Hi, there.
A few days ago, I posed a question about factor analysis of survey results, but I left many important details unspecified. HERE is the more detailed one.
I develop a theoretical model of language assessment literacy (LAL) based on literature. In this model, LAL is captured in seven dimensions. A survey of 56 items with a 5-point Likert scale (unknowledgeable, slightly knowledgeable, moderately knowledgeable, rather knowledgeable, very knowledgeable) is designed to measure primary EFL teachers' LAL. This survey is first piloted with 71 target teachers, and 65 cases are valid. Many of the respondents have strings of similar responses to successive different items. An exploratory factor analysis (SPSS) with generalized least squares and direct oblimin methods shows that nine factors have Eigenvalues greater than 1, but the first item has excessively large loadings (about 60%) on the first factor. The correlation matrix shows significant correlations between every two items (ranging from .22 to .92), suggesting multicollinearity. The corrected item-total correlations range from .58 to. 85, indicating good discrimination among respondents.
How should I revise the survey items to make the survey work so that I can confirm my model?
THANK YOU VERY MUCH.
I believe that you should consider the most representative indicators of each factor with greater factorial load (+.50), another indication is that because it presents many factors you should try to maintain an equal amount of elements per dimension, this allows you to generate an instrumental model with better adjustment indexes that are presented when performing the confirmatory analysis.
Also, due to the number of samples, it is not recommended to use a multidimensional model, so it is recommended that if you can theoretically reduce items and factors, it would be more appropriate in your study.
Hi! it is important to check the Cronbach's Alpha and pre testing to refine the items before performing the EFA. Then the lower loading items should checked out as well to reveal the clear underlying dimensions. Thanks
First of all, since you have 56 items, the number of samples / respondents for the study (pilot) should be more than that, with recommended number of 100 respondents or more (Bartlett et al., 2001). Or as mentioned by the previous answer, you can look or analyze construct by construct using the same sampel. Apart from that, since you used direct oblimin technique, which means that you only want to know that whether the item measure the construct or variable, so just look at the loading factor that carry minimum value of 0.40 (some suggest 0.50 or greater).
You should be careful in data screening session. Have to check all the data inserted in a proper manner. Along with it have to check the scale, nominal ordinal oe rating. After these have to check mean, standard deviation and presencec of outlier. Have to revise carefully. Also the reliability value of each items. Then start with diteeminant value , if you find its acceptable range then you will proceed the process of factor abalysis. Most importantly revise the sentence structure of individual question. And the respondent you select is right person or not.
See the following articles which answer your question. Best, Ali
Article Developing and Validating a New Multi-Dimensional Scale for ...
Article Email is Evil! Behavioural Responses towards Permission-base...
Article Validating a New TQM-Benchmarking Measurement Model in an In...
Article Assessing Patients’ Perception of Health Care Service Qualit...
Article Political Advertising Effectiveness in Wartime Syria
Article When empathy hurts: Modelling university students’ word of m...
Article A Confirmatory Factor Analysis for SERVPERF Instrument based...
Article A Confirmatory Factor Analysis for SERVPERF Instrument based...
- Badges
- Science topic
- Similar topics
- Statistics
- Survey