Hi, can anyone help me in this
I conducted Generalized estimating equations (GEE) analysis to assess the effectiveness of an educational intervention between two groups.
The results of group, time, and group * time were significant. However, non of the pairwais comparison results in the intervention and the control were significant. Also the p value is equal to 1.000 in most of the results..
So why did I get non significant results eventhough the overall model was significant. And why the p value is always equal to 1.000..
Note: pairwais comparison were adjusted for bonferroni
Hello Samia,
I can offer some general observations:
1. If your analysis suggested that there was a group * time interaction, then that would ordinarily imply that group differences are inconsistent across time, and/or time differences are inconsistent across group. If you follow these up with less powerful (bonferroni adjusted for multiple tests) simple effects tests, it is possible that poorer power is the explanation for why no test showed as significant. Without knowing more about your data set, variables, and analysis software, I can't offer any guess as to why a p-value shows up as "1.000."
2. The larger question is, why go with GEE in the first place? What was the data condition that motivated this choice of analytic method, and was there no other viable option available?
Good luck with your work.
Thank you David Morse for your answer,
To reply why did I choose GEE is because I am doing my study on a clustered data (schools) and GEE is the best choice for clustered data. I used Spss software to analyse the data.
Regarding the first question, I tried to do the time analysis for each group separately and I found that both groups are not significant but they are close to significant (0.05 and 0.07), so my explanation was because when I found a time for both groups togather maybe the overall results become significant because both groups are closed to significant, is this correct?
I am sorry but I do not understand the structure of your data. I am guessing that it is three levels - schools at level 3, pupils at level 2 and repeated measures of pupils at level 1. Is that the case - and how many observations do you have at each level? How are the groups defined - are they measured at the school or pupil or occasion level? If that is the structure you need to deal with two types of dependence (pupils nested in schools) and ( repeated measures within pupils). Have a look at
http://www.bristol.ac.uk/cmm/learning/multilevel-models/data-structures.html
Random effects modelling readily handles such complexities
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#mlmbkgrd
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS