE: To test the effectiveness of a group intervention that teaches parenting skills to moms
- OUTCOMES: There are a bunch: most can be considered interval/ratio level,but one is binary
- PARTICIPANTS: 120 treatment participants, 120 control participants
- PROCEDURES: Participants are randomly assigned to experimental condition.
Those in the treatment condition are nonrandomly placed in groups based on schedules and location. There are 12 groups, each consisting of 10 participants. Treatment is offered in two different locations (which feature different demographics), and each group is led by one of four parenting trainers.
The problem we find the most vexing is how to deal with group effects and trainer effects. We are wrestling with a couple of challenges. One challenge is that there are not many groups (only 12) and even fewer trainers (only 4). I know that most multilevel statistical techniques require more than 12 clusters. However, even more challenging is the reality that those in the control condition are not placed in groups and have no interaction with trainers. Thus, the very notion of group effects and trainer effects does not apply to participants in the control condition. For this reason, it doesn't seem to make sense to us to use multilevel regression in this case--since there is no such thing as a "group" for control participants. We tossed around the idea of creating fake groups for those in the control condition (randomly assigning such participants to one of 12 fake groups), but we're not convinced this makes sense.
In short, our question is, How should one handle group and intervener effects when (a) there are a small number of groups and interveners, and (b) only half of the participants are grouped and intervened upon?
京公网安备 11010802022788号
