Hello,

I have done a survey with likert scale data and an experiencing difficulties in the factor analysis and testing.

Some info about my data: i have 7 constructs which are measured with 3 or 4 item scales. I have 5 independent constructs (4 items each) and 2 dependent constructs (3 items each), resulting in a total of 26 items.

I have a theoretical model with 7 constructs.

When i do my factor analysis on all 26 items it should yield a 7 factor model. Should i include the dependent variables here? If i do so, the model does not fit so well. When i limit the analysis to the independent variables the model fits better. Is this normal and correct? Since i have a theoretical model with 7 constructs i have chosen to extract using 5 or 7 factors depending if i included the dependent factors instead of using the Eigenvalues. Is this correct?

I have two major problems here.

I have done a survey with likert scale data and an experiencing difficulties in the factor analysis and testing.

Some info about my data: i have 7 constructs which are measured with 3 or 4 item scales. I have 5 independent constructs (4 items each) and 2 dependent constructs (3 items each), resulting in a total of 26 items.

**Factor analysis**I have a theoretical model with 7 constructs.

When i do my factor analysis on all 26 items it should yield a 7 factor model. Should i include the dependent variables here? If i do so, the model does not fit so well. When i limit the analysis to the independent variables the model fits better. Is this normal and correct? Since i have a theoretical model with 7 constructs i have chosen to extract using 5 or 7 factors depending if i included the dependent factors instead of using the Eigenvalues. Is this correct?

**Actual testing**I have two major problems here.

**1)**I am not sure what test to use. The data analysis showed the variables are not normally distributed, so i think i should use nonparametric tests, but does this also hold for Likert data? I need to show the relationships between the constructs. I am unsure if i should use Kruskal-Wallis or Spearman.**2)**How do i use the variables in a given test? Several variables measure one construct and i want to test the relations between the constructs and not individual variables. Since i have 3 or 4 variables that measure 1 construct i am not sure how to use this in a test. Should i use 4 independent variables (integrated into one using factor analysis output) on one dependent or test each individual variable seperatly? And if i use them seperatly how do i integrate the results to the factor level?
Last edited: