Please enable JavaScript.

Coggle requires JavaScript to display documents.

Flowchart: How to do the correct analysis, CFA with Mplus - Coggle Diagram

- - - - Are there several groups? (more than 1)
        
        no
        
        do we have more than 2 measures?
        
        no
        
        one group pretest-posttest :check:
        
        paired t-test
        
        2 more items...
        
        yes (e.g.: 1 pre-test, two measured during intervention, 1 post-test, but just one group ->we only have the within-factor no grouping aka between-factor)
        
        one-way ANOVA (within)
        
        assumptions
        
        2 more items...
        
        General Linear Model ->Repeated Measures -within-subject factor name: the DV, number of levels: number of measurements -within-subject variables: put here all the measurements -PLOTS: horizontal axis: DV ->add line chart -EM-MEANS (to get the post hoc tests): select the DV under "display means for" and select "compaire main effects" , Bonferroni -OPTIONS: descriptives, effect size, power, parameter estimates
        
        yes
        
        is there a pretest and a posttest?
        
        yes
        
        Are there many measurements before and after intervention?
        
        yes (many measures before intervention and many directly after)
        
        1 more item...
        
        no (but for instance we could have one/several measurements during)
        
        1 more item...
        
        no
        
        static group comparison :check:
        
        Is there more than one IV?
        
        2 more items...
  - - - hierarchical Regression with pre-test as control variable
        
        assumptions
        
        model specification (model is based on sound theory and literature), not stat. tested
        
        linearity: the relationship between DV +IV is linear
        
        Graphs ->Legacy Dialog ->Scatter/Dot ->Simple Scatter ->Define->x-Axis: pretest, y-Axis: posttest the graph should show a linear relationship for both CG and EG
        
        check whether there is a discontinuity in the graph (if not but there is still something e.g. a change of trend: check a moderation! 3.step in Regression add an interaction term of pretest*intervention
        
        no multicollinearity: if there are several predictors, those are not perfectly linearly related
        
        when predictors are so perfectly correlated that they become redundant
        
        Regression Menu: STATISTICS: select collinearity diagnostics ->VIF should be lower than 10 & Tolerance should be higher .10
        
        normal distribution of residuals
        
        "subjective" test: in the Regression Menu: PLOTS: y-Axis: Zresiduals X-Axis:ZPredi, click on histogram & normal probability plot ->should follow normal distribution and p-p should be all on the line
        
        test: if you have in Regression menu under SAVE clicked unstandardized residuals you can use the Descriptives Analysis ->under OPTIONS select skewness & kurtosis (should be between -1 and 1)
        
        if it doesn't hold: Bootstrapping: 1.same steps as with regression before than: -BOOTSTRAP: select "perform bootstrapping", number of samples 1000, level 95%, select "bias corrected accelerated", Sampling: "simple" Important: in the part SAVE: take out the residuals (don't save unstandardized residuals anymore otherise Bootstrapping doesn't run)
        
        independence of residuals
        
        test: Durbin-Watson Test should be ideally 2, accepted if between 1.5 and 2.5 (selected in Regression Menu under STATISTICS)
        
        homoscedasticity of residuals = variance of residuals is homogeneous at all levels of the DV
        
        in the Regression Menu: PLOTS: the plot between residuals (y) and predictor (Z) ->the dispersion should be equal for low and high levels of predictor
        
        if violated: PROCESS tool
        
        aanalyze -> Regression ->Linear ->IV: put first the pretest in , method: ENTER, click on NEXT, than put in the intervention variable -STATISTICS: estimates, model fit, R²change, descriptives, collinearity diagnostics, Residuals: Durbin-Watson -SAVE: unstadardized Residuals -PLOT: Y:ZResid X:ZPred, select Histogramm and p-p plot
        
        creating the variable for the two groups based on pre-test: recode into a different variable: select pretest as variable ->name the new variable ->old and New values ->range from 1-10 is control group e.g.