It means that in the group some items correlate negatively. If alpha is negative and some items correllarate positively other negatively I would conduct exploratory factor analysis. You probably have at least two negatively correacorrelated dimensions.

With regards

I would check to make sure that items are all scored in the right direction. Have negatively worded items been reverse scored?

I agree with Paul, the primary reason for negative correlations is that the items are scored in different directions. For alpha, all of the ideas need to be scored so that they yield positive correlations.

There have been so many questions about alpha and related issues here that I have collected a set of related resources:

https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items

Thank's for all your responses,

By the way, i have checked the right scoring directions of all items before. But, did you think that the sample size has an effect on this results, because exceptionally i have just a 10 units (respondents) in my sample ..!!.

Regards.

I also agree to the opinion of my predecessors. Before calculating Cronbaach's Alpha make sure that all your items taken into account for this purpose are coded in the same direction. You will than receive a positive reliability value.

No, sample size does not make a difference in calculating alpha. The key issue is the average size of the correlation (along with the number of correlations).

I agree sample size does not make a difference. 

I agree with David and Agnieszka that sample size does not make a difference in how alpha is calculated. In practical terms, however, with just 10 respondents it would just take a couple of people to answer randomly or in a deliberately inconsistent way to make a nonsense of your data, Rules of thumb for how big your sample should be vary; some authorities would say 300 people minimum, others 200, others at least 3X as many people as you have items.

So the real answer is to get more respondents to complete your questionnaire. This will apply also to the other statistical methods that have been suggested. For now, I would, as others have said, check your scoring direction and also ensure that all respondents complete the questionnaire in a genuine, non-random way.

I agree, I have calculated the Cronbach alpha of two measurements with only 5 subjects and my Cronbach alpha is positive.. My statistic supervisor told me that a negative Cronbach alpha just means there's something wrong with the calculation or there is e reversed coding needed, because a Cronbach alpha is always a value between 0 and 1. 

Dear John, 

i have to be more clear in my case. So basically, i have chose a "purposive (judgment) sampling method ", it means that i have doing a case study of determined activity between others under a Group of companies, in which it contains a numbered of subsidiaries that i could be able to select only 10 of them in my study.

my Regards

Have you examined the correlation matrix to determine whether there are negative correlations? Alpha takes as its input the average correlation across the matrix of correlations among your variables. So, if the average of the correlations is negative, the value of alpha will be negative.

So what can you do about that David, because I also am having an issue with a very low Cronbagh's score because many of the variables in my instrument have a negative correlation. They aren't negatively coded...they are just negative correlations, so I ran a PCA but now I am not sure how to run a Cronbagh's on that data. 

Cronbach's alpha assumes that the items all measure the same thing in the same way, but a negative correlation means that the two variables are acting in opposite ways.

With regard to saying that something is positively or negatively coded, note that this refers to the direction of correlation with the theoretical concept that the item should be measuring. So, you might have two items that were both worded positively, but their substantive content would be "reversed" with regard to the theoretical concept.

With regard to PCA (or any other form of Exploratory Factor Analysis), if you have a mix of strong positive and strong loadings on a factor after rotation, then this indicates that the meaning of some of your items are reversed by comparison to the other items.

FYI, there are a variety of ways of calculating reliability from factor loadings (theta, omega, etc.), and even though these are arguably superior measures, none of them has ever achieved anything close to the popularity of alpha.

some question on your survey have opposit direction or sens tray to correct the sens of all questions (negative form or positif form)