In a study I am using a likert scale which is consist on 12 items and its reliability test is showing .65 alpha. How can I increase alpha without deleting item as the items are playing significant role in my study?
Hello Zoiya,
With the information you provided, a definite answer is not possible. However, I can make some suggestions. I am assuming you have done an item analysis. The item-total correlations (discrimination indices) are the statistics most related to alpha. I would begin by inspecting the item with the lowest item-total correlation. Is the item clearly related to the scale? Is the item worded clearly? Does the item include jargon that might not be understood by all respondents? Do some of the words in the item have double meanings (can be interpreted differently)? You would repeat this process moving from items with the least discrimination to the most discrimination. There is probably no need to do this with items that are discriminating satisfactorily. If you cannot improve an item using these questions and the discrimination index is low, the item probably should be dropped. If this is the case, you might be able to identify a more appropriate replacement item.
Good luck,
J. McLean
A Cronbach's alpha value of .65 seems rather low for a 12-item scale. Besides low reliability, multidimensionality of the scale could be a reason for the low alpha value. That is, the items may measure more than just a single factor/dimension.
Have you checked to see whether the items are unidimensional, that is, whether a single-factor confirmatory factor analysis (CFA) model fits the 12 items? A single factor model with equal factor loadings is a prerequisite for a meaningful interpretation of Cronbach's alpha. If the single factor model does not fit, this would be an indication that it does not make sense to compute alpha for the entire item set combined, as there may be multiple factors underlying the item responses.
If a single factor model with equal loadings does fit your 12 items, then the only way to increase alpha is to add more items or to replace items with large error variance parameter estimates (that is, items with low reliability), with more reliable items.
The item to total correlation is not a replacement for the test of unidimensionality through confirmatory factor analysis.
If the author can't remove the items, then try to use enough questions to assess competence. Such a situation is common if your Cronbach Alpha is low, which means some of your items are not representatives of the domain of behavior.
James E. McLean thank you so much . I have done scale reliability test and it item total correlation is -.366.
There is no jargon and it is clearly defined statement in this item . The scale is about the attitude of people towards infidelity.
The Item is already reverse coded.
Hi—With the new information, i have another suggestion. Since you have a moderate negative correlation, it is likely that item should be reverse scored. That is, 5=1, 4=2, 3=3, 2=4, 1=5. That should increase your alpha markedly. J. McLean
Another option is consider your sample. Cronbach's alpha is NOT a characteristic of a scale. It is a characteristic of a scale for a sample/population. You haven't talked about your sample, but if it is limited in some way this can also lower the alpha.
Daniel Wright sample is 450 and married male and female and study is about cheating behaviours in marrriage. the item is adding a new information in my study so it should be add in my results that all male and female are not considering an item as cheating in marriage so I have to mention in my research but this item is not correlated with the items of the scale so what should I do? should I mention the result of it seperately if it is not correlated.
Zoiya Naz, from the above, I am not sure what attributes you want the scale you are using to possess, but I think it's important to realise there are a number of misconceptions concerning coefficient alpha (also known, less appropriately, as Cronbach's alpha).
Certainly, adding items to your scale will probably increase the value of alpha, but whether that "fools" you into thinking that something desirable has occurred to your scale is quite a different matter. A high value of alpha, for example, by no means guarantees a high degree of association among the items on a scale.
The following two articles might be helpful in that regard:
Schmitt, N. (1996). Uses and abuses of coefficient alpha. Psychological Assessment, 8(4), 350–353. https://doi.org/10.1037//1040-3590.8.4.350
Sijtsma, K. (2009). On the use, the misuse, and the very limited usefulness of Cronbach’s alpha. Psychometrika, 74(1), 107–120. https://doi.org/10.1007/s11336-008-9101-0
Apart from that, if I were you, I'd look carefully at a matrix of interitem correlations to see whether that would shed any light on what you are aiming for. Exploratory factor analysis might also help - particularly if you follow best-practice procedures when conducting it By that, I mean I think you should not rely on the Kaiser criterion for identifying the likely number of factors in your data (the scree plot in conjunction with parallel analysis is much better), and also not use an orthogonal (e.g., varimax) rotation (for starters, I'd recommend a promax rotation).
All the best with your research.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude