I have 10 variables, for each of which there are 3 7-point Likert scale questions, so 31 questions in total (for the last variable there are 4 questions). I have 241 responses. I need to conduct SEM but first test for reliability and validity.
For now all I have is the results for 31 questions from 241 people, which steps should I implement in order to do reliability, validity and SEM?
Thank you!
Hello Polina,
Unless you're already certain of the structure of this set of questions/items/stimuli, the first order of business is likely factor analysis (not principal components) to identify structure. Then, worry about estimates of internal consistency reliability (or McDonald's omega, especially if you have a multi-dimensional structure). However, note that the reliability estimates, if derived after having played around with subscales, possible dropping of items, and so on, may be biased upwards. Additional data would help you to establish a potentially less biased estimate.
People lucky enough to have large samples can split them, perform an exploratory factor analysis on the primary sample, then run a confirmatory factor analysis (to affirm the stability of the structure identified by the EFA) and estimate score reliability. Your sample of 241 isn't quite large enough (given 31 individual variables) to do that.
Good luck with your work.
Hello Polina,
Unless you're already certain of the structure of this set of questions/items/stimuli, the first order of business is likely factor analysis (not principal components) to identify structure. Then, worry about estimates of internal consistency reliability (or McDonald's omega, especially if you have a multi-dimensional structure). However, note that the reliability estimates, if derived after having played around with subscales, possible dropping of items, and so on, may be biased upwards. Additional data would help you to establish a potentially less biased estimate.
People lucky enough to have large samples can split them, perform an exploratory factor analysis on the primary sample, then run a confirmatory factor analysis (to affirm the stability of the structure identified by the EFA) and estimate score reliability. Your sample of 241 isn't quite large enough (given 31 individual variables) to do that.
Good luck with your work.
I agree with David
You can use SPSS, JAMOVI, JASP, must take into account that the cronbach alpha is for reliability for the validity of the questionnaire have to perform the descriptive data of the items, the exploratory factorial analysis.
The AFE allows to generate structures of theoretical models and hypotheses that can be empirically tested (Gorsuch, 1983), without having previous specifications of the model or considering both the number of factors and the relationship between them (Gerbin & Hamilton, 1996).
The technique used by the AFE is to automatically extract the factors with a certain statistical criterion, obtaining the simplest factor structure in terms of its easiest and most significant interpretation (Thurstone, 1931 and 1947).
Confirmatory Factor Analysis (AFC)
The AFC allows to correct or corroborate, if any, the deficiencies of the AFE leading to a greater contrast of the specified hypotheses (Bollen, 1989), in the same way it analyzes the matrix of covariances instead of correlations, which helps establish whether the indicators are equivalent (Batista and Coenders, 1998).
The AFC is represented by flowcharts (path diagram), according to their particular specifications. The rectangles represent the items and ellipses, the common factors. Unidirectional arrows between common factors and items express saturation and bidirectional arrows indicate the correlation between common or unique factors (Jöreskog, 1969).
In conclusion, these models provide the appropriate statistical framework to assess the validity and reliability of each item instead of performing a global analysis (Batista, Coenders and Alonso, 2004), helping the researcher to optimize both the construction of a measuring instrument and The analysis of results.
In the AFC it is necessary to observe the factor loads that allow establishing the correlation between the variables and the factors (Garson, 2012). The closer they get to one, the greater the correlation will be. An empirical rule in the AFC states that the charges must be> = a 0.07. Although it is high and some factors and variables may be left out of the model, these should be taken at the discretion of the researcher, (Raubenheimer, 2004).
Taking the criteria of Garson (2012), Raubenheimer (2004) and Widaman (1993) in which they establish that, in order to eliminate collinearity between variables to the maximum, it is necessary to take the variables whose loads are> = 0.07, especially in the case of a Model proposal, in which case, its construction is based on the existing theory.
Hello Polina Melnikova
Similar question:
https://www.researchgate.net/post/What_statistical_analysis_should_I_use_on_a_Likert_scale_data
For read:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4096146/
example:
http://lavaan.ugent.be/tutorial/cfa.html
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling