factor analysis
Locked
 
|
I have a likerts scale with 5 questions for each variables. I have 6 variables in total including the DV. I ran the principle component analysis and my items loaded to the component is did not belong to.(eg: Q1 should be in the component 1 but the result showed in component 2)
Am I run the test wrongly or is that any other way to test factor analysis? Any question is that correct to run the variable one by one or is it a must to run all the variables together? |
|
If you are dealing with
scales, you are mainly interested in what is common among the
set of items that are designed to fit that construct. So you
would use the principal axes form of factor analysis not
principal components which tries to account for the common and
unique variance.
Where did the scales and Items come from? If these are well established scales the main use of a factor analysis is to check the scoring key. what did you use for stopping rule? I.e., how did you decide how many factors to retain? Did you use varimax rotation? Did that item load cleanly on the wrong factor, or did it split with the factor you thought it belonged on? Is your group of respondents much different from the group(s) the scales were originally established on? How many respondents do you have? Did the scale developers have large sets of respondents? Art Kendall Social Research ConsultantsOn 5/3/2014 3:54 AM, vonvon [via SPSSX Discussion] wrote: I have a likerts scale with 5 questions for each variables. I have 6 variables in total including the DV. I ran the principle component analysis and my items loaded to the component is did not belong to.(eg: Q1 should be in the component 1 but the result showed in component 2)
Art Kendall
Social Research Consultants |
|
|
I did not get the original post, but I have no idea what the OP is trying to accomplish. Is there a single construct the OP believes these five questions are intended to measure? And what about the "DV"? Does the OP believe this construct is associated with the "DV"? Want to evaluate the psychometric properties? You could employ Classical Test Theory methods or Item Response Theory methods. I am almost always in the favor of the latter because it allows one to perform a more refined analysis (e.g., careful evaluation of thresholds between adjacent response option categories). Moreover, studies have shown that IRT measurement models may result in more sensitive measures of the construct than raw scores in statistical testing.
Side note: It isn't clear to me how the original post is connected to SPSS. It's always helpful if posts can be tied to SPSS in some fashion so as not to go too far afield. Ryan On Sat, May 3, 2014 at 7:58 AM, Art Kendall <[hidden email]> wrote:
|
|
|
In reply to this post by Art Kendall
Thanks for your replied.
the questions are based on self administered questionnaires. My supervisor asked to use factor analysis to reduce the number of items in the questionnaire. I want to retain all of the 5 factors if possible. Yes, I used the varimax solution as well. Some of the items have higher loading score on the component it does not belong on. (eg: one variables for 5 questions, 3 of them have high loading (>0.5) in component 2, while 2 of them have high loading in component 3. and I have 1 variable that have only 1 question is greater than 0.5. I have a total of 254 respondents in my research. |
|
|
Where did the
questionnaire come from?
Were did the set of 5 constructs come from? What are the 5 constructs you are trying to measure? What will the resulting scales be used for? Is it correct that you retained and rotated 5 factors because you intended to measure 5 constructs? What did the eigenvalues look like? How do they compare to the eigenvalues from parallel analysis? Art Kendall Social Research ConsultantsOn 5/4/2014 11:02 PM, vonvon [via SPSSX Discussion] wrote: Thanks for your replied.
Art Kendall
Social Research Consultants |
|
|
Additionally… How many items are there? Is this sample similar to that/those upon which the original factors were determined? From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Art Kendall Where did the questionnaire come from? Art
Social Research Consultants
On 5/4/2014 11:02 PM, vonvon [via SPSSX Discussion] wrote:
Art
View this message in context:
Re: factor analysis |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |