The reliability of the data collected through a well structured questionnaire (instrument) can be statistically high but may not be free from various biases; especially from the respondents side. Generally, the data is being collected by distributing the questionnaire to the respondents and then collecting it from them on a suitable time, and you can only request them to give to their best (time and effort) to fill it in the best possible manner. But, whether they have filled it with full interest, that is, reading all the questions carefully, and understanding them thoroughly, is not guaranteed. Even after removing the outliers you cannot catch all misjudged/ overlooked and answers ticked without understanding the question/s completely can leave the whole exercise to be futile and unworthy for generalizations.

Therefore, for a better result, in terms reliability of the data I suggest that a group of respondents (may be a bunch or all at once) should be called on a common platform such as a get together or an outing and they should be made clear about each and every question of the instrument (or questionnaire) to be used in the analysis with proper discussion and query handling (if required) so that they get thorough understanding before actually filling it. Otherwise, I feel the data can never be without biases be it unintentional or not.