Adding New Sample Enterprises to the All Enterprises Tankan*
March 2, 2000
Bank of Japan
Research and Statistics Department
From the March 2000 Tankan survey, 50 additional enterprises will be added to the All Enterprises Tankan (Short-term Economic Survey of Enterprises in Japan). As a result, the total number of sample enterprises for the March 2000 Tankan will be 9,205 enterprises.
In order to prevent a decrease in the statistical accuracy from a decline in the number of sample enterprises due to bankruptcies, mergers, and so on, the following procedures are adopted for the All Enterprises Tankan. First, we regularly check (once a year) the statistical accuracy to see whether the error ratios of the population estimates of sales are within the target range (3 percent for the manufacturing industry and 5 percent for the nonmanufacturing industry) and whether the distribution of the sample enterprises has diverged from that of the population enterprises. Second, when the statistical accuracy has decreased, we add new enterprises sampled at random. (Refer to "The Methodology of the Sampling and Aggregation of the All Enterprises Tankan" for details.)
This time, we have checked the statistical accuracy of the last Tankan survey (December 1999), and decided to add another 50 sample enterprises. The details are as follows:
(1) Error ratios of the population estimates
The error ratios (of sales1) of the population estimates all remained within the target range for each of the six divisions by industry and scale.
1 Population estimates based on sample surveys, like the All Enterprises Tankan, inevitably have some estimation error. Thus, analysts need to take this into consideration when they use the data. The All Enterprises Tankan is designed for the actual figures to be basically within the range of 3 to 5 percent of the population estimate.
|Error ratios (of sales) in the December 1999 survey||unit, %|
(2) Distribution of the samples
In the All Enterprises Tankan, there are 118 divisions by industry and scale (divided by the number of regular employees). Each division is checked regularly to see whether the distribution of the sample enterprises (distributed according to the number of regular employees) appropriately reflects the distribution of the population enterprises.
In 8 divisions, the distribution of the sample enterprises diverged from that of the population enterprises.2 Moreover, there were some divisions whose distribution of sample enterprises indicated a high likelihood of diverging from the real conditions of the enterprises in line with the future decline in sample enterprises. To maintain the statistical accuracy, we decided to add additional sample enterprises to these divisions (50 enterprises; refer to the appendix).
2 The All Enterprises Tankan adopts the "test of the goodness of fit." This method tests the null hypothesis, H0"The distributions of the sample and population enterprises are the same pattern." When the hypotheses are rejected, or when the number of sample enterprises needed for the test is unsatisfactory, we judge that there is a difference between the distribution of the sample enterprises and that of the population enterprises.