Putting zero in place of a missing value is not justifiable anywhere and anytime.

Ette Etuk would it be better if i changed this question instead?

If you replace missing data with 0, you'd be saying your participants consider those items even more important than the one ranked 1. If I understand you correctly, participants randomly ranked only 5 of the items, so it's *not* reasonable to assume the randomly missing items just so happen to match what each participant considers most important. If your goal is to find out how participants generally rank the importance of the 10 variables, then I see three approaches. First, find the average ranking for each item. If each item was equally-often presented randomly, this should give you a reasonable ordering for all 10 items, though there will not be a measure of statistically significant differences. Second, you could do a mean imputation for all missing values and conduct your ANOVA. This is a bit iffy because of so many missing values. Third, you could conduct all combinations of paired-sample t-tests (including only participants who had each particular pair) and report when one item was significantly different from the other. This is also iffy given multiple comparisons. Though you can adjust the results with a Bonferroni correction. For academic rigor, I'd go with option 3 (including the Bonferroni correction) but for everyday intuitive use (e.g., feedback within a company), I'd probably go with option 1 for simplicity. Hope this helps some, Evan. ~ Kevin

To your actual question, of course you CAN, but you probably do not want to for the reasons Kevin Grobman gives. So the goal may be to impute missing values (or use multiple imputation) and therefore you need to describe the missing value mechanism(s). Kevin said " If I understand you correctly, participants randomly ranked only 5 of the items " but I can't see in your question where you say which were shown. I think it is important to know what your research question is (or questions are) to know what to advise. From your description I am assuming that you are wanting to get something like a 2d plot that can include by items and questions, or are you assuming the variables have, for example, some kind of latent structure underlying them (i.e., some differences among the people that lead to variation in response patterns). From your question I cannot tell, but on the face of it I am not sure Friedman's is appropriate.

Thanks for pointing out multiple imputation, Daniel. I hadn't been aware of this technique and just read more about it. It looks like it was recently added to SPSS. I'll look forward to trying it when we're allowed to return to our offices. ~ Kevin