The procedure preferred by statisticians for pre-post designs, is to use an Ancova, i.e. a regression:

post = pre + group (in that order)

For the reasons why statisticians prefer the ANCOVA, see:

Senn, S. (2006). Change from baseline and analysis of covariance revisited. Statistics in Medicine 25: 4334-4344.

If you had only one DV, then I would recommend ANCOVA, as Jos Feys has suggested. But you need to say more about the 3 DVs. For example:

• Is one of the DVs a primary outcome with the others being secondary?
• Do you have three univariate questions, or one multivariate question? I.e., do you care about the results for each DV individually, or are you interested only in a linear combination of the 3 DVs?
• Are they all quantitative, or are some of them categorical (e.g., success vs failure)?

Thank you for clarifying.

Were people randomly allocated to the control and experimental groups? If not, the answer is more involved than those suggested above. Also, what are your research questions? This will determine, for example, if a gain score approach is preferred. Are you interested in causation? So, unlike Jos Feys I think more information is needed before suggesting an answer. So, add my questions to Bruce Weaver 's questions.

ps. There is lots written on this situation and when different approaches may be preferred.

pps. One more qs, say a little about your data. Looking at your profile you are in education. Are these teachers, students, etc., and if so are they clustered (e.g., students in different classrooms, teachers in different schools)?

Jos Feys Daniel Wright Bruce Weaver Thank you so much for your answers and the proposed questions. So here is a little more about my data and research to clarify some issues. I have two groups of students (around 30 for each), one of them is control and the other one is the experimental group. I did not have the leverage to assign students randomly, as with most the situations in education, they were already divided into sections. I did however randomly assign the experiment/control option to the sections and the data shows that the two sections share lots of similarities.

I am using Pekrun's Control and value theory to investigate whether a task value intervention could enhance students perceived value of English as a foreign language which should lead students experiencing more positive emotions (enjoyment) and less negative emotions (boredom and anxiety). To measure students perceived value i have used Eccles' conceptualization of value which involves three types of value (Intrinsic, utility, and attainment value). I am also measuring students perceived control as CVT posits that achievement emotions arise due to an interaction between students perceived value and control. So those are my independent variables. As for the dependent ones they are the above mentioned emotions (enjoyment, anxiety, and boredom).

I hope this clarifies things. I have tried using analysis of variance but for some reason i feel that my data should be handled in a different manner, However, i might be mistaken as i am still learning.

Thank you for all the help

ANOVA or PERMANOVA :)

With cluster sampling and only two clusters, you are likely to have systematic differences (bias) between the treatment and control groups, and it is important that you adjust for them in your analysis. The model that you use can have a substantial impact on the results that you get.

Bradford Chaney thank you for your answer. What do you suggest that i do?

I would (1) compare your treatment and control groups to find what differences appear, and adjust for those variables in your model; (2) think about what actors may influence the impact of the treatment, and consider whether/how they should be included in the model; and (3) build all relevant variables into an appropriate model. I'd use regression analysis, but my experience is that choosing how to frame the research question and resulting model is generally more important than the specific statistical tool that is used. The fact that you have pretest values is very helpful, and will help to adjust for differences that you otherwise might not be able to model.

Bradford Chaney thank you so much for your answer.

Knowing your precise topic will help understand the 4 independent variables and 3 dependent variables, along with your treatment in your study and how individuals or subjects can be assigned to groups, both experimental and control to arrive at the best fit. I don't think the Mplus is that complicated to use if you have those number of variables in your own study. I think what you need to do is to relax, study the application of Mplus in experimental designs and see if it can work in yours. Wishing you the best. God bless

SPSS package can be used as a statistical analysis tool in experimental study like yours. When pretest posttest is the design, it means that all the participants are administer test to enable the researcher to have baseline score prior to the application of the treatment or intervention. After the treatment or intervention to the experimental group, another test will be given to all the participants as posttest. The two scores will then be compared so as to determine when the intervention has cause a change or not. When more than two variables are involve in the study like yours. I suggest you use Analysis of Variance (ANOVAR) or Analysis of Covariance (ANCOVAR) to test the difference in the hypotheses. It is also important to run post hoc analysis using Scheffe post hoc analysis. This will show level or degree of the effect of your interventions on dependent variables.

Absolutely Analysis of Covariance (ANCOVA) is appropriate, when you consider that the pre-test is the covariate. So the process is first build the ANOVA model leaving out the covariate, then what you're really asking is, does the pre-test explain sufficient variance? Of course, your data needs to meet the assumptions, probably the most stringent being homogeneity of regression slopes, which you would test by looking at the Test of Between Subjects Effects and looking at the interaction term, which should have a p-value of GT .05, i.e. the null hypothesis here is that there is homogeneity of slopes so we wish to fail to reject the null. I have done this analysis using SPSS v26 using the General Linear Model Univariate (in my case) or one can specify Multivariate, depending on your exact model.