Here is a review article, which includes the topic "Don't Know," by two acknowledged experts in the field. 

The obvious response is that it depends on the question, but you really need to consider what 'maybe' or 'don't know' means. For example, there is a big difference between an interviewee stating that they don't know something because they weren't interested in the question and being undecided about an issue. In the former case, you would need to exclude these cases from your analysis as missing variables, while the latter is actually a valid response and therefore needs to be included in your results.

'Don't know' of any good studies on this though. 

Dillman has a wonderful book on designing and conducting survey research. 

Thank you to those of you who have answered so far! I will look up the Dillman book. I did find this article, somewhat useful, perhaps for you, too :-)  http://academic.brooklyn.cuny.edu/economic/friedman/rateratingscales.htm

Here is a review article, which includes the topic "Don't Know," by two acknowledged experts in the field. 

Excellent, David. Thank you!

As a follow-up, what I ended up doing: I used a 4-point Likert-type scale of Strongly Agree - Somewhat Agree - Somewhat Disagree - Strongly Disagree and a fifth option of "Not Sure or N/A." I chose "Not Sure" rather than "Don't Know" based on some early tests in which users felt more comfortable admitting that they were "not sure." Based on statistical analysis, I would recommend adding a middle "Neither Agree nor Disagree" to the Likert-type scale; I plan to do this in the next iteration of my instrument. The "Not Sure or N/A" threw off the statistical analysis for interrater reliability because every time someone checked that, that response counted as "0" and had to be excluded. Fine for scoring of spaces, not so fine for analysis. So while I think, for this specific application where the spaces being evaluated were so varied, it was important to give people an "out," it was a real pain from a statistical analysis point of view. Thanks again for your input.

Naomi A. Sachs, based on your experience, is it okay to use "Maybe" as an alternative word for "Not Sure"? I did used "Maybe" in one of my survey and during the presentation of the data collected, it was argued that "Maybe" may not be an appropriate word to use - causing the researcher to further assumptions rather than firm and factual reasoning.

Thanks in advance :)

M Syafiq Iskandar that's a good point. And in that case, I don't think the use of "not sure" would make things any better. The question is, which is worse: forcing participants to choose or make a guess about something when they really don't know the answer, or allowing them to say "maybe" or "not sure" and then allowing researchers to potentially make inaccurate assumptions. Both options could be a threat to validity. In some surveys, I have sometimes used an open-ended option ("if you checked "not sure," please tell us a little more/explain/etc.") but that can be unwieldy and adds to respondent burden. Best of all is to eliminate the need for DK/NS altogether as much as possible but sometimes, that's just... not possible.

I prefer to use other and leave space for them to make a comment. But it then might become a qualitative answer.