Reversed Likert scales

Question

I have a dataset using a four point likert scale. I'm using Wilcoxon's signed ranks test for significance between repeated measures. I'm intending to use Spearman's Rho for correlation.

My problem is that some questions had reversed scoring.

i.e. 'Everyone is nice to me' never=0, sometimes=1, usually=2, always=3
'People are horrible to me" never=0, sometimes=1, usually=2, always=3

Some of the reversed questions are part of subsets which are then also compared.

My question is should I convert the negatively scored items i.e. 3-(score), so 0 becomes 3. so I can compare like with like? I'm inclined to do this partly because I can't see how subsets can otherwise be compared.

Is this valid?

Answer 1

• When working with Likert scales that have some reverse-coded items, it is usually a good practice to make sure that all items are scored in the same direction before beginning analysis. For example, if you do not, your Cronbach's Alpha reliability will be very much downwardly biased.

• Next, in terms of correlations, assuming you have already reversed negatively scored items so that all items are in the same direction, you want to do the following. Suppose you have 5 items that measure scale A, and another 5 items that measure scale B. To compute correlations, you need to scale-score each construct first. That is, for the scale A it should be (item 1 + item 2 + item 3 + item 4 + item 5)/5. Then repeat the same process for scale B. Finally, compute your correlation of interest by using these two scale scores.