Measurement and Application of Income Variables in Political Science Survey Research
As a key indicator of socioeconomic status, income plays an important role in social sciences and has been a routine socio-demographic item on many opinion polls. However, in Taiwanese political science studies, the use of the income item is far less common than the use of other socio-demographics such as gender, age and education. Data quality is arguably a major problem that puts researchers off from using income data from surveys. In this paper, we examine this problem for the purpose of laying the foundations for future research to improve survey measurement and application of income variables.
First, we discuss the issues of income measurement at every stage of the survey process from a methodological perspective and with the support of empirical evidence from various fields of literature. In seeking solutions to those issues, we trace the evolution of income measurement of the American National Election Studies from the 1940s onwards, along with several long-standing, large-scale academic survey projects in Europe. This review identifies a number of designs aimed at a more comprehensible wording of the income question, a more standardised procedure to elicit income information, and a more sensible way to keep income responses confidential. We recommend Taiwanese pollsters apply these potentially useful designs in future surveys for better income measurement.
Second, we investigate the existing income data of Taiwanese surveys. Our analysis of the 2001-2017 data from Taiwan's Election and Democratization Studies (TEDS) found noticeable discrepancies between official statistics and survey estimates of the average population income. Further analysis suggests that those discrepancies are partly due to the positive correlation between income and unit nonresponse — in line with our previous methodological discussion. In addition, there has been a non-trivial amount of item nonresponse in the income data of TEDS, which adds another layer of complexity to the use of those data in research. We suspect that these problems are not unique to TEDS, but rather common among political opinion polls in Taiwan, because most of them have measured income in a similar way.
Despite those problems, we found that the income variables of TEDS are useful as a strong indicator of respondents' subjective socioeconomic status, hence making a good complement to objective indicators such as education and occupation. This finding, together with those aforementioned, provides a general guide to using the existing income data of political surveys in Taiwan: avoid treating them as a measure of absolute wealth, but use them as an indicator of socioeconomic status. In this regard, these income data, though imperfect, should still serve the needs of many social science studies.
Last, with respect to using income data of surveys in research, item nonresponse is an inevitable problem. A common practice to handle this problem is the "complete-case method" — that is, excluding non-respondents from analysis and focusing solely on respondents without any missing data. In view of its widespread use, we highlight the pros and cons of this method, and make explicit reference to the condition under which it works and the condition it does not work in, in order to provide researchers a quick guide to making good use of the complete-case method to deal with item nonresponse in income data.