I am doing a between groups exp. (4 groups).
Each participant will be asked to complete part I of a Questionnaire A measuring attitudes toward 3 different topics, and part I of Questionnaire B, measuring 2 other constructs. These will provide a baseline.
Each group will then be shown one of four video clips.
Parts II of questionnaire A and B will then be completed.
Hypothesis:
- Group membership predicts a change in score in Questionnaire A.
- Which of the three topics of Questionnaire A is most affected by group membership.
- Group membership predicts a change in score in Questionnaire B.
- Change in score in Questionnaire A predicts changes in scores in Questionnaire B.
- Change in which of the three subsets of Questionnaire A is most predictive of changes in Questionnaire B.
What statistical test is most appropriate to answer these hypotheses.
If I need to run a regression what type of regression do i need.
Would this approach be correct?
1) Group membership predicts a change in score in Questionnaire A.
ANOVA
2) Which of the three topics of Questionnaire A is most affected by group membership.
t-test: one t-test per group.
3) Group membership predicts a change in score in Questionnaire B.
ANOVA
4) Change in total score in Questionnaire A predicts changes in scores in Questionnaire B.
Linear Regression: Changes A is predictor and changes B is the outcome (ie. “Changes your attitudes predict changes in your behaviour”).
5) Change in which of the three subsets of Questionnaire A is most predictive of changes in Questionnaire B.
- In order to test which Questionnaire A constructs best predicts changes in questionnaire B scores we will perform a hierarchical regression. Appropriate post-hoc tests will be conducted If required. Predictors include changes in scores on Attitude to women, Attitude to men, Attitude to Sex. Outcomes is changes on Questionnaire B.
Variables will be coded as categorical for group membership, and as Scale for the questionnaires.
Which of the three topics of Questionnaire A is most affected by group membership? To answer this question I would do a linear regression analysis and
check the part correlations of attitudes toward 3 different topics related to the dependent variable group membership. The part correlations have to be squared and then they give the % they contribute to the variance in R2.(which is both shared and unique variances and part correlations do not add up to R2) I give you an example: Look at Tables 4 and 5. R 2 is about .88 and when you look at experience of mental illness and square it you see that this variable contributes with about 56% of the R2 So you can check which of your 3 variables goes most with group membership.
Jochen Wilhelm, Dr, is very good on this so you can ask if he is so kind and tell you his opinion on your problems.
I love MANOVA so for many groups I would do that.
Thanks lot for the advice, its incredibly helpful.
Would you say that im in the right track with the other analysis? s. Béatrice Marianne Ewalds-Kvist
Dear Africa Gil Socorro,
You have better ask Jochen Wilhelm if he could be so kind and give you an advise!
Dear Dr Ewalds-Kvist,
I contacted him privately and unfortunately he said he was unable to help.
Let's go step by step
1) Group membership predicts a change in score in Questionnaire A.
ANOVA or ANCOVA adjusted for baseline value of Questionnaire A and probably also for gender and age
2) Which of the three topics of Questionnaire A is most affected by group membership.
First of all you can apply the same ANOVA / ANCOVA model as above for each of 3 componensts of Questionnaire A and that with the lowest p-value (highest test statistics) will be winner = component most affected by group membership.
Another point of view is to take one of 4 groups and which of the three topics of Questionnaire A is most affected by this subgroup. By decriptive statistics select one with the highest absolute mean change. Then you can calculate difference between this component and each of other 3 components of Questionnaire A. For these 3 difference you can apply paired t-test (or paired Wilcoson signed rank test if distributions are far from gaussian)
3) Group membership predicts a change in score in Questionnaire B.
ANOVA or ANCOVA adjusted for baseline value of Questionnaire B and probably also for gender and age
4) Change in total score in Questionnaire A predicts changes in scores in Questionnaire B.
Linear Regression: Changes A is predictor and changes B is the outcome is one option. You should start with scatter-plot Questionnaire A on x-axis, Questionnaire B on y-axis. But class variable group (there are 4 categorial groups here) should be also included and probably also age and gender. First of all you should check if the linear relation is an appropriate model. If not then some non-linear model could be more appropriate (scatter-plot pathern could help to select which non-linear model)
5) Change in which of the three subsets of Questionnaire A is most predictive of changes in Questionnaire B.
Linear Regression model - stepwise selection with all 3 components of Questionnaire A, group effect (4 groups), age, gender (and may be another baseline characteristics as predictors if available) will be the best option.
Ladislav Pecen , thank you so much for your advice. This has cleared things up for me immensely.
Best,
Africa
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running