I have recently conducted an experiment to compare the efficacy of 3 treatments and I had a whait-list group too. I had three interventions and a I wanted to compare their efficacy on cognitive performance (I used 5 assessment tools to assess this) of elderly, Ive selected 100 people and randomly assigned them on these 4 groups. Ive conducted a pretest for baseline, a post test after the lasts session of interventions and a follow up test after 6 month.

my experiment design is like below: where X are my interventions, and O1 is pretest, O2 is post test and O3 is follow-up

treatment1 O1 X1 O2 O3

treatment2 O1 X2 O2 O3

treatment3 O1 X3 O2 O3

wait-list O1 - O2 O3

I want to compare the effect of these 3 interventions and see if each intervention is significant or not, and I want to see if the effect of interventions are lasted after 6 month in my follow up. now whats my best option to analyze the collected data in SPSS? I have used the repeated measures in general linear model section, but it has its shortcomings and don't give me the data I need.

I have attached the PAIRWISE COMPARISON by BONFERRONI, the problem is I want to know how each treatment worked and which treatment effect lasted best over time, in this table I just see that there are differences in time 1 vs time 2 and etc, but it wont give me the data based on each treatment. can anyone help me with this please? Thanks in advance

  • $\begingroup$ What does O1 X O2 O3 mean ? You said "I want to know how each group performed and which treatment effect lasted best over time" - what do you mean by "how each group performed" mean, and what does "lasted best over time" mean ? Also, please can you describe the study in more detail. $\endgroup$ Jul 19 '20 at 11:30
  • $\begingroup$ dear Robert, thanks for your reply, I've edited my original question and added more details about my research. $\endgroup$ Jul 19 '20 at 15:31

From the updated description, you have 4 groups of participants, where each group receives a different intervention (including 1 control group. Each participant was measured 3 times, the first being a baseline measure, and the other two following the intervention.

Interest is centred on the relative differences between the "treatment effects" of the groups as well as any difference(s) over time. The pairwise comparison model that you fitted will not answer these questions.

One way to analysis these data to help answer your research question is with a mixed effects model:

  • there are repeated measures within participants, so random intercepts should be specified for participant. This will control for the non-independence of measurements within participants.

  • treatment group and time should be fixed effects since these are the focus of the research question. The estimates for treatment group will provide inference about the mean outcome measure in the groups (similar to the model you fitted). The estimates for the time variable will provide inference for the change in the outcome measurement over time.

  • If there is interest in whether the change in the outcome measurement over time differs between the groups then you would also want to fit an interaction between the treatment group variable and the time variable.

SPSS has a mixed effects command, however I don't use SPSS - and I would advise you not to. A better approach, in my opinion is to use R or Python, both of which are free. In R, the model would look like this:

outcome ~ group + time + (1|participant)

[this example does not fit the interaction]

  • $\begingroup$ I appreciate your taking the time to answer my questions. It has really helped me. I'm trying to learn R, but I'm still a novice. $\endgroup$ Jul 21 '20 at 6:43
  • $\begingroup$ No problem, and you are welcome :) We are all still learning. One very good thing about R is that whatever the problem, you can probably find the answer fairly quickly, but if not then you can always ask on stackoverflow (if its purely about R) on on here if it is a statistical question involving R. $\endgroup$ Jul 21 '20 at 6:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.