I received the following question by email:
I was wondering should I use tick the
option for pairwise exclusion of
missing data when I carry out
regression analyses (or any analyses
for that matter) rather than using [some other missing values replacement strategy].
Julie Pallant recommends pairwise
exclusion of missing data in her SPSS
textbook.
I have a few thoughts, but I was interested in first hearing your thoughts.
Best Answer
Pairwise is a dangerous method in this case, IMO. If you delete pairwise then you'll end up with different numbers of observations contributing to different parts of your model, which can make interpretation difficult.
That being said, casewise deletion tends to discard lots and lots of information, so I suppose it depends on both the proportion of missing responses, and your sample size.
Personally, I would probably use the multiple imputation procedure in SPSS and run the analyses for each dataset, then combine if nothing looks odd.
This would be my strategy of choice with a high proportion of missing values, whereas if the number is small, case-wise would probably be my first choice.