On emotive terms in questionnaires Together with some of my colleagues, I am responsible for an introductory course preparing students for university studies in mathematics. For this course we have the policy that if a student finishes the course, we ask him or her to evaluate the course by answering what we call a course evaluation.
Two years ago I and my colleagues updated the course evaluation. The results have been stunning in the sense that we get more responses and more useful feedback. However, there are some problems with the course evaluation we are currently using, I think, though not all of my colleagues agree. Let me present my problem.
My knowledge in survey methodology says that one should not consistently ask only questions formulated in positive terms, and vice versa for negative terms. Instead, items in a questionnaire should be phrased both positively and negatively, for otherwise the survey may suffer from acquiescence bias. 
For example, in our course evaluation we make the statement

The course web page was clear and well designed. [Translated from Swedish.]

and ask them to grade how much they agree with this statement.
Another statement is

The assignments helped me in my learning process. [Translated from Swedish.]

In trying to eliminate acquiescence bias, I suggested that we phrase the second statement as

The assignments disturbed my learning process. 

(Note that this is just an example to illustrate an idea.)
When answering these questions the respondent choose an integer between 1 and 5. 1 means "Do not agree at all" while 5 means "Completely agree" (both translated from Swedish). (The reasons for constructing the questionnaire in this way were based on a review made by Stockholm University.)
Most of the evidence I have seen for problems regarding acquiescence bias is based in psychological research. I have no problem with this, but my question is if there is any empirical and statistical study trying to examine or estimate the difference in how people respond to a course evaluation questionnaire if one consistently phrase the questions in positive terms, compared with consistently phrasing the questions in negative terms, or positive and negative terms 50/50?
(From our own questionnaire we see a strong tendency to pick integer 5, i.e. "Completely agree". My hypothesis is that some of this tendency can be attributed to acquiescence bias. Our sample contains around 270 observations, and we ask six questions of the sort described above.)
 A: Here is a study of negatively and positively phrased questions in a course evaluation: Bradley, Royal and Bradley (2008), "An Investigation of 'Honesty Check' items in Higher Education Course Evaluations." The authors conclude that including negatively phrased questions is more likely to introduce increase measurement error than counter acquiescence bias. 
You may also find this relevant: Stewart and Frye (2004), "Investigating the Use of Negatively Phrased Survey Items in Medical Education Settings" The study concerns a survey of students' perception of learning environment in a medical school, including a set of questions related to "meaningful learning environment", and concludes that negative phrasing is "associated with lower scale reliability." 
The negatively and positively phrased questions in both studies were all from the same survey, and were all answered by the same students, if I understand well, which may not be ideal. They do not compare different versions of a survey across equivalent populations. Do look at the works cited, though. Maybe there's something of interest there that you haven't come across yet. 
My two cents: You could simply measure some attitudes/perceptions with positive questions and some with negative questions, or have a survey with all negative questions, but then you will probably find yourself asking "What would I get if I phrased it the other way?" Even if you believe you've mitigated acquiescence bias, you may get something worse. The first thing to consider with asking negatively phrased questions to which people agree or disagree is that they can be confusing, because people need to reverse things in their heads; instead of saying "I disagree that X is the case", they have to say "I agree that X is not the case" - it's not a very intuitive way of thinking for us, and this may give you noise. 
Another way you can use negatively phrased questions is to have pairs of questions measuring the same attitude/perception, one of each pair being positively and the other negatively worded. With this, you can identify acquiescence and eliminate those who blindly vote "I agree" or "5" because their answers will be inconsistent. However, there are problems with this, too. The challenge is to "reverse" each positively worded question, but this is rarely straightforward. Take your example "The assignments helped me in my learning process". Is the negative of this "The assignments did not help me..." or "The assignments disturbed..."? The problem is that if a student responds to "The assignments helped me in my learning process" with "I disagree completely", this says nothing about whether the assignments "disturbed" the learning process. And vice versa! That the assignments did not "disturb" the learning process does not mean that they were helpful! So the two questions here really are measuring different concepts: to not be helpful is not the same as to disturb; the former is just an absence of a positive thing, whereas the latter is a presence of an explicitly negative thing. 
If you can, do consider the scale too in relation to each question. Maybe measuring a negative and a positive with one question is less likely to result in acquiescence bias. E.g. "How helpful or disturbing the assignments were for the learning process?" - Very disturbing/Somewhat disturbing/Neither disturbing nor helpful/Somewhat helpful/ Very helpful. (Although, I imagine this would present some challenges if you do things like factor analysis or principal component analysis with the data.)
