I am conducting a scale development research and recently had collected pilot data. Should I remove items that improve the alpha values? If I insist to remain the items, is there any reference to support it?
Thank you in advance for your assistance!
Dear Mohd,
as always, the answer is: It depends.
What is you current Cronbach's Alpha value WITH the problematic item?
To what value would it be improved WITHOUT it?
The rule of thumb of our department is: Only eliminate an item if the a-values "jumps" to the next decimal point, e.g. from 0.85 to 0.90.
As far as "good" or "bad" a-values are concerned, Field (2013, 709) states:
"You’ll often see in books or journal articles, or be told by people, that a value of .7 to .8 is an acceptable value for Cronbach’s or; values substantially lower indicate an unreliable scale. Kline (1999) notes that although the generally accepted value of .8 is appropriate for cognitive tests such as intelligence tests, for ability tests a cut-off point of .7 is more suitable. He goes on to say that when dealing with psychological constructs, values below even .7 can, realistically, be expected because of the diversity of the constructs being meas-ured. Some even suggest that in the early stages of research, values as low as .5 will suffice (Nunnally, 1978)."
Futhermore, Tavakol & Dennick (2011, 54):
"There are different reports about the acceptable values of alpha, ranging from 0.70 to 0.95."
That said: Maybe you a-value already is not that bad after all.
I do not have a source at hand for the following, but I trust almost all reaearchers will agreee on it: NEVER eliminate an item simply based on numbers. ALWAYS consider what the item stands for in your (theoretical) construct and what its loss would mean for the questions you try to answer. I mean, there is a reason why you put it in in the first place.
Hope that helps. Please come back to me if I can help you further. Note though that I am not a statistician but employ a rather practical approach.
Best wishes
Marcel
Sources:
Field, A. (2013). Discovering Statistics Using IBM SPSS Statistics (4. Aufl.). Los Angeles Calif. u.a.: SAGE.
--> 5th edition (2017) available
Tavakol, M. & Dennick, R. (2011). Making Sense of Cronbach's Alpha. International Journal of Medical Education, 2, 53–55.
Many thanks for the response Marcel. It helps a lot.
There is anecdotal evidence that shows that if an item has a low ' corrected item-total correlation’ , for example less than .3, it needs to be reworded rather than removing it. What is your opinion?
Dear Mohd
glad I could help you out. If you are at an early stage in designing a questionnaire and have the time (!) to actually revise it that is what I would go for. However, I would recommend a structral approach, rather than just rewording it. I employed cognitiv interviews, more specifically a method called "thinking aloud" (Campanelli 2008) to improve my items before I tested them in a pilot study. You can still do that prior to a main study, of course.
Best wishes
Marcel
Campanelli, P. (2008). Testing Survey Questions. In E. D. de Leeuw, J. J. Hox & D. A. Dillman (Hg.), International Handbook of Survey Methodology (S. 176–200). New York: Taylor and Francis.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire