1

I do have a dependent variable with two distinct dimensions (factors). In total the dependent variable do have 10 items. The first 5 items belong to dimension 1 (as sub-dimension of the overal dependent variable) and the second 5 items belong to dimension 2.

To clarify:

  • Overall construct (10 items)
  • ----First dimension (5 items, each item measured on a 7-point likert scale)
  • ----Second dimension (5 items, each item measured on a 7-point likert scale)

Which statistical test is appropriate?

First thoughts:

  • On one hand, ANCOVA deals with one dependent variable (yes I do have one dependent variable)
  • But on the other hand, MANCOVA could be also appropriate as I can split my dependent variable into two dimensions, and maybe they can be seen then as two dependent variables?
Raphael
  • 163
  • 1
  • 7
  • You have left out an important thing: what you hope to accomplish by using a statistical test. – rolando2 May 15 '17 at 14:02
  • I have 3 experimental conditions and want to assess the effects (or group differences in means) to the dependent variable(s) – Raphael May 15 '17 at 15:53

2 Answers2

1

  1. I would try confirmatory factor analysis to make sure that your overall construct really is two factors.

  2. If they are two factors, I would try a path model. In R, I do this with the lavaan package. You can create regression equations predicting both subscales of the dependent variable. Then, you can specify that these subscales correlate with one another. If I wanted to see if $Y_1$ and $Y_2$ were predicted by $X_1$, $X_2$, and $X_3$, I would specify this model in lavaan:

    model <- "y1 ~ x1 + x2 + x3 # y1 is predicted by all three IVs y2 ~ x1 + x2 + x3 # y2 is predicted by all three IVs y1 ~~ y2" # y1 and y2 correlate with one another

Mark White
  • 8,712
  • 4
  • 23
  • 61
  • Thanks for your reply. What is your suggestion with regards to the ANOVA and MANOVA problem. Is it better to conduct two separated ANOVA's on both subscales or one MANOVA. I am a little confused with your solution (I am new in this topic). – Raphael May 23 '17 at 05:56
-1

Your dependent variable having the nature of binary; in my opinion Logistic Regression will be more appropriate in such scenarios. Or maybe you can use Linear Discriminant Analysis.

Hamid syed
  • 1
  • The construct of the dependent variable has two subscales (such as intelligence is the overall construct, while the first dimention is quantitative and the latter is verbal). It is no that the dependent variable is dichotomous, so a logistic regression is not appropriate. – Mark White May 15 '17 at 13:21