Lavlu Mozumdar, it might be helpful if you provided more information about your "problem". For example, how many items comprise your latent variable? With only a small number of items, it is quite possible for them to have a satisfactory degree of interrelatedness without having a high coefficient alpha. (Incidentally the term "coefficient alpha" is probably more appropriate than "Cronbach's alpha.)

Did you conduct something such as exploratory factor analysis to identify your latent variable, and, if so, what were the results of that?

Robert Trevethan: Six items comprise the latent variable. Even I test this by deleting item (one by one) option and get the maximum alpha value 0.13, which means items are not consistent anymore !

Dear Lavlu Mozumdar , with max alpha coefficient 0.13, it looks clear that the 6-items instrument is not reliable at least in your case. It might be the result of either poor understanding of items by your respondents or misunderstanding of the negative items as positive on your side (please check if you overlooked reverse coding). If all is well and still your data shows that the construct in question is not reliable then better drop this variable from your model analysis. TQ

Lavlu Mozumdar, thank you for getting back. The alpha value of .13 certainly does not look good. Please consider the possibilities referred to by Muhammad Aslam. I regard all of those possibilities as well worth heeding.

By rights, the interitem correlations should lie between .15 and .50 - at least according to Clark and Watson (1995):

Clark, L. A., & Watson, D. (1995). Constructing validity: Basic issues in objective scale development. Psychological Assessment, 7(3), 309–319. https://doi.org/10.1037/1040-3590.7.3.309

Let us know how you go!

Thank you Robert Trevethan and Muhammad Zia Aslam for your suggestions ! However, this is an important variable for my research ! As this doesn't pass the reliability test, I am thinking to use one or two of the items as separate variables ! What do you think about this ?

Hello Lavlu,

I can appreciate your difficulty, but I think there can be problems with using a single item to represent a variable. However, although single-item variables might be criticised for not capturing the richness of a construct, they have been defended by Willits, Theodori, and Luloff (2016):

Willits, F. K., Theodori, G. L., & Luloff, A. E. (2016). Another look at Likert scales. Journal of Rural Social Sciences, 31(3), 126–139. https://egrove.olemiss.edu/jrss/vol31/iss3/6

Maybe citing that article would defend your course of action. I hope that helps.

But did you do the checks that Muhammad mentioned?

Dear Robert Trevethan , I checked the issue mentioned by Muhammad Zia Aslam ! There is no problem of such. I also acknowledge the problem of using single item. When we don't have the other options, we can have a try by using the only option ! Anyway thank you again for your reference ! Greetings, Lavlu

Best of luck dear Lavlu Mozumdar to give a try with single item, however you might face resistance from reviewers. Single item measures are not a big issue BUT if a researcher takes one item out of 6 to measure a variable after data collection to overcome reliability problem, it is really a big issue in my understanding. Anyways, I wish you luck again.

Dear Muhammad Zia Aslam , Thank you for your opinion on this issue. I completely agree with you on the issue mentioned. I wish I could use all the items, but I can't due to the reliability problem. The reviewer must raise the issue of consistency of the items as well ! Please let me know if you have other suggestions so that I can use all the items to represent the variable. Greetings, Lavlu

Try to put the questions of this variable (only) in to factor analysis and see what groups of factor loading you may get, it might give you an idea with question to use or which question to use as a single item

Dear Lavlu Mozumdar , did you check Composite Reliability (CR=.50) by deleting items in a sequence from lowest loading? And if you can find CR around .50 then overall reliability even .40 would be fine as well.

If it does not work simply add values (summated scale) and treat this variable as observed variable to move forward, in a way we use in SPSS. Furthermore, with such a bad reliability, I am afraid this variable would not have significant relationships in your study.

Dear Muhammad Zia Aslam , Thank you once again for your suggestion. By deleting items in a sequence from the lowest loading, Max overall reliability I get .23 with two items ! With the summated scale, this variable is not statistically significant.

Dear Lavlu Mozumdar in your situation it looks very hard for you to move forward with this problematic variable. Your collected data tells you that this variable is not reliable and so cannot be used to draw any inferences. Better just run your analysis with summated scale of this variable and report whatever the results. It is not always necessary that we get results as suggested by literature or theory. As researchers, our work is only to justify the outcomes.

Could it be a RESULT to be discussed in your research that this important scale does show the reliability expected from the litterature?