Reliability with reverse coded items

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Posts: 2
Joined: Fri Jan 16, 2009 8:25 pm

Reliability with reverse coded items

Postby yuelibb » Fri Jan 16, 2009 9:36 pm

I have collected a questionnaire data from some 3rd graders asking about their views towards science.
The whole scale has 11 items.
The data were coded as 1 (disagree), 2 (neutral), and 3 (agree).
There are 3 negatively worded items out of the total 11 and I reverse coded them when I ran the Cronbach's Alpha.
However, the Cronbach's Alpha came out really low, and by looking at the "Cronbach's Alpha if Item Deleted" statistics, the cronbach's Alpha will increase if I deleted these three items.
Then I ran the reliability test again without reverse code these 3 items (i.e., code them the same way as the positive worded items). The CA came back higher than when I reverse coded the items, but again, if I deleted these 3 items, the CA will be even higher.
I don't know what I should do with these negatively worded items. They are worded in a way that so obviously the reverse coding is needed. But came back with poor reliability, which means they should be deleted. I wonder if there's any one knows how to handle this situation, or if I can find any solutions from the literacy.

Posts: 2757
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Postby statman » Fri Jan 16, 2009 9:47 pm

Generally one should have all scales running the same way, however, in your case I would highly suspect that alpha due to 1) a 3-point scale and 2) the neutral point. In short, IMHO, testing internal consistence in this case lacks merit so move on to your tests but they also could be less than viable.
See the note below

NOTE: Please read the Posting Guidelines and always tell us your OS, the SPSS version and information about your study and data!

Statistical Services
Posts: 41
Joined: Sat Jan 17, 2009 4:18 pm
Location: Canada

Postby sasky » Sat Jan 17, 2009 4:42 pm

Have you looked at the raw data? Is it possible that the 3rd graders did not notice the reversal of the scales? It sounds like you should delete these 3 questions. Reverse scoring should increase the alpha versus not reverse scoring. It would help if you had numbers here to illustrate what you mean by 'low' and how much the alpha increased. With 11 items one would expect a respectable alpha even with moderate correlations among the items.
Posts: 30
Joined: Fri Nov 14, 2008 6:17 pm

Postby bourne » Sun Jan 18, 2009 10:07 am

or what u can do is analyse these three separately. (not with remaining questions)
check the reliability of these three question separately.
what i guess from ur problem is, u need to group out these three items from the remaining items
Posts: 41
Joined: Sat Jan 17, 2009 4:18 pm
Location: Canada

Postby sasky » Sun Jan 18, 2009 8:39 pm

Be careful with this. Is the scale supposed to be unidimensional or multidimensional? If you followed proper scoring, maybe the scale is poor (in this sample) or maybe you should perform a factor analysis. Again, not knowing the name of the scale or the numbers from your output, we can only speculate.
Posts: 2
Joined: Fri Jan 16, 2009 8:25 pm

Postby yuelibb » Sun Jan 18, 2009 11:30 pm

Gosh! Can't believe I got this many nice suggestions during the weekend.
I'll post some numbers once I get back to office tomorrow. Hope some of you can help me diagnose.

Who is online

Users browsing this forum: No registered users and 1 guest