Hello Neema,

Hair, Black, Babin, & Anderson (2010). Multivariate data analysis (7th ed.). Upper Saddle River, NJ: Pearson.

p. 47: "Missing data under 10 percent for an individual case or observation can generally be ignored, except when the missing data occurs in a specific nonrandom fashion (e.g., concentration in a specific set of questions, attrition at the end of the questionnaire, etc.)"

Good luck with your work!

Neema -

There is no magic number.  It depends upon variability (population standard deviation), whether the missing data, in statistical terms, are "nonignorable" (as indicated by David, or from a modeling point of view), and the accuracy requirements of your application.  You may need to consider that your data may be broken (stratified) into different subpopulations because of varying influences. 

If you can take your missing cases (nonrespondents in surveys) and manage to obtain some of those data, you might see how much difference it makes.  If there is an appreciable difference, you might surmise that the remaining missing data would make even more of a difference. 

Perhaps you could do some sensitivity  analyses, saying approximately to what degree the missing data could be different, without too much negative impact on your study, and what that impact might be, based on other knowledge. 

Cheers - Jim