Dear Colleagues,
we are planning a school-based study with intervention and control classes.
If we have control classes and intervention classes in the same schools, we are worried that the control classes will be contaminated by the intervention classes / their teachers.
This is why it would be ideal to have intervention and control schools and randomize them instead of intraschool randomization of classes.
However, our sample size is not large enough to test on school level (not enough schools). We have to analyze on class level in order to have a realistic chance to get significant results.
The schools we picked for the study have been analyzed before and are quite comparable.
Would it be possible to randomize into control schools and intervention schools but still take the classes as the unit for analysis?
Are there other methodological ways to solve our problem? (contamination and sample size can not be solved)
Thank you,
Titus
My answer will be brief. You may be able to use HLM (heirarchical linear modeling) to do your analysis. This allows for "nesting" of data at various levels. Sample size will effect the robustness of the analysis, but this will give you a good starting point.
There seem to be seprate issues here: One of research design (allocation at school or class level; contamination effects), one of statistical analyses. Steven Wall has given you the statistical analysis you need (in UK terminology, "multilevel modelling (MLM)"; see the Univeristy of Bristol Website). Issues of sample size/lack of power can be mitigated by you reporting your results in terms of effect sizes rather than just p-values (as is the current standard in my fields of science and social science). Your desire to randomly allocate at the school level but then test for effects at the class level is achieveable, but you'd be best placed not to agglomorate pupil data to whole classes apriori. Instead specify a 3-level HLM/MLM with pupil scores varying at the pupil, class, and school level, and with experimental condition being specified as a school level measure/variable.
Hi Titus,
Thanks for your question, one brief question in return...are you sure your samples are too small, as I did not see anything about power analysis. Depending on your effect size, you may be pleasantly surprised. I see James has mentioned this, and his answers sound good to me...what are you scoring, and how are you scoring what you are scoring?
All the best
Rich
The answer to your question can be addressed properly if you can share with us if your study is whether under qualitative or quantitative type. .If it is qualitative type, then you dont need bigger sample but if you have your study under quantitative, then you should opt for qualitative study due to the fact that you have few sample schools available in your area.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases