Treatment of missing data

Computation and analysis of CAS scores can be difficult in situations where missing data is present. The following guidelines are recommended for dealing with missing data on the CAS. These guidelines were developed and tested by Dr Patty Chondros, statistician at The University of Melbourne (for a full discussion, see Appendix 6).

 1/ Should missing data be substituted?

 If a participant has missed fewer than 30% of items on a subscale, data substitution is likely to be appropriate for that subscale (see Table 4 for equivalent number of items on each subscale).

 However if the proportion of items missing is 30% or greater, it is recommended that data substitution not be used. In this case, the full subscale would normally be treated as missing.

 Table 4. Acceptable number of missing responses for data substitution in each subscale.

2/ What type of data substitution should be used?
There are two alternative types of data substitution which can be implemented, and which each yield similar results. These are “zero” substitution and horizontal mean substitution.

 Zero substitution
If the above criteria are met (see Step 1), missing responses for that subscale are simply substituted with zero. This approach may give a more conservative estimate of the number of women meeting cut-off scores for each subscale, than horizontal mean substitution.

Horizontal mean substitution
If the above criteria are met (see Step 1), missing responses for that subscale are substituted with the mean score for non-missing items in the same subscale. Note that the overall mean for all items on the CAS should not be used to replace missing data.