I do not agree with the previous answers.

First of all, you should think about your hypotheses. Why do you want to start a fist intervention only with an experimental group?? If you would find an meaningful (effect size) difference between pre and post measurement, there is no way to show that this change (i.e. the " efficacy ") is caused by the intervention and not only by time. This is not a statistical but a methodological problem, although in that case a dependent t-test would be correct. It is nothing you would get published (maybe as a first pre test additional to your "real" experiment, to show that "something" is changing). So, why wasting ressources for a result which is a priori not interpretable. Additionally, maybe you should think about two control groups to be convincing: a waiting control group and a sham intervention control group. This would really show the impact of your intervention.

Second, whatever your design (two or three groups) would be, if you have more than one factor/independent variable, you should not test this via simple t-tests (alpha error inflation)! In your case, you have clearly two factors, time and group, where the former is a repeated measures factor. Therefore, you should use a split-plot ANOVA (i.e. independent and repeated measures). Your hypothesis would lie in the interaction, since you believe that both (or all three) groups have the same value before the intervention, but that only the experimental group will change from t1 to t2 (or at least more than the control group(s)). If you find a significant interaction, you can use simple effects analyses (similar to t-tests) to determine which groups or time points differ at a given level of the other independent variable (e.g. differ the groups at t2).

Third, you state that you have several dependent variables. Again, similar to the alpha error inflation due to multiple t-tests instead of an ANOVA, you have the same problem with multiple ANOVAs. Therefore, if the dependent variables are moderately correlated with each other, you should think about an MANOVA approach (Tabachnick & Fidell).

Everything said assumes that all assumptions for the statistical tests are met and adequate sample sizes are given. Therefore, you should calculte sample sizes a priori with a power analysis, for example using G*Power.

Hope this helped a bit.

I would like to suggest the following:

To compare pre and post effects, you need a paired t - test.

To compare 3 or more groups, ANOVA.

Both procedures have assumption that have to be met in order to obtain an unbiased conclusion.

Paired-t is perfect to compare pre and post effects.

However, if you add a control group you should turn to the unpaired t-test between Intervention vs. Control groups.

You can approach your analysis in two ways:

1)

a) Unpaired t-test Intervention vs. Control Pre-intervention (to test if the two groups are homogeneous, i.e. have the same characteristics BEFORE the intervention)

b) Unpaired t-test Intervention vs. Control Post-intervention (to test the effectiveness of the intervention)

2)

Unpaired t-test Intervention vs. Control of Post minus Pre values. In this case you would expect no changes (Post - Pre = 0) or few differences in the Control group, but that in the treated group you would expect a relevant change according to the effectiveness of the intervention.

An ANCOVA would be the usual way to analyze, i.e. a regression: post = pre + group

group would be: treatment vs. control

I do not agree with the previous answers.

First of all, you should think about your hypotheses. Why do you want to start a fist intervention only with an experimental group?? If you would find an meaningful (effect size) difference between pre and post measurement, there is no way to show that this change (i.e. the " efficacy ") is caused by the intervention and not only by time. This is not a statistical but a methodological problem, although in that case a dependent t-test would be correct. It is nothing you would get published (maybe as a first pre test additional to your "real" experiment, to show that "something" is changing). So, why wasting ressources for a result which is a priori not interpretable. Additionally, maybe you should think about two control groups to be convincing: a waiting control group and a sham intervention control group. This would really show the impact of your intervention.

Second, whatever your design (two or three groups) would be, if you have more than one factor/independent variable, you should not test this via simple t-tests (alpha error inflation)! In your case, you have clearly two factors, time and group, where the former is a repeated measures factor. Therefore, you should use a split-plot ANOVA (i.e. independent and repeated measures). Your hypothesis would lie in the interaction, since you believe that both (or all three) groups have the same value before the intervention, but that only the experimental group will change from t1 to t2 (or at least more than the control group(s)). If you find a significant interaction, you can use simple effects analyses (similar to t-tests) to determine which groups or time points differ at a given level of the other independent variable (e.g. differ the groups at t2).

Third, you state that you have several dependent variables. Again, similar to the alpha error inflation due to multiple t-tests instead of an ANOVA, you have the same problem with multiple ANOVAs. Therefore, if the dependent variables are moderately correlated with each other, you should think about an MANOVA approach (Tabachnick & Fidell).

Everything said assumes that all assumptions for the statistical tests are met and adequate sample sizes are given. Therefore, you should calculte sample sizes a priori with a power analysis, for example using G*Power.

Hope this helped a bit.

Thank you all so much, this is definitely something for me to consider further.

Rainer, I had considered this but had worried about small samples as it would be time-limited and self-funded. I will reconsider though.

So. If I was to add a control group to the first intervention and keep the second as it is I will need to conduct two MANOVAs?

If the two interventions are independent from each other, yes, then these are two independent questions and you can conduct two separate MANOVAs.

Limited sample size is also a problem for t-tests, so you need to caclulate a power analysis BEFORE you start your experiment. Using t-tests intead of ANOVA is not a solution for small sample sizes.

For example, using G*Power, if you have a one-tailed question (since you believe there will be a change in one specific direction), a medium effect size (according to Cohen, but this value has to be considered and chosen by you and according to typical results in your research field) of d=0.5 and a Power of .80 you will need 27 participants. If you assume smaller effect sizes, e.g. d=0.3, you will need 71 participants! As you can see, it is mandatory to know what you are looking for and to choose the right parameters. Imagine your "true" effect is 0.3 in your population, but you only have a sample size of 30, this is associated with a Power of .48, so if you repeat your experiment very often, only 48% will yield a significant results, which means you are better off with tossing a coin to determine if there is an effect or not.

Brilliant thank you. I feel like I am learning more here than from text books and assignments!

You are welcome, feel free to ask.