The type of data must be clarified in order to choose the appropriate statistical analysis

I wish you success

The first test to perform on any and every continuous scale variable, after the cursory descriptive statistics of frequency, mean, median, mode, standard deviation, variance, min, max and missing values is: the Kolmogorov-Smirnov (KS) tests to determine the nature of the distributions of the samples (normal vs non-normal or non-uniform). It appears that be six such distributions? That is three different "condition" groups each with a Pre score and a Post score? If the KS test is significant at all, use non-parametric paired sample hypothesis tests to evaluate for significant differences between Pre and Post (example Wilcoxon Signed Rank, etc.) if KS is not significant (i.e. all the score distributions are normal, uniform) use the paired sample T-test. If there is a question of controlling for the effect of the condition (i.e. comparing the three different conditions) for non-normal distributions, use the Kruskal-Wallis non-parametric ANOVA with Mann-Whitney U post hoc exhaustive pairwise comparisons and if a parametric test. For normally distributed samples that would not violate the assumptions of General Linear Model tests (... like T-test, ANOVA, Uni and Multi-variate multilinear regression, etc.) use one-way ANOVA with exaustive post hoc test. Understand that randomization into three groups is no guarantee that the resulting convenience samples will be normally distributed. Sample size matters as does time frame. If you do not have a minimum of two years of data, you will not be able to effectively control for the most well-known external threat to experimental validity, namely seasonal variation. When testing human subjects, NOT being able to compare the subtle impact of winter to winter, spring to spring summer to summer and fall to fall, is a limitation of the study. Especially so if you plan to publish the study.

I hope this helps.

The type of data must be clarified in order to choose the appropriate statistical analysis

I wish you success

Dear Sir

It depends on the type of variables, however, data mining is the most attractive method

Hello Victoria,

It sounds to me as if you're interested in whether a (baseline) numeracy score X treatment interaction exists, with respect to the final DV scores. That's akin to evaluating whether numeracy moderates the treatment condition/s. If you're comfortable with the usual assumptions, OLS regression could be used, Andrew Hayes' PROCESS module could be used (an anova add-on, available for SAS, SPSS, and R, see: http://www.processmacro.org/index.html), or any of a number of anova subprograms could be used.

On the other hand, if you were more concerned about whether treatment groups, after accounting for randomly-occurring differences in (baseline) numeracy were different on the outcome measure, then ordinary ancova would seem appropriate.

Good luck with your work.

It depends on the objective of the study and the type of data so as to choose the appropriate statistical test .

Prof Morse gives excellent advice. I suggest however that this question should be part of your research design. Proper Prior Planning Prevents Poor Performance. Best wishes, D. Booth

The behavior of the data (level of measurement, type of distribution, etc.) will define the most appropriate test.