HI,

It simply means the items of your instrument (Questionnaire) are not able to grasps the essence of the construct your trying to evaluate. A simple role of thumb is to start with two items and adding one by one. As increase in item start deceasing the Alpha value. You have to decide that what Alpha value is suitable to start next level of analysis. I hope may this help. About benchmark value of Alpha you need to study Nunnly 1998.

thank you very much. I followed your instruction and still I get the negative value or lower positive value. could you please suggest me any other methods to go for next analysis, In deed, I want to analysis Hotelling t test for my multiple variables data to test hypothesis.

 I am obliged that my advice help you in some way. Here comes the basic question from the items you used in your instrument (Questionnaire). You are using some previously developed instrument (Questionnaire) or you developed it your own? If it is your own developed instrument (Questionnaire) then how you developed it. How much of  previously published literature was used in development this instrument (Questionnaire). If you are able to pass these test then there are more tests to consider. 

I used my own instrument following the facts of literature survey those have previously published. Using those facts and considering specific facts relevant to my study, I have developed my instrument. pls any suggestion to resolve my issue

Hi, 

I think you are able to do it. If still not i do it for if you send me the data file in spss or excel. I do it and send it back for you.

Regards...

I do think that you need to recheck your likert scale. Mabe a 5-point likert scale from 1-5 should do the trick

Sample size? A small sample size may caused this issue

Next would be the negatively worded items - recode.

Hi,

check,

a) if the items are coded in the same direction (not mixing " i am dissatisfied with my job" and "I like my job"). That is, be sure that a high number means the same thing for all the items. Recode if necessery,

b) Alpha requires that all items measure one and the same underlying factor. This should be tested by means of a confirmatory factor analysis. An alternative is that your scale is a simple composite consisting of items that refer to different things. In this case, computing alpha is nonsense and reliability should be checked by repeating the survey (retest reliability).

HTH

Holger

Askarbek Mambetaliev did you see that this it is a 5 year old thread?

Remember: Whenever someone digs up an old dead horse, all participants are notified. This happens to me 10 times a day.

Best,

--Holger

I have encountered a situation in a small data set in which scores for all 3 items (on 5-point Likert scale) were high, but all participants gave a 5 for one item. So the correlation between this item and the other two items was negative, and the Cronbach's alpha was negative. So it appears that a negative alpha can happen when the scores for one of the items are at the ceiling.

https://www.bachelorprint.eu/statistics/cronbachs-alpha/

If you get a Negative Cronbach's α, you can use factor analysis to check the factorial structure and correlations between the extended factors.

Hi Chuck,

good solution but actally, a test and support of the one-factor model should be done prior to calculate alpha :)

Best,

Holger