(Un)bounded rationality of inequality recognition: Perception through the use of the stratification index and percentiles
How to cite
Nakabayashi, Masaki; Kawata, Keisuke; Okubo, Shoki; Fujihara, Sho. 2021. "(Un)bounded rationality of inequality recognition: Perception through the use of the stratification index and percentiles." AEA RCT Registry. AEARCTR-0008317. https://doi.org/10.1257/rct.8317-1.2000000000000002
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Our decision-making behaviors toward public policy depend on our perception of society. Preferences for policies that address inequality and poverty rely on how we perceive inequality. Thus, it follows that our preferences could depend on our cognitive capacity to perceive facts. If our capacity is large enough compared to the presented measurement units used to describe society, then such measurement units are irrelevant to our decisions regarding public policy. However, if our cognitive capacity is bounded across the measurement units used to describe society, then our political decision-making in fact begins with the choice of measurement unit used to describe reality prior to making decisions that are recognized to be sensitive to partisanship.
Thus, whether measurement units based on the same fact affect our perceptions or impressions of reality is critical to examine further, i.e., beyond behavioral and experimental economics and extending toward social sciences in general and policy-making.
Income inequality is a keen issue in both developed economies and relatively wealthy emerging economies. We need to address this issue. Thus, understanding how people feel about inequality from different measurement units based on the same fact is critical to policy-making in order to address the issue. To investigate how we perceive income inequality, we compare two measures of inequality, namely, the stratification index (Zhou 2012) and percentiles.
We use data on income distribution in Japan from 1985 and 2018, which were obtained from the National Livelihood Survey conducted by the Ministry of Health, Labour and Welfare of the government of Japan. Using this data, we calculate percentile proportions and the stratification index.
Then, we implement a randomized conjoint experiment on the internet. We show the household income distribution of households with children and households of elderly people of two "societies"; one's income distribution from 1985 to 2018 is described according to percentile proportions and the other one's income distribution from 1985 to 2018 is described according to the stratification index. Both measures actually describe the same Japanese society in 1985 and 2018. Then we ask respondents which society seems to have become more unequal between 1985 and 2018.
If we find different evaluations between two descriptions according to income percentiles and stratification index, that means that our perception of inequality is susceptible to the presentation of statistics of the very same distribution. Otherwise, we would conclude that the human recognition of inequality does not show a significant difference between the percentile measure and stratification index presentations and that we need more trials to investigate whether the irrelevance of measurement units applies to a broader range of measures of inequality.
A policy implication of this study is that if the former result is obtained, then it can be seen that we essentially begin to decide a public policy choice when we calculate a measurement of inequality and present this measurement to the public, which occurs far before citizens and their representatives begin to discuss such policies consciously. Otherwise, citizens who are exposed to different measurements are more likely to achieve a consistent perception of a certain issue and share their recognition of the facts, which sets a better ground for policy-making.
Zhou, Xiang (2012) "A nonparametric index of stratification," Sociological Methodology, 42, 365-389. https://doi.org/10.1177/0081175012452207
Intervention
We create two measures of household income inequality using the same distribution of household income data obtained from the National Livelihood Survey conducted by the Ministry of Health, Labour and Welfare of the government of Japan in 1985 and 2018. The measures are percentiles and the stratification index. We randomly cite a point from each measure of 1985 and 2018 and show these points to respondents. We do not indicate that the measures are drawn from the same distribution but rather explain that each point measures the inequality of "a" society. For randomization, we adopt a randomized conjoint experimental design. Looking at the two descriptions according to the percentile and stratification index that are generated by the randomized conjoint design, the respondents are then asked to evaluate which "society" became more unequal between 1985 and 2018. Our intervention is a combination of two samples, each of which is based on the percentiles and the stratification index taken from the same household income distribution. Thus, our intervention is based on a factorial design.
Experimental design
We create two measures of household income inequality using the data obtained from the National Livelihood Survey conducted by the Ministry of Health, Labour and Welfare in 1985 and 2018; these two measures are the percentiles and the stratification index. We cite 5 points from each measure in both 1985 and 2018. We randomly pair a point from each measure, show them to each respondent (who is not informed that these are from the same society), and then ask the respondent which "society" seems to have become more unequal between 1985 and 2018. For the randomization in regard to pairing the measures, we adopt a randomized conjoint experimental design as a factorial design. We assign each respondent five tasks of choice between the randomly matched pairs of inequality measures.
Along with the experiment, we collect information about the respondents' background characteristics such as age, gender, prefecture, working status, employers' size, job title, education, income, household income, political preferences, partisanship, whether reading a newspaper, Internet news, or books, and self-perception of social class. Considering the possible heterogeneity of intervention effects across the background characteristics, we deploy a generalized random forest algorithm.
Design details
Cited by
No citation data available for this registration yet.