Nonresponse occurs if we do not get data for all respondents that were included in the sample. Nonresponse is one of the biggest problems of modern survey research because it is rising. A high nonresponse rate does not necessarily mean a high error rate. Errors that occur due to nonresponse rate happen because certain individuals were either not contacted or did not participate in the survey after they were invited. That can potentially be a problem. If the respondents are not representative to the general population due to nonresponse, the response rate must be increased if we wish to increase sample representativity.
We differentiate between unit nonresponse or element nonresponse (for chosen unit we didn't get any information about individual’s values) and variable nonresponse (we don't have answers to certain variables- the respondent answered to a survey, but didn't answer all questions - variables). These types of nonresponse have different causes.
Reasons for unit nonresponse:
- rejected invitation (principles, being too busy, etc.),
- participation not possible (sickness, deafness, foreign language, etc.),
- unit absence (travel, business trip, etc.),
- unit cannot be located,
- unit cannot be contacted,
- other reasons (lost questionnaires, etc).
Reasons for variable nonresponse:
- respondents do not know the answers,
- respondents refuse to answer (e.g. sensitive questions),
- haste or mistakes of the interviewer (the question is skipped, the answer is not written down by mistake, etc.),
- inconsistent answers being excluded at a later stage.
In addition, several generic reasons for nonresponse exist. One of them can be sending invitations unsuccessfully (e.g. the e-mail is recognised as spam).
Due to the consequences of nonresponse (bias and inaccurate estimates) it is important to decrease it as much as possible by:
- using larger samples (where the nonresponse rate is taken into account beforehand),
- attempting to prevent nonresponse (e.g. high quality questionnaires and/or interviewers),
- replacing missing units with other units from the same stratum during the process of data collection,
- weighting units (data) from strata with low response rates with reciprocal proportional values (but only if the right conditions exist),
- replacing missing variable values, e.g. by inserting stratum average values, by using methods of multiple insertion (regression, EM algorithm), etc.
Web survey tools offer automatic prompt verification of response validity. Advanced software offers criteria of verification such as verifying nonresponse that enables question design in a way that some questions must be answered (mandatory questions).
1KA offers easy implementation of answer validity checking. If a respondent does not answer a question, the application can display a reminder (soft reminder) or even not allow the respondent to continue without answering (hard reminder). This represents an important way of improving the quality of collected data.