Dear colleagues,
For my master's thesis, I am experimenting on three different groups of which I have recorded the weight before and after a particular treatment.
I am confused as to whether I need to perform 3 paired t-tests on the data and compare the results or perform the repeated measures ANOVA test. As each group only has two variables (the weight before and the weight after), my first thought was to perform 3 sets of paired t-tests on the data. However, I have read that the incidence of error increases for each t-test you perform on the same data set.
So my next thought was to perform an ANOVA test. The problem I encountered with this was that I didn't have six independent variables, I had three that were pair-matched. I next turned to the repeated measures ANOVA test; however, my understanding is that for this there needs to be data recorded at three different time intervals for each group. I only have two: the before weight and the after weight. I have read, however, that repeated measures ANOVA can be performed if there is only two time intervals are although this is not usual. So my conundrum is do I perform the three paired t-tests on the data and compare them, allowing for the incidence of errors to occur, or do I proceed with the repeated measured ANOVA test?
Would you please advise how best to proceed with the statistical analysis of my data? Any advice given would be greatly appreciated!
Many thanks,
Nida
It sounds like you have a mixed within-between design so you would use a mixed ANOVA with one between-subjects factor (group, 3 levels) and one within-subjects factor (before vs. after treatment, 2 levels). In other words, this is a 2-factorial ANOVA where the first factor is a between subjects (group) factor and the second factor is a "repeated measures" (within subjects) factor. In SPSS, such a design can be analyzed using the "Repeated measures" option.
In general it is almost always better to use one model that takes into account all the data, rather than doing multiple e.g. t-tests.
Hi Christian, thank you for your prompt response, I really appreciate you taking the time to answer my question. My follow- up question to that is, as it is the weight differential I am mostly interested in would it be possible for me to calculate this for example if subject A weight 5kg before and 3kg after the treatment, the value would be 2kg etc, and then on this weight differential perform an ANOVA test as there is three groups then? Or is that removing the validity of the experiment and should I perform, as you suggested, a repeated measures ANOVA test on the data?
Nida Kamal You could do that (1-factorial, non-repeated measures ANOVA with three groups = between subjects design only) on the difference score as dependent variable. That way, you could test whether there are group differences in average weight loss or gain.
An advantage of the 2(within) x 3(between) mixed ANOVA would be that you could simultaneously test three effects:
1) The main effect of experiment (treatment) --> overall effect of treatment (pre-post weight difference) regardless of group
2) The main effect of group --> overall group differences regardless of treatment condition
3) The interaction effect experiment x group (which would tell you whether there are group differences in gain/difference scores)
So in a way, the 2-factorial mixed ANOVA approach would be more comprehensive.
Salvatore S. Mangiafico thank you for your response.
Christian Geiser thank you for clarifying further for me. Being a dentist and not a statistician I was in way over my head trying to understand the reasons behind why we do these tests. I will try for the repeated measures ANOVA test, to analyse this section of my results. Thank you once again for your help!
If I understand correctly (and I could be wrong), you have pre and post weight data and three groups. Weight is actually the variable Lord uses in his hypothetical example (https://psycnet.apa.org/doiLanding?doi=10.1037%2Fh0025105). He discusses using ANCOVA or the ANOVA that you describe, so my first question is for Christian Geiser and Salvatore S. Mangiafico since they both seem to prefer ANOVA here rather than 3 t-tests (which I agree with). Would you also advice for this over an ANCOVA without any further information? For me, as far as the choice between ANCOVA and ANOVA (and there are other options), more information is necessary before answering. For example, how were people allocated into the conditions (e.g., Article Allocation to groups: Examples of Lord's paradox
,Article Gain Scores, ANCOVA, and Propensity Matching Procedures for ...
).
Thank you, Daniel Wright and Teshita Uke Chikako for providing your insights and sharing the information you have. I will certainly look into the information you have provided and make a cohesive decision based on what all of you have suggested. Thank you all once more.
Here's another article supporting the use of ANCOVA for a pre-post experimental design.
Vickers AJ, Altman DG. Analysing controlled trials with baseline and follow up measurements. Bmj. 2001 Nov 10;323(7321):1123-4.
https://www.bmj.com/content/323/7321/1123
Did you try two way ANOVA, because you have three group with two variables. right?
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA