MANCOVA and MANOVA Discussion > Post-test only design & ANCOVA/ANOVA
Hi Katherine,
Thanks for your question. You have a very interesting and complex study. To answer your first question, I still would not recommend doing a post test only design. Instead, I think I would try to control for as much as I could (for example, perhaps controlling for attitudes Towards technology in general)...
With regard to analysis, I think I would definitely recommend using ANCOVA, as opposed to ANOVA, even if you believe there are no differences between groups on any individual variables, because with the ANCOVA you could still examine possible interaction effects if they exist...
November 3, 2012 |
Jeremy Taylor
Jeremy Taylor
Hello Jeremy,
I am working on a research proposal. I would like to clarify certain doubts before finalizing my proposal. Otherwise, later stats make me cry... ha ha ha.
Let me give some details about my proposal.
If, in a post-test only quasi-experimental study having 3x2 factorial design, three groups of students are involved & each group is supposed to be instructed by a different instructional method.
In this post-test only design, there are:
1. Two independent variables:
(a) instructional methods (3 levels/groups)
(b) gender (M & F)
2. Three dependent variables:
(a) achievement (3 subscales)
(b) attitude towards Science (measured by a likert scale consisting of 7 subscales)
(c) attitude towards use of technology in Science instruction (measured by a likert scale consisting of 5 subscales)
3. Four covariates (for all the dependent variables):
(a) IQ
(b) previous year achievement grades in Science
(c) previous year OVERALL achievement grades
(d) pre-attitude scores towards Science
QUESTIONS:
I. Pre-test achievement can not be measured as the students are supposed to be taught certain topics which are entirely new to them & pre-attitude towards use of technology in Science instruction also can not be measured as students have so far not experienced the use of any kind of technology by their Science or other subjects teachers in the classroom.
Is the use of post-test only design justified in this case?
II. Please suggest me the most appropriate statistical technique for data analysis by taking into consideration the following:
(i) if students are assigned to groups randomly as well as instructional methods are assigned to groups randomly (true experimental study)
(ii) if intact classes are treated as groups & only instructional methods are assigned to groups randomly (quasi-experimental study)
III. Can ANCOVA be conducted on each dependent variable by adjusting for covariates (regardless of true or quasi-experimental study) ?
Or, will ANOVA be sufficient enough (if groups are found to be equivalent on all covariates)?