The most likely reasons are 1) that you have not keyed all of your items
in the same direction, i.e., the higher numerical item score should
indicate a higher score on the summated scale.
or 2) that you have not put the correct items into the correct scales.
Check the matrix of interitem correlations, if there are any negative
correlation among items they should be of trivial magnitude.
Check the corrected item-total correlation, squared multiple
correlation, and alpha-if-item-deleted.
Hope this helps.
Social Research Consultants
University Park, MD USA
Spencer Baker wrote:
> I am using an instrument that has lower level scales and an overall
> scale. It was originally created for adults and I am using it with a
> much younger population to compare. I am receiving some low and
> negative coefficient alphas at the lower level scales but the overall
> scales are in the .70s.
> Is there an article that discusses this phenomenon? It is a small
> sample (N=151) so far.