- May 24, 2019
- May 24, 2019
- May 24, 2019
Home > Statistics > Outline of Statistics and Statistical Release Schedule > Notices of Changes and Corrections 2004 > Revision of TANKAN in the March 2004 Survey -- Comparison between the pre- and post-revision in the December 2003 Survey
Comparison between the pre- and post-revision in the December 2003 Survey
March 8, 2004
Bank of Japan
Research and Statistics Department
Note: U-tags is used to draw underlines in this HTML document.
Click on nttk13.pdf (124KB) to download the full text.
TANKAN will be revised from the March 2004 Survey (to be released on April 1, 2004). In addition to the regular revision of sample enterprises which is conducted every 5 years, the revision includes some changes in the framework of the survey. The fundamental revision aims at improving the statistical accuracy by reflecting the industrial structure of the economy and incorporating the recent accounting system.
This paper introduces the main points of the revision and primary changes in the sample design of TANKAN. It also reveals the difference between the pre- and post-revision surveys in terms of aggregation results.
The five major changes will be introduced in this section.
The surveyed industries will be revised to reflect the recent structural changes in industries. We select the surveyed industry if it is related to the trend in business conditions and is profit-making. The classification of industries is decided in accordance with the Japan Standard Industrial Classification (Rev.11) released by the Ministry of Public Management, Home Affairs, Posts and Telecommunications in March 2002. Details of the revision in TANKAN are as follows (See Figure 1).
While the pre-revision TANKAN adopts the number of employees as the criteria for enterprise size, the revised TANKAN will adopt enterprises' capital as its criteria.2 This revision is to reflect the rapid diversification of employment styles in enterprises. For instance, there appear to be some enterprises with a small number of employees but large capital, such as holding companies and Internet-related companies.
Survey items are designed to reflect changes of economic conditions. On the other hand, we have to avoid duplication against other statistics and to decrease the burden on respondents. To accomplish these objectives, we will revise several items. Some examples are shown below (See Figures 2 and 3 for all revised points).
"Principal Enterprise TANKAN" will be abolished from the revised TANKAN. Since the surveyed enterprises3 are basically fixed and simple aggregates are used in "Principal Enterprises TANKAN," the results may diverge from the actual economic conditions due to the changes in the industrial structure. Because of this reason, "Principal Enterprises TANKAN" has been treated as a reference survey since the March 1999 Survey. As a result, a majority of TANKAN users have come to focus mainly on "All Enterprises TANKAN." Under this situation, we have decided to abolish "Principal Enterprises TANKAN" (sample enterprises of "Principal Enterprises TANKAN" will continue to be surveyed in "All Enterprises TANKAN").
Unanswered responses (missing value) have been treated simply: the responses were excluded from the aggregation. For the quantitative items, in which the population estimation method is taken for aggregation, this treatment corresponds to substituting the average figures of the stratum,4 to which the unanswered respondent belongs, for the unanswered items. This is because the population estimation method calculates the aggregated figure by multiplying the average figure by the number of the population.
In order to decrease non-sampling errors, we are making efforts to keep a high response rate by asking for the understanding and cooperation of responding enterprises. With these efforts, the current response rates for all quantitative question items are basically at a high level, so we have not had any problems with the current measure. However, we have been searching for a better measure for unanswered responses, in the event of lower response rates.
After the revision, we will introduce a new measure for unanswered responses for all annual projection items such as Sales, Current Profits, and Fixed Investment. This substitutes the most recent response of the corresponding enterprise to the unanswered items. According to our analysis, the new measure enhances the accuracy more than the current one as a whole.
To revise the survey, we have chosen new sample enterprises from the updated population based on the 2001 Establishment and Enterprise Census of Japan.5 We have adopted the sampling method described below. The method aims at achieving statistical accuracy with less burden on respondents and less staff members when compiling data.
In consequence of sampling, 3,831 sample enterprises are added and 1,187 are eliminated. As a result, there are 10,848 sample enterprises in the new TANKAN (Figures 4 and 5).
Changes in the survey framework and sample enterprises create a discontinuity between the time series data up to the December 2003 Survey and data of the March 2004 Survey. To recognize this discontinuity, we have conducted a preliminary survey for all sample enterprises of the new survey. With the survey, we reaggregated the December 2003 Survey on the new basis. There seemed to be no large differences between the pre- and post-revision surveys in both judgement items and year-to-year changes of the annual projection items such as Sales and Fixed Investment. Note that there are some gaps in several items such as Business Conditions DI for large nonmanufacturing. See "Data comparison between pre- and post- revision of TANKAN in the December 2003 Survey" (released on March 8, 2004) (nttk13a.pdf [35KB] or nttk13a.zip [46KB, MS-Excel]) for details.
The discontinuity between the pre- and post-revision of TANKAN is caused by the following three factors (Figure 6):
As for the Business Conditions DI, there is an upward discontinuity in all mainly due to adding new sample enterprises which exerted upward pressure on the DI.6 On the other hand, a discontinuity differs in each category, because changing criteria contributes both positively and negatively to the DI.
Let us look at how three factors mentioned above affect business conditions DI concretely.
At first, the eliminating sample enterprises factor contributes positively to the new DI. As a whole, the eliminated enterprises, which employ 50 and over but have less than 20 million yen capital, are the ones with lower DIs than other sample enterprises (See Figure 7). Secondly, the changing criteria factor does not have any general implications on the Business Conditions DI. It contributes both positively and negatively to the new DI (See Figure 8). Finally, the adding sample enterprises factor contributes positively to the new DI, except for large manufacturing, in which a few are added to sample enterprises. As a whole, new sample enterprises, especially those newly established, have higher DIs than older ones. New enterprises have become samples of TANKAN, since the population of the survey was updated. To put it another way, the updated population may include some newly established enterprises with outstanding performance among all enterprises and thus have positive effects on the new DI (See Figures 9 and 10). TANKAN on the new basis accurately reflects business conditions in this way. In other words, the old population may cause bias to the survey with the course of time.7
Now, let us look at categories such as large or small, and manufacturing or nonmanufacturing in detail.
We analyzed the discontinuity in fixed investment among annual projection items in the same way as we did in the Business Conditions DI. Eliminating sample enterprises and changing criteria do not have any significant effects on fixed investment. On the other hand, adding sample enterprises has some positive effects on the year-to-year change of fixed investment of small enterprises in both manufacturing and nonmanufacturing (Figure 11).
To sum up the analysis described above, significant discontinuities are not observed between the pre- and post-revision surveys. Note, however, that there are some discontinuities which may be caused by the bias in the pre-revision survey, because five years have passed since the last revision. The discontinuities may reflect the various structural changes of the Japanese economy which occurred in the last five years.
Recently, the structure of the Japanese economy and industry has been changing rapidly. In order for TANKAN to capture the actual business conditions, a more frequent revision may be required. As such, we will shorten the interval between the revisions of sample enterprises from every 5 years to every 2 or 3 years.
Even after the revision, there are still several issues we have to deal with. For example, although the procedure for responding mergers and spin-offs is partly changed from the revision, we think it incomplete even after the revision. We will continue to examine this issue from a wide perspective including the framework of the survey. Another example would be that we should consider collecting survey forms on-line, as they are still collected by mail. Considering the recent IT innovation, an online survey may bring certain benefits to the respondents.
The Bank of Japan will continue to enhance the accuracy of its statistics by incorporating the changes in economic structures and conditions. Moreover, we will also make efforts to overcome the various problems to meet users' needs and to alleviate the burden on respondents.