What criteria to use when having with 'unengaged' , ‘straightlined’ or ‘patterned’ responses in questionnaires? Please consider the following situation, which happens a lot in real life applied research in social sciences. 
Some percentage of the respondents give ‘straightlined’ or ‘patterned’ responses to a self-reported questionnaire, consisting of several Likert scales. 
What criteria should be employed to make a decision as to keep or remove these respondents from the database? (I refer here to possible criteria related to percentages, type of patterns, etc.). Also, this is not a situation of missing responses, so no imputation is needed.
It should be noted, in my opinion, that a distinction must be made between scales containing only direct items and scales containing also reversely formulated answers (in the latter case, the decision would be simpler, I presume, since a ‘straightlined’ pattern in answers would indicate clearly either no cognitive processing of the question/item or malicious intent).
I’ve searched for references, but aside for taxonomies and how to identify these responses, I’ve come up empty regarding how to deal with them (except ‘throw them out’).
 A: In my opinion, I think the data should be analyzed as is. To some extent, these patterns are a reflection of the survey as well as a reflection of the population. Basically: crap in, crap out. Long winded questionnaires are guaranteed to fatigue respondents, sensitive or invasive questions will likely induce selective non-response, and younger participants generally do not have the same interest in sharing information, or as much information, as their older counterparts.
Secondly, it is impossible as a statistician to go about "data cleaning" by hand, or even looking at the data to inform a set of rules to anticipate just how badly some people will mis-respond on a survey. When you do this, you are very likely to "cheat" and p-hack so that findings conform to your own preconceived notions of reality.
This is the importance of interpreting the findings with a subjective nature. For instance, diet and physical activity questionnaires are notoriously bad. People lie, exaggerate, and do not respond. I did an analysis which found a positive correlation between reported physical activity and weight: fatter people exaggerated their level of physical activity to such an extent that they exceeded what a healthier person reported. Those results are not surprising when you remember that it is self-reported physical activity. I interpreted those findings directly, pointed out that I don't believe them, proposed improvements to the questionnaire, and called it a day.
You can certainly use some statistical methods--or just practical methods--to do a sensitivity analysis. I can't see how advanced algorithmic approaches would be beneficial, principal components analysis would probably reveal all you need to know.
