Check the wording of the items are some negatively worded and others positive? If not it suggests although you think your items are measuring the same construct you have different subscales? The best way to test this out would be to do Rasch analysis if you can find someone with those skills to help you? this will also tell you if you have misfitting items i.e. measuring different constructs.

Morning, Muhammad,

If you have items from established questionnaires, then you don't need to make an EFA. If you mixed items or if you developed the items on your own, then you should do it. 

1. Case,, established questionnaires.  Actually you need the Ctonbach's alpha for the internal consistency. Convergent and discriminant validity would be great, but I assume that you will have problems with these statistics because of the results of your EFA. Try to think theoretically, whether a rotation  (varimax, promax, direct oblimin) or no rotation could help. The rotation should be performed if you theoretically assume, for example, a polynomial relation between items.

2. Own questionnaire.

2.1. Hmm,  you have to try rotations.

2.2. delete items one by one. Some of items can cause problems. Sometimes if you delete a reflective item, the construct starts working. You need to have at least 3 items per construct in order to legitimise your measurement.

2.3. Then make the CFA. In this procedure you can predestined the constructs and the related items. This procedure will help you to find out which items are causing problems.

2.4.check for the common method bias. It could happen that people simply answer in the same way, by, for example, checking items on the right. If you clean your dataset by deleting participants with no/too low variance in answers, it might help. But if the problem remains, then you have to remaster your questionnaire or find another way to collect your data.

Sincerely, 

Eugene 

 Thank you both Yevgene and Alina for smart reply.

Pretest your questionnaire might reduce this problem.

Dear Muhammad,

You have been given some really good suggestions above.  I would like to go back to basics a bit, however. I assume you started with two separate, previously developed instruments, and it's on the basis that I'm making the suggestions below.

First, are you really sure that your two variables are DEFINITELY different from each other? Maybe if you factor analyzed all 18 items together you would obtain a different structure from what you anticipated. After all, job satisfaction and job stress are probably not all that different from each other, at least for some people, and your sample might be different from the samples on which the job satisfaction and job stress instruments were originally developed. You might discover, after looking carefully at your factor analytic output from the two instruments combined, that something new was emerging.

Second, do you have enough people to run your factor analysis with? As a (not fully sanctioned) rule of thumb, in order to obtain trustworthy results from a factor analysis researchers should have 10 times the number of participants as there are items in the analysis? So, if you combined the items from both instruments, you'd need at least 180 participants.

Third, when you ran your factor analysis, did you ensure that your KMO index and Bartlett's test of sphericity were OK? Also, did you perhaps use the scree test rather than number of eigenvalues greater than 1 to determine how many factors you wanted to extract from your data?

And, as Yevgen above suggested, have you tried rotating your data in different ways to see what seems to produce the most intelligible solution?  As part of the "sorting out" process, have you omitted items from the analysis that were cross loading or not loading on any factor sufficiently?

The above are just some suggestions to keep things moving for you. I hope they're helpful.

Your research sounds interesting.  Every good wish with it.

Robert.

Thanks a lot all of you. For the better understanding  an SPSS file is attached which may kindly be analysed for confirmation of factors extraction. First 13 questions are related to Job stress while onward to Job satisfaction.

Beside it, should I measure Cronbach Alpha individually of each variable or put in all of the items of Job Stress and Job Satisfaction at one time to obtain overall Cronbach Alpha.

Hello Muhammad,

I would like to help you, but I think you have attached an SPSS data file.  I'm not able to open it or work with it.  If, however, you had the output from SPSS, I might have a better chance.

(Some other people might feel inclined to have a look at your data and even do some analyses. They're quite easy if you know what you're doing - but unfortunately it is easy to do silly things if you're not informed about factor analysis.)

I think it is important to see how many factors you have within your data before you do anything about Cronbach's alpha.  Obtaining that statistic is appropriate only after you have a clear idea of which items you want to group together to form a unified scale.

I hope that makes sense.

Robert.

Thanks dear. I tried to upload an SPSS file but it created problem while uploading. However, very soon I will upload it by removing the problem.

Thanks again.

Imran 

Hello Imran,

First, I gather that you prefer to be known by this name.

Second, if you want to upload some SPSS output, I think it might be easiest to transfer the original SPSS output into a Word document, and then send the Word document.  It is easy to do the transfer from SPSS to Word.  Just have a blank Word document open, click on the relevant section of SPSS output, copy it, and paste it into the Word document.

Robert.

Yes

AoA Imran,

You got some prudent suggestions specially Robert's advice makes good sense to me. First you need to see whether the instruments you are using are reflective measures or formative? Indicators/items in reflective instruments tend to correlate and are uni-dimensional whereas formative instruments tend to multi-dimensional. Refer to previous studies for the factor structure. Having said that, you also need to ensure that respondents clearly understand the language and constructs mentioned therein. I personally had serious issues when I collected data for MS thesis. You know in our culture, people just carelessly fill in the questionnaire. Then to repeat Robert, you need to see the value of KMO Bar test to check if the data are good enough for EFA or not? Lastly, its also advised to have sufficient data for EFA.

Stiil if you are unable to resolve your issue, you may drop in and I will help as much possible. I am in Islamia College, just few minutes from Qurtaba Univ.

Regard,

ARIF