Are there predictors of not knowing and refusing to answer on surveys in Georgia?

Are there variables that predict who is likely to report “Don’t know” or to refuse to answer survey questions more often in Georgia? This blog post looks at this question, using un-weighted Caucasus Barometer 2017 (CB) data for Georgia.

Only questions that were asked to all respondents were considered during the analysis presented in this blog post (a total of 177 questions), and variables were generated for:

1. The number of times a respondent answered “Don’t know” and;
2. The number of times a respondent refused to answer a question.

On average, respondents refused to answer one question on CB 2017. The median number of refuse to answer responses was zero. Respondents answered “Don’t know” to nine questions on average, and the median number was five. As a previous blog post highlighted, people most often report they do not know when asked about political questions and areas of reasonable uncertainty (e.g. their economic futures). When it comes to refusing to answer, people are most likely to refuse to answer questions about their income and politics.

To analyze whether or not demographic variables predicted “Don’t know” and “Refuse to answer” responses, Poisson regression was used. The following demographic variables were included in both regressions: age group (18-35, 36-55, 56+), gender, ethnicity (ethnic minority or ethnic Georgian), settlement type (capital, other urban, rural), and level of education (secondary or lower, vocational, tertiary). Besides demographics, the analysis also included a variable for whether people besides the interviewer and respondent were present during the interview. After the regression analysis, the number of times a respondent would be expected to respond either don’t know or refuse to answer was calculated, controlling for other factors included in the models.

A number of demographic characteristics are associated with higher expected rates of “Don’t know” response. Ethnic minorities, people in rural areas and urban areas outside Tbilisi, people without tertiary education, women, and people over the age of 55 provide more don’t know responses than people in Tbilisi, those with tertiary education, and people under the age of 55. Besides demographics, whether someone is present at an interview that is not participating in it is also associated with how often people report they don’t know. If additional people are present at an interview, then respondents report one fewer don’t know response on average. This may reflect a large number of factors (e.g., maybe people with large families are more certain of their views), however, one plausible explanation is that people do not want to admit they do not know in front of other people.

Some demographic variables also predict “refuse to answer” response options. People in the 36-55 age group refuse to answer questions slightly more often than in other age groups as do people outside Tbilisi, those with tertiary education, and ethnic Georgians.  Besides demographics, the presence of people besides the interviewer at the interview has a significant impact on the frequency of refusing to answer. This again may reflect social pressure of some sort. Rather than respondents being worried about appearing uninformed, one plausible explanation is they are more likely to be worried about appearing socially uncooperative.

While don’t know answers and refusing to answer are both often treated as non-response, these arguably are different types of responses. If they indeed are different, one would expect them not to be strongly correlated. The results of a correlation analysis suggest a very weak (ρ=0.009) and non-significant association, supporting the contention that don’t know and refuse to answer options are indeed different types of responses rather than replacements for the other.

This blog post has looked at whether and which demographic groups respond “Don’t know” and “Refuse to answer” to survey questions more. The results suggest that a variety of demographic variables are significant predictors. In addition, the presence of other people at the interview appears to have an impact on how often people report they don’t know or refuse to answer survey questions, with both declining when people are present.

To download the data used in this blog post, click here.