Geographical Patterns of “Not Reported” Cases in Small Area Statistics of the 2010 Population Census

Abstract

Population censuses are the only surveys that have been conducted for all residents across Japan, and the data have often been used in geographic studies for creating sociodemographic indices at the neighborhood level (cho-cho aza). However, there have been concerns about the increase in “not reported” cases in recent population censuses, which are nonresponses due to a refusal to complete the survey. In this study, we aimed to: 1) examine geographical patterns of the not reported cases at the small area level; 2) discuss their influence on regional analysis; and 3) explore some ways to deal with this problem. Using the 2010 population census, percentages of not reported cases were calculated for the five variables of age, marital status, labor force status, education, and past residence that have been repeatedly used in geographic studies. We then aggregated the percentages of not reported cases by the degree of urbanization on multiple scales and visualized their geographic distribution using maps. Finally, multilevel negative binominal regression models were used to estimate incidence rate ratios of the number of not reported cases in neighborhoods using the degree of urbanization on multiple scales as independent variables and municipalities as a group variable. The results showed that there were clear urban–rural gradients in the distribution of the not reported cases, with the highest percentages in urban areas and the lowest in rural areas. We also found significant municipality-level variance (spatial clustering) in the not reported cases as a result of the multilevel analysis. This may suggest that the survey methods and resources utilized by municipalities, which are in charge of conducting census surveys, could influence the survey responses of residents and potentially introduce systematic errors into the data. These results emphasize the necessity of taking into account the bias caused by the not reported cases when conducting small area analysis using population censuses. Exploring the determinants of the not reported cases by conducting social surveys independently from the census as well as considering statistical methods of adjustment for the missing cases are also required.