# Statistics

Home > Statistics > Outline of Statistics and Statistical Release Schedule > Notices of Changes and Corrections 2008 > Regular Addition of New Sample Enterprises to TANKAN

# Regular Addition of New Sample Enterprises to TANKAN

February 21, 2008
Bank of Japan
Research and Statistics Department

From the March 2008 TANKAN (Short-term Economic Survey of Enterprises in Japan), 113 enterprises will be added as samples of the survey.

We examine the statistical accuracy of TANKAN once a year. This is to prevent a decline in the statistical accuracy due to a decrease in the number of sample enterprises caused by bankruptcies, mergers, and other factors. We add new sample enterprises to TANKAN when the examination reveals that statistical accuracy has in fact declined (see "Sample Design and Sample Maintenance of TANKAN," June 7, 2004, Research and Statistics Department, Bank of Japan, for details).

We have decided to add 113 enterprises to TANKAN (see appendix) after the recent examination of the statistical accuracy of the last TANKAN (December 2007). As a result, the total number of the sample enterprises for the March 2008 TANKAN will be 10,913 (including 208 financial institutions). The details are as follows:

## (1) Sample Distribution

Population enterprises are classified into 383 strata by industry, scale based on capital size, and number of regular employees to calculate the population estimate.

For each stratum subject to the sample design1, we examined similarities between the distribution of sample enterprises and that of population enterprises. The result revealed that there was no stratum in which the distribution of the sample had deviated from that of the population2.However, there were strata of which the sample distribution indicated a high likelihood of deviating from the distribution of the population with further decline in the number of sample enterprises.

To maintain statistical accuracy, sample enterprises have been added to these strata.

1. Strata with less than five enterprises (46 strata) have been excluded.
2. The "test of the goodness of fit" using chi-square distribution is used in the regular examination. This method tests the null hypothesis, H0: "The distribution of the sample and that of the population have the same shape." When the hypothesis is statistically rejected, it is judged that a deviation exists between the distribution of the sample enterprises and that of the population enterprises.

## (2) Error Ratios of the Population Estimates

For all of the main six divisions by industry and scale (Manufacturing/Nonmanufacturing and Large/Medium-sized/Small, see below), the error ratios of the population estimates3 (measured by sales data), after the addition of 113 samples, remained within the target range (Manufacturing: 3 percent, Nonmanufacturing: 5 percent).

1. 3Population estimates based on sample surveys, like those in TANKAN, inevitably entail certain estimation errors. Users are advised to take this into consideration when using the data.