The Unit Non-respondents to Telephone Survey in Taiwan Kuagn-hui Chen, National Chung Cheng University, Taiwan
Su-Feng Cheng, National Chengchi University, Taiwan
Question:
Researchers have been perplexed by the issue of unit non-response for a long time. It is common that selected
respondents do not participate in the interview for such reasons as failure to make contact with the respondents, refusal of the respondents to participate, and inability of the respondents to participate (e.g., language barrier and absence). In addition to respondents’ socio-demographic background and psychological predisposition, the social environment where the survey was taken, the design and topic of the survey, and interviewers’ skills and motivation also influence whether respondents would participate when they are contacted (Groves and Couper 1998, 29-42).
If the occurrence of unit non-response is not randomly distributed, some factors systematically determine whether a selected respondent will successfully participate in the survey.
Individuals with specific characteristics are more likely to be non-respondents. When these characteristics are related to the issues the researchers are studying, the presence of unit
non-response will diminish the representativeness of the final
sample and lead to biased estimation, and researchers cannot draw valid inferences from the sample coefficient to the population parameter of interest.
A variety of methods and techniques have been introduced to correct the bias resulting from unit non-response. It is common for researchers to use the method of weighting in an attempt to reduce unit non-response bias. However, since researchers have little or no information about the unit non-respondents, it is unlikely for them to examine the difference between participants and non-respondents in order to estimate the loss of
representativeness of the final sample as compared to the target population. Therefore, the lack of information regarding the absentees makes the unit non-response bias very difficult to deal with in a satisfactory manner.
We need to know more about the unit non-respondents in order to properly assess the possible impact of unit non-response. This paper is thus designed to describe the demographic characteristics and political attitudes of the non-respondents with the assistance of the information provided by the unit non-respondents’ spouses in a telephone survey conducted in Taiwan. Non-respondents and participants are compared with regard to their demographic characteristics and political attitudes. If non-respondents and participants are similar, unit non-response does not hurt the representativeness of the collected data. If there exists significant difference between the two groups of individuals, the
33
issue of representativeness is needs to be further examined and addressed.
Data:
A telephone survey data set is analyzed in this paper to examine the characteristics of unit non-respondents. This telephone survey was conducted by the Election Study Center at National Chengchi University in Taiwan in June 2011, and a national representative sample of 354 pairs of husbands and wives were successfully interviewed in this survey. The questions asked in this telephone survey include each participant’s demographic background, interaction between husbands and wives, political and social attitudes, and perception of spouse’s political attitudes.
In addition to these participated couples, while we also
successfully interviewed 804 married adults at the same time, their spouses failed to participate in the interview because of inability, refusal, or unavailability. In other words, these absent spouses were unit non-respondents, when the unit of analysis is individual respondent.
Table 1 shows the reasons for non-response. While nearly half of them did not participate because they were not successfully contacted by the interviewers, 37.8% of them immediately refused to participate when they were contacted or refused to continue during the interview. The remaining 13.7% of non-responses happened because of other unsure reasons or technical problems. In addition to the demographic variables such as gender, ethnicity, income, and
occupation, we inquired their spouses about the non-respondents’
political attitudes such as party preference and political support.
We are therefore able to examine the non-respondents’ political attitudes based on the information provided by their spouses.
Table 1. Reasons for unit non-response
Reasons n %
Non-contact Phone not answered 111 13.8
Unavailable 279 34.7
Lack of co-operation: refuse to participate 304 37.8
Others 110 13.7
Total 804 100.0
In this paper, the two data sets will be analyzed and compared to assess the characteristics of non-respondents regarding their demographic characteristics and political attitudes. The results of this analysis will be helpful for researchers to estimate the impact of unit non-response on representativeness of the final sample with regard to their political attitudes.
Analysis Strategy:
Given the availability of the 804 non-respondents’ information provided by their spouses who participated in the survey, we develop the following analysis strategy shown in Figure 1 to examine the characteristics of the non-respondents. The analysis strategy is based on the idea that the two sample data sets under study are equivalent and representative of the target population, if there is no unit non-response in both data sets. This analysis strategy is composed of three steps of comparison.
35 Figure 1. Analysis Strategy
First, the demographic variables of the 354 pairs of husbands and wives (hereafter referred to as couple data) are compared to those of the 804 individual wives or husbands (hereafter referred to as individual data). If the occurrence of unit non-response in the individual data is randomly distributed, the participants and the non-respondents should be similar, and therefore there exists no significant difference between the individual data and the couple data (relationship A in Figure 1). Second, demographic background and political attitudes of the non-responding spouses in the individual data set are displayed in accordance with the information provided by the participated married individuals. (relationship B). Third, the non-respondents in the individual data set are compared to the respondents of the couple data to see if unit non-respondents are different from respondents between the two data sets, and to further assess the possible concern of biased
estimation resulting from unit non-response (relationship C).
Data Analysis:
Following the three steps of analysis mentioned above, we compare respondents of the two data sets. If both samples were originally equivalent when they were randomly selected, and if the unit non-response randomly happened during the interview,
respondents of the two data sets should be similar. As it is shown in Table 2A, sex is an important factor contributing to the occurrence of unit non-response. While there are even number of males and females in the couple data, the ratio of men to women in the individual data is 4:6. As compared to females, males are more likely be unit non-respondents.
Table 2A. Demographic characteristics of the two data sets: sex Data set:
Sex Individual Couple Total
Male n
χ2=12.51 (Yates’s continuity correction) d.f.=1 p<.001
Since the original unit of observation in this survey is couple, instead of individual respondent, the variable of sex is controlled in the rest of analysis. All the respondents from the two data sets are compared in terms of their demographic background, and the results are listed in Table 2B.
Table 2B. Demographic characteristics of respondents: education, age, and sex
Male Female
Education Individual Couple Total Individual Couple Total Elementary n
37
Age Individual Couple Total Individual Couple Total
20-29 n
Social class (Occupation)
Individual Couple Total Individual Couple Total Upper middle
white collar n
Middle lower white collar
n
Individual Couple Total Individual Couple Total
Hakka n
Overall, there exists no substantial gap between the two groups of respondents with regard to their demographic characteristics.
None but one difference reaches the .05 level of significance.
Males of the couple data set are significantly different from their counterparts of the individual data set regarding their level of education. Male respondents of the couple data are more likely to be best educated than those of individual data. This difference in education between the two groups of males reveals that best educated males are more likely to be unit non-respondents.
In addition, age seems important to a certain degree affecting whether selected respondents would participate in the survey, although the difference is not significant. The same pattern regarding age is found in men and women: there are more people in their thirties in the couple data than in the individual data. This difference indicates that people of this age are more likely to be unit non-respondents. Since the target population of the survey was married adults in Taiwan, there were only a few couples who were in their twenties. Therefore, it seems reasonable to infer that younger people are more likely to be unit non-respondents.
Although respondents of the individual data set and those of the couple data set seem similar with regard to their ethnic background and social class, the two groups of respondents are significantly different in term of sex and education. Males, best educated males in particular, are more likely to be unit
non-respondents. Furthermore, although the result shows that age
39
cohort is not significant at .05 level, it is obvious that younger people are also more likely to be non-respondents.
The next step of analysis focuses on the information provided by the respondents of the individual data set to investigate who the unit non-respondents are (relationship B in Figure B). Each respondent was asked by the interviewers about his or her spouse’s occupation, ethnic background, Taiwanese/Chinese identity, partisanship, and evaluation of the two major presidential candidates, Ma and Tsai. Table 3 lists the male and female non-respondents’ background and attitudes reported by their spouses.
Table 3. Information about non-respondents provided by spouses Social class
(Occupation)
Male Female Total
Upper middle white collar
n
Middle lower white collar
n
Non-response n
column %
Taiwanese n
Non-response n
column %
Independent n
column %
Non-response n
column %
Non-response n
column %
Non-response n
column %
With regard to social class, recoded from their occupation, more than forty percent of male non-respondents belong to upper-middle class, and nearly a quarter of them are blue collar workers. This
41
is reasonable because work is an important factor that hinders the selected male respondents from participating in the survey. It seems another story among the female non-respondents, and a substantial number of them are housewives, so most of them are listed in the last category. On the on one hand, ethnic background of the non-respondents reported in Table 3 reflects the general
distribution of ethnic groups in Taiwan. On the other hand, there are less mainlanders than we usually observe in survey reports.
It is possible that mainlanders are more likely to be
non-respondents. With regard to partisanship, most of the less but similar proportion of non-respondents support the pan-blue camp.
In terms of their evaluation of the Ma and Tsai, the two major presidential candidates in 2012, there are about the same proportion of non-respondents holding positive and negative attitudes to the two major political figures.
The above distribution of non-respondents’ attitudes needs to be compared to that of participants in order to assess whether the two groups of people are politically different. The following analysis compares the unit non-respondents of the individual data and the respondents of the couple data (relationship C in Figure 1). We focus on the factors related to the political cleavage in Taiwan. These factors are ethnicity, partisanship, and
Taiwanese/Chinese identity. In addition, we also compare their evaluation of Ma and Tsai. The results are reported in Table 4.
Table 4. Comparison of non-respondents and respondents
Male Female
Ethnic
Total Non-resp ondent
Identity Non-resp
ondent
Respon dent
Total Non-resp ondent
Partisanship Non-resp
ondent
Respon dent
Total Non-resp ondent
Independent n
column %
Evaluation of Ma
Non-resp ondent
Respon dent
Total Non-resp ondent
Evaluation of Tsai
Non-resp ondent
Respon dent
Total Non-resp ondent
Respon dent
Total
Strongly like n 25 25 50 31 27 58
43
It is shown in Table 4 that respondents and non-respondents are significantly different in terms of their Taiwanese/Chinese identity, partisanship, and evaluation of Ma. While 61% of male non-respondents are reported by their wives as Taiwanese, only 45.7%
of male respondents have this identity. Therefore, the presence of unit non-response leads to an under-estimation of Taiwanese identification and an over-estimation of dual identification among the male respondents. A similar pattern appears on the male non-respondents’ partisanship, which is related the issue of identity: they are more likely to be pan-green supporters to a certain extent, although the difference is not statistically significant. Related to the issue of identity and partisanship, the majority of male non-respondents dislikes Ma, who belongs to the pan-blue camp and is viewed as a pro-China political figure.
However, female non-respondents and respondents are more politically similar than males. The only one exception is partisanship. Different from male non-respondents, female
non-respondents are more like to be independents and less likely to be pan-blue supporters. In brief, it is obvious that male non-respondents are politically different from male respondents, and this may lead to biased estimation of the population parameter.
Conclusion and Discussion
In summary, it is found in this analysis that such demographic characteristics as sex, education and age are related to the phenomenon of unit non-response. Males and young adults are in general more likely to be non-respondents in telephone survey. It is also found that the best educated men are also more likely to be non-respondents. The three demographic factors that lead to the unit non-response reflect the fact that males, young people, and better educated individuals are usually busy with their work and career, so they are less likely to be available or less willing to cooperate to participate in the survey.
Furthermore, we examine the political attitudes of the unit non-respondents and find that three political attitudes that are related to the political cleavage in Taiwan are associated with unit non-response among males. That is, males who identify themselves as Taiwanese, support the Pan-green camp, and dislike Ma are more likely to be non-respondents.
Briefly speaking, not only demographic variables but also political attitudes are related to one individual’s possibility becoming a unit non-respondent in telephone survey. It is common for researchers to use the method of weighting to adjust the sample and make the collected sample and population have the same distribution of demographic variables. The current analysis shows that demographic characteristics are related to unit non-response, so using demographic variables to compute the weight of each case
45
in order to make the sample looks like the population is a reasonable way to make the sample more representative of the population.
However, since political attitudes are also related to the phenomenon of unit non-response, the method of weighting cannot be a satisfactory solution. This analysis shows that males who are on the opposite sides of the political cleavage have unequal possibilities being a unit non-respondent. This bias cannot be properly addressed with the information of demographic variables.
We suggest using household instead of individual person as the unit of observation in telephone interview, so that we can retrieve extra information of the non-respondents to estimate and correct the possible bias resulting from unit non-response.