Hi, most of the textbooks prefer high Cronbach's Alpha. You are right, it is also preferred to be above .70, but this is not a magic number, so being around this number is acceptable. 

There are some reasons for having low alpha such as small sample size, or if the measure is translated, then some items could differ in the meaning. 

It is also good to ask the stat programs such as SPSS to give you in the output "alpha if item deleted" this will show you which specific item has affected your total scale alpha, and you can manage accordingly

good luck

Actually there is no scientific acceptable limit for Alpha. So it is up to you (personal endeavour). Some scientists consider that 0.6 is the minimal acceptable limit for Alpha. I usually use 0.6 as the minimum limit in my researches.

Hi, most of the textbooks prefer high Cronbach's Alpha. You are right, it is also preferred to be above .70, but this is not a magic number, so being around this number is acceptable. 

There are some reasons for having low alpha such as small sample size, or if the measure is translated, then some items could differ in the meaning. 

It is also good to ask the stat programs such as SPSS to give you in the output "alpha if item deleted" this will show you which specific item has affected your total scale alpha, and you can manage accordingly

good luck

I agree that there is no hard and fast limit for alpha. One thing you do is see if any of your indicators are lowering your alpha rather than contributing to it. This is particularly easy to do in SPSS where you can use the option "alpha if item deleted."

If you have a large number of items, you might consider the possibility that they are measuring two (or more) things, rather than a single construct. The best way to check this is with an exploratory factor analysis (EFA). There are lots of options for EFA, but you might start with a Principal Factors analysis with an oblique rotation of the factors.

Cronbach alpha can be visualized as the average correlation of all items used in the test. Further, it is known to be the function of number of items used in the test. Before we go for cronbach alpha, it is important to know how the individual items are correlated to total score. It means what is the contribution of each item to total score. It is quite possible that few items only are contributing significantly to total score while others are not. To see this, find out the individual correlation of each item score with total score. The SPSS software allows you to find out the correlation matrix for the given data. The items whose correlations are significant can be retained in the test while others to be deleted. Deleting the not so correlated items to total score will definitely improve the overall cronbach alpha. Please ensure that answers to all items are in same directions i.e., positive or negative direction in relation to the construct which you are measuring. 

As mentioned already, there is no acceptable scientific limit for cronbach's alpha. In case of low level of alpha, efforts should be made to include more items or to delete the items which are not so correlated with total score. 

Zahra,

low alpha may be due to violations of its assumptions. Check first whether your scale is actually uni-dimensional. Then, use Guttman's Lambda 2 (in SPSS the pull-down menu where alpha is the first option, chose Guttman) and compare Lambda2 and Lambda3 (which is actually alpha). If these coefficients differ, then the assumption of tau-equivalence is violated. Tau-equivalence means that the covariances between your items should all be the same - a quite restrict assumption, which is almost never met. Lamda2 is not based on this assumption, therefore it may be a better coefficient of internal consistency.

Coefficient omega may of course be an alternative, also known as McDonald's omega.

Regards, Karin

According to Malhotra and Birks (2007), 0.60 indicate a satisfactory internal consistancy of scale. 

Cronbach's alpha coefficient is sensitive, in some circumstances poorly worded item led to miss understanding of the item. So, it is required to look to the alph if item deleted. 

Pallent (2011) scales fewer than 10 items expected to have low Cronbach's alpha up to 0.50. and in this case it is required to check the inter-items correlations, and should be between 0.20-0.40.  

Warm regards, 

I suggest that you examine your correlation matrix to understand the role of the two items that you could possibly omit. It particular, you want to re-order the variables in the matrix to start with the ones you are including in the shorter scale, then the two you might drop, then the dependent variable.

The key sets of relationships here involve the two variables that might be dropped from the scale. On the one hand, they should show notable correlations with the dependent variable, because dropping them from the scale caused the R-sq to go down. On the other hand, they should show weak correlations with the variables in the reduced scale, because omitting them caused the alpha to go up.

If that is the case, then you should consider the nature of the two items that you separated from the original scale and ask whether there are theoretical grounds for treating them as a new, two-item scale.

It is satisfactory if you are getting loading on your respective constructs.

Regards

Rafique

Desired value (Cronbach’s Alpha)

Above 0.70

Nunnally (1978)

Nunnally and Bernstein (1994)

Hair et al. (2006)

Urbach and Ahlemann (2010)

Above 0.60

Lyberg et al. (1997)

https://www.researchgate.net/publication/267871651_PLS_analysis__Guidelines_to_evaluate_measures?ev=prf_pub

Article PLS analysis – Guidelines to evaluate measures

As you can see in the formula, Cronbach's alpha is dependent on the number of items in your scale, so if your scale has-for instance-10 items then .66 is very low, if the scale has 4 items, .66 is rather acceptable.  Exploratory or forced Factor analysis of all the items of the scales in your model is also a good way to know if your scale is reliable, if all items are highly loading on the factor and if items do not load also on the other scales  you use. Those items should be removed. 

I think with scales of fair length the attached chart might work!

If the construct had never been tested, sure u can proceed with Cronbach's Alpha of 0.665 going by Thumb rule (1:10)

Hair (2006), we can accept values near of .60. Especially if the factor have only few items.

We have just published an article about scale purification, i.e. the process of eliminating items from multi-item scales. We have used the example of SCM, but our framework can be applied to any other discipline. Maybe you need to eliminate items from your scale to improve alpha. Download: https://doi.org/10.1108/SCM-07-2016-0230 (or request via my ResearchGate page).

A low Cronbach alpha indicates the sample of items performs poorly in capturing the construct that motivated the measure, and vice versa. In exploratory research , modest reliability in the range of 0.5 to 0.6 is considered sufficient.(Hair et al, 2006)

Hair etal suggests a 0.70. there are many other literature that suggests 0.70 as a cut off value. however, if the scales are new and tested for the first time (in exploratory studies) alpha values can go low as 0.60. there is literature to support the above argument. similarly, if you are using a scale which has been widely tested then your cut off values should be higher than 0.70. there are some literature that calls for cut off values as high as 0.90. however, the cardinal rule in developing single dimension latent factor is not to have too many items. Most literature would suggest a minimum of 3 items. when you have more items it is more likely there is going be other factors/dimensions in the scale leading to poor alpha values.