Experimental Design: Which Design Should I Choose?

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

If I want to test impact of new math teaching method on kids' learning i can do one of two designs: 1. I pick 20 classes and use the new method in ten of them and teaching as usual in the other ten. I do pretest and post-test and then compare the change scores between the two groups of classes (kids)
2. I pick 20 classes and use new method for HALF of the students of each class and teaching as usual for the other half of students in the class. I do pretest and post-test and compare the change scores between the two groups of kids.
Is one design preferred to another? My colleague claims that the second design allows me to not worry about the effect of the classroom. What do people think? Any reference to written discussions on the topic are appreciated.

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：Experimental experiment Design Mental choose learning between classes compare Design

WRT the second design.  With today's software why not build classroom into the analysis. Is it really practical to split a class into two methods of teaching?

Art Kendall, Social Research Consultants

I agree with Art. The second option (randomize within classrooms) doesn't per  se get rid of the classroom effects anyway. (Classroom effects could be from  teachers, or certain people self-selecting into classrooms, or contamination between students among others I'm sure.) So I'm not sure what you gain from it besides head-aches in implementation. (On-line classes it seems doable, in person off-hand it seems very difficult.)

Also two presents a confound with the status quo of teaching that could have undesirable effects. Say for instance both experiment and control do worse than prior data would suggest. It could be from splitting the class in half it causes adverse effects.
I'd suggest picking up any book on multi-level modelling. Experiments of  this kind were one of the main motivations for development of the field. Controlling for classroom isn't difficult with current software (as Art mentioned), and it can potentially add useful insight into the experimental  analysis. (It certainly is more true to how it would be implemented in  practice anyway.)

Andrew W

I also do not see how students in the same physical class could be taught different methods.  
A variation I would consider would be to have the same teachers administer both methods to different classes. Such a design would allow you to evaluate the teacher-BY-method interaction effect. You would then likely have a greater risk for diffusion of methods.

I agree that you ought to learn about and/or consult with an expert in research designs for hierarchical designs

Ryan Black

This makes much more sense in terms of eliminating the bias due to a specific teacher. Although it does seem that some measure of implementation would be useful since the same teacher may excel at one method and not do well at the other (e.g. s/he prefers/understands/resonates with one method over the other.)

Melissa Ives