scores for comments

[andreaskutka]

The following features have been extracted, but no utility has been found for them in the testing data (too few observations). They require more testing data and further investigation to be converted into scores.

comment_length

Feature

f__comment_length contains the total length of comments set for an item, extracted from active interviewing events of type CommentSet in the paradata.

Rationale

Very short comments (e.g. length <= 3) or comments only containing numeric values are often due to interviewers writing the answer to the question into the comment. This may be due to questionnaire mistake, in which case we should see comments frequently for the item, or interviewers being confused, which we would like to flag. Longer comments may provide more information.

comment_set

Feature

f__comment_set contains the total number of comments set for an item, extracted from active interviewing events of type CommentSet in the paradata.

Rationale

In principle, comments should provide additional information to the Supervisor/HQ/data user, e.g., when the interviewer cannot solve a problem or wants to confirm a unusual answer. Item level anomalies from other features with comments set for the same item may be less of an issue. If comments are frequent, the absence of comments may be suspicious.

comment_duration

Feature

f__comment_duration is constructed similar to answer_duration, summing instead the intervals for all events of type CommentSet.