Survey nonresponse is a type of survey error that is caused when there is a discrepancy between the participants within the sample group who do not answer the survey and the participants within the sample group who do answer the survey. When put in more technical terms, we can say that non-response error refers to the variation between the true main values of the original sample list (those who receive invites to the survey) and the true mean values of the net sample (those who actually respond to the survey).
Within this article, we will explore the two key types of survey non-response as well evaluate the different approaches that can be used to treat it.
Types of Survey Non-Response
When studying survey non-response, we are examining two distinct but related phenomena:
Unit Non-Response
Unit non-response occurs when an entire observation unit is absent from a sample.
Let’s examine the different causes of unit non-response and its implications.
Causes of Unit Non-Response
There are many different causes of unit non-response and they may vary based on the mode of the survey being used. In face-to-face surveys, there are many physical barriers that are present when trying to access a household. Variables such as the patterns of times of when members are present in a household as well as the timing and frequency of attempted interviewer contacts can influence the non-response rate.
Refusal patterns also generally follow specific patterns, with age, socio-economic status, and socio-economic environment influencing willingness to respond.
Respondent Characteristics
- Age: Older and younger respondents are more likely to cooperate relatively to middle-aged respondents.
- Socio-Economic Status: Individuals with low socioeconomic status are more likely to cooperate relatively with those with high status.
- Socio-Economic Environment: There is generally lower respondent cooperation among high-density areas.
Interviewer Characteristics
Experienced interviewers who are able to develop a good rapport with respondents at first contact are likely to secure interviews.
Survey Design
Monetary incentives and/or providing an advance warning of the interview using survey software can increase the likelihood of respondents cooperating.
Item Non-Response
Item-nonresponse, on the other hand, occurs when one or more measures of interest are missing while the other required measurements are present.
Let’s now examine the different causes of item non-response.
Causes of Item Non-Response
When considering the causes of item non-response, a pivotal factor that we must consider is the ‘don’t know responses. The determinants of item non-response can be categorized into the following three groups:
Respondent Characteristics
A respondent’s personal characteristics have the biggest impact on whether they select the ‘don’t know option. The following factors impact the rate of ‘don’t know’ responses:
- Respondent exposure to the interview topic
- Respondent’s level of education
- Respondent’s interest in the survey topic
Questionnaire Design Issues
This factor is out of the control of respondents and refers to when the ‘don’t know’ response is caused due to question-wording. Some respondents may find the wording of certain questions to be offensive, or even difficult to understand. They may select the ‘don’t know’ response simply because it is easier to do so.
Interviewer Behavior
Interviewer characteristics also influence unit non-response. Interviewers that are skilled and interviewers who believe it would be easy to administer a questionnaire are more likely to obtain responses than those who are not as skilled and believe it would be difficult to obtain responses. This affects modes wherein interviewers are present like CATI surveys, offline surveys, telephonic surveys, etc.
Implications of Non-Response
High Costs
The measures taken to reduce survey non-response are generally associated with very high costs.
Reduced Representativeness
The existence of unit non-response can potentially reduce the representativeness and reliability of the survey.
Potential Biases
Missing data can lead to survey results being biased and unreliable unless treated correctly.
Treatment of Non-Response
Researchers inevitably come across item and unit non-response when conducting surveys. Hence, they must know how to analyze incomplete data in the best way so as to not let it damage the reliability of results.
Treatment of Unit Non-Response
The way unit non-response is treated depends on the availability of information. There are two key methods used to treat unit non-response:
Model-Based Approaches
Model-based approaches, such as selection bias techniques, are used when researchers have a theory of the patterns of missing data as well as detailed information on the characteristics of non-respondents. With this approach, the researcher can account for the differences between the sample and the wider population.
Weighting Adjustments:
In most cases, researchers will only have information about the population relative to the sample in the form of auxiliary information that is extracted from the census. In such cases, the model-based approach cannot be used, and instead, weighting adjustments are applied to reduce the bias in survey estimates caused by unit non-response.
Treatment of Item Non-Response
Item non-response can also be treated using statistical techniques. There are two key methods used to treat item non-response:
Application-Specific Approach
The application-specific approach is employed when the researcher has a particular theory about the process using which individuals decide whether or not to answer certain survey questions. This information can then be used to model the missing data.
Imputation Procedures
In cases where there is an absence of a theory about the specific patterns of missing data, the imputation approach can be taken. Imputation procedures involve using different methods to estimate the values of the missing data. Once these values are ‘filled in’, researchers can go on to analyze the data using standard methods of analysis.
FAQs on Survey Non-Response
What are the two main types of survey non-response?
There are two key types of survey non-response. The first one is unit non-response, which occurs when an entire unit of data is missing, while the second one is item non-response where one or more measurements of interest are missing.
How to treat unit non-response?
There are two key methods that can be used to treat unit non-response and they are model-based approaches and weighting adjustments.
How to treat item non-response?
There are two key methods that can be used to treat item non-response and they are application-specific approaches and imputation procedures.
Follow Techiemag for more!
Table of Contents
