INSTITUTE FOR GLOBAL PROMINENT RESEARCH

SMALL AREA ESTIMATION AND ITS APPLICATION

IMPROVING DATASETS USING SMALL AREA ESTIMATION

-Statistical techniques can improve the accuracy and scope of limited information, saving time and money.
Research Keywords: Economic Statistics, Baysian Statistics, Small Area Estimation

Collecting comprehensive data on populations is not always possible, or even the best use of resources. Surveys can be expensive and time-consuming, so improving the accuracy of data with “small area problems” can stretch data resources further.

“Small area problems happen when an estimator, such as sample mean for a small area, becomes very unstable because of small sample sizes,” explains Associate Professor Genya Kobayashi, who is part of a new Chiba project called Small Area Estimation and Its Application to the Analyses of Poverty, Public Health and Disaster.

His group is focused on overcoming such issues by developing small area estimation (SAE) techniques that borrow statistical strength from supplementary data.

OVERCOMING INEQUITY AND PUBLIC HEALTH INSIGHTS

While the applications of SAE are broad, “small area problems arise particularly within official statistics on which governments make important political decisions. Hence, small area estimation techniques could lead to more effective policy making,” notes Kobayashi.

An example crops up in the monthly Family Income and Expenditure Survey (FIES) conducted by Japan’s Statistics Bureau. Although FIES provides the latest information on household income and expenditure, at a small area level, such as a city level, the sample size of FIES becomes so small that direct estimators are unstable and can’t grasp detailed trends.

But FIES, says Kobayashi, can borrow statistical strength from data in the National Survey of Family Income and Expenditure (NSFIE). The sample sizes in the NSFIE are much larger than in the FIES and thus more reliable, but because the survey is time consuming and costly, it’s only carried out every five years.

The FIES and the NSFIE collect area-level aggregated data, so Kobayashi’s group used the Fay-Herriot, a mixed regression model routinely used in SAE.

However, when data is in other formats more appropriate SAE models must be developed. Recently, for example, Kobayashi’s group has been working to improve an income map of Japan based on the Statistic Bureau’s Housing and Land Survey (HLS). Using a new statistical model that accounts for the grouped data format of the HLS and additional data from the Population Census, they measurably reduced the uncertainty of the HLS’s income estimates.

SAE’s scope isn’t limited to stabilizing data, adds Kobayashi. SAE techniques can add information. For example, the HLS model could also infer income in areas where data is missing. Using a mixture modeling approach, the team recently estimated the distribution of land prices around train stations in and around Tokyo by making inferences across areas.

Contributing to public health discussions, in 2018 Kobayashi collaborated on a paper that used a spatial Poisson regression model to look the links between access to clean water and dropping mortality rates in the early 1900s. Kobayashi and collaborators at the Tokyo Institute of Technology and Duke University in the United States looked at death rates in Tokyo in 1930 against a dataset of taps. The group found that access to clean water accounted for 41.3 percent of the improvement in Tokyo’s crude death rate between 1921 and 1937. This paper adds to discussions on the relative importance of nutrition and clean water on mortality. Kobayashi hopes that everyone from policy makers to businesses will soon come to the group to add a finer resolution to their datasets.

Members

Principal Investigator

Co-Investigatior