The interitem correlations for a scale should rarely have any but
trivial negative coefficients.
The most common causes are including items that belong on other scales,
or failing to reflect items as needed.
I find it easiest to create a new set of variables and reflect a subset
to those so all variables in the transformation.
recode olditem001 to olditem200(else = copy) into newitem001 to newitem200.
*this is a rare instance where it is ok to use RECODE without INTO
because you can still retrace your steps.
recode newitem001 newitem003 newitem004 ... newitem200
(1=5)(2=4)(3=3)(4=2)(5=1)(else = copy).
reliability . . .
Be sure to inspect the minima on item correlations and the item-total
correlations to spot failures to reflect and items that weaken the scale.
Art Kendall
Social research Consultants
Madeline Becker wrote:
> I collected data via a measure that looks at adaptation to college. The
> measure has one full scale and 4 subscales. I ran a relibility analysis
> for all the items in the scale and then each of the subscales. I keep
> getting a message that says: "The (alpha) value is negative due to a
> negative average covariance among items. This violates reliability model
> assumptions. You may want to check item codings." I have spent hours
> recoding the variables (some of the items needed to be reverse scored) and
> rescoring the full scale and the subscales (just to be sure that I didn't
> mess anything up) and then looking a the data itself to make sure that it
> was entered correctly. Also the negative alpha's I am getting are: -
> .332, -.637, -.244, -.612, -.565, .115; not good. Does anyone have any
> ideas? Is it possible that I have done something wrong or do I just have
> a crummy data?
>
>
>
Locked
