2
$\begingroup$

I randomized my survey, and now I have my data on SPSS (and all cleaned up) but when trying to run my Cronbach's $\alpha$ I have a problem.

Because I have two treatments (sexualised and normal) all my measures are present twice in my data set. For example, “similarity” has 3 items but now I have 6 variables: 3 for the sexualized and 3 for the normal condition. 

Sim1  
Sim2  
Sim3  
aSim1  
aSim2  
aSim3

SPSS will not allow me to combine 2 treatments because I have no data that has answers for the 6 (obviously).

Do I calculate them separately and then get the average between the two, or do I report both Cronbach's $\alpha$s for the 2 treatments?

Improve this question
$\endgroup$
8
  • $\begingroup$ What do you want to do with $\alpha$ and how do you plan to use these data afterwards? $\endgroup$
    – Gala
    Commented Jul 18, 2013 at 20:39
  • $\begingroup$ Also what's your sample size? $\endgroup$
    – Gala
    Commented Jul 18, 2013 at 20:59
  • 1
    $\begingroup$ Can you present examples of items? The likely answer is that you calculate the alpha separately. $\endgroup$
    – Behacad
    Commented Jul 18, 2013 at 21:09
  • $\begingroup$ @Behacad +1 to your comment. In fact you definitely cannot meaningfully do anything else that calculating alpha separately on each sub-sample in my opinion but the likely answer is that you shouldn't be calculating alpha at all. $\endgroup$
    – Gala
    Commented Jul 18, 2013 at 21:16
  • $\begingroup$ Ok for example I am using purchase intention (DV) made up of 3 measures: one being perceived social norm. To make sure I am using valid and reliable measures I need to group all my items of social norm to get the cronbachs. but Because I have 2 treatments, sexualized and normal it doesn't allow me to combine the items for social norm. I have 3 items for social norm but because of the 2 treatmens it has multiplied my social norm items by 2. Thus to present a cronbachs alpha should I do them separately and then get the average? or report the 2 separately? thanks!! $\endgroup$
    – Carmen PM
    Commented Jul 18, 2013 at 21:54

1 Answer 1

Reset to default
2
$\begingroup$

Given what you have stated, you should calculate the alpha's separately. For example, you could state that the measure of social X had an alpha value of .75 in sample 1 and of .80 in sample 2. If there are very few items (e.g., around 3), average inter-item correlation may also be used instead of Cronbach's alpha.

Improve this answer
$\endgroup$
2
  • 1
    $\begingroup$ It is always possible to interpret alpha in terms of a quasi-average correlation, and vice-versa. alpha/(1-alpha) = kr/(1-r), where k = the number of items and r = (average covariance)/(average variance). From this we get alpha = kr/(k*r + 1-r) and r = alpha/(alpha + k(1-alpha)). $\endgroup$
    – Ray Koopman
    Commented Jul 19, 2013 at 7:42
  • $\begingroup$ Thank you everyone. Ray you have kind of lost me! possibly simplest thing to do is Behacad's first solution. $\endgroup$
    – Carmen PM
    Commented Jul 19, 2013 at 9:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.