Using National Health Insurance Database for Sampling Surveys
Censuses are usually the only method for collecting information on a national population, and they are conducted every ten years for most countries. Taiwan's government also conducts censuses every ten years to collect information on the de jure population, covering the shortage of the household registration system. However, there are potential problems in the traditional census, such as low response rates, poor data quality, outdated information and rising survey costs. Taiwan, like a lot of countries, has been seeking alternative methods for collecting national data and decided to use sampling surveys (of about 16% of the national population) to replace the whole population census in 2010 and 2020, in order to cope with the problems of low response rates and rising survey costs. Still, the sampling survey cannot provide timely population information in Taiwan. The rolling census is one of the new methods which can offer up-to-date information, and was first adapted by the U.S. and France in 2010. The rolling census in the U.S., American Community Survey, collected data on 3% of the national population annually for five years, and this can deliver a time series of national estimates for those five years.
In this study, we aim to explore new data collection methods which can serve as an alternative to the traditional census and provide timely population information. In particular, our goal is to evaluate the possibility of using data from the National Health Insurance (NHI) Research Database for acquiring information on the de jure population in Taiwan. In particular, we use the Longitudinal Health Insurance Database 2005 (LHID2005) in this study. In addition to records on upper respiratory tract infections, we also propose three other methods based on the records of outpatient visits to identify the place of normal residence. We choose outpatient rates and other criteria from the NHI data for method evaluation, and find that the medical cost of outpatient visits less than 555 points (one point is close to one New Taiwan Dollar) is the most appropriate method.
To demonstrate the feasibility of using NHI data to acquire the population information, we apply the proposed methods to explore the domestic migration in Taiwan. The official records of Taiwan's domestic migration are available only at the level of province and special municipality, and are not accessible for counties and townships. To make up for the deficiency of official records, we estimate the domestic immigrants and emigrants at the level of electoral circumscription (73 seats) for members of the Legislative Yuan. The population of each electoral circumscription is between 200,000 and 500,000, smaller than that of counties and larger than that of townships. The results show that the domestic migration in Taiwan can be separated into three sub-regions: northern, central, and southern Taiwan. The migration activities are active within each region, but not as active between regions. Also, the six major cities (or special municipalities) in Taiwan, each with population of at least two million, attract most of the domestic immigrants, and the remote areas (such as eastern counties) have the highest emigration rates.