A statistic to describe a correlation with progression I want to see if the proportion of 'I don't know' responses correlates with the progression through the survey. Generally we expect this proportion to increase through the survey due to a fatigue effect. However, the following chart suggests this effect is not present in my data: 

I would however like to back this up with a statistic. I figure I need a correlation measure between the question order and proportion of DK. Does anyone know about such measure, and maybe how to produce it in SPSS (R could also be an option)?
PS: I eventually also want to control for question and respondent characteristics.
 A: Putting aside my comment on drop-outs, which should be considered together with this if there are any; I would start with a scatterplot of proportion of don't know versus question rank; and follow up with Spearman's rank correlation coefficient.
A better approach would also take into account complexity of questions.  One way of doing this might be to use data (if you have it) on how long it takes on average to answer each question.  This could then be put into a logistic regression model as an controlling explanatory variable (you would expect an uninteresting positive coefficient - more don't knows for harder questions), and sequence of the question as an explanatory variable (with the idea of testing that one for significance).  Neither effect is likely to be linear, so some careful model building would be required; a generalized additive model might be one option.  The same approach could incorporate other question characteristics too eg number of options.
Ultimately, as suggested in my comment, if you need to prove this conclusively one way or another the only way is to randomise the order of questions (or groups of questions).
