Cronbach Alpha for 3-item scale?

[livethruthis]

Hi there,

A fellow student in my program is reporting a .75 Cronbach Alpha for a 3-item scale that is part of a larger measure. Is it even possible to have such a high alpha (yes, I know that .75 is barely acceptable) for only 3-items? I'm not questioning her ethics - but her analyses. I'm simply wondering if it is possible because the number of items is so low...well below the minimum 10 items recommended by Nunnally (1978).

Many thanks
C.

---

Members don't see this ad.

 

[WisNeuro]

Yes, you can technically run a cronbach's alpha for any number of items. I wouldn't see that as the problem. Rather, this may be a unidimensional "scale." Would have to look at the total variance explained by item to see.

 

[Justanothergrad]

Running a coefficient alpha for a 3 item measure is not a great approach to capturing scale variability because alpha is influenced by the number of items in the scale. Alpha measures the average of all possible split half, so with reduced items you get reduced variability. With that few items, you get poorer measures of reliability.

 

[WisNeuro]

Running a coefficient alpha for a 3 item measure is not a great approach to capturing scale variability because alpha is influenced by the number of items in the scale. Alpha measures the average of all possible split half, so with reduced items you get reduced variability. With that few items, you get poorer measures of reliability.

Hey now, we diagnose a variety of disorders with things like the PHQ-4 here in the VA. Are you saying we've been doing it wrong all this time 😉

 

[Justanothergrad]

Hey now, we diagnose a variety of disorders with things like the PHQ-4 here in the VA. Are you saying we've been doing it wrong all this time 😉

Yeh- you're using WAY too many items!

 

[blakjak12]

As have others have said, it is possible to compute alpha with 3 items (and hey, you could even do it with two!). Good info here on how coefficient alpha is affected by scale length: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.527.7772&rep=rep1&type=pdf