A Study of Missing Value Imputation in Business Surveys
: The Case of the Short-Term Economic Survey of Enterprises in Japan (Tankan)
May 16, 2013
Click on Full Text [PDF 436KB]
In business surveys, how to treat the data on which no responses are obtained (missing values) is an important theme in maintaining and improving the accuracy and reliability of statistics. The Short-Term Economic Survey of Enterprises in Japan (Tankan) enjoys a high response rate thanks to the cooperation of the enterprises it covers. Nevertheless, each survey inevitably includes items with missing values. When there are incidences of missing values in quantitative data items, such as sales and fixed investment, the current practice is to compile the survey result by imputing the missing value with the "previous fiscal year's value obtained from the non-responding enterprise."
This imputation method is thought to be appropriate when the economic environment is stable, as the change in the values between the previous term and the current term is relatively small. On the other hand, during a phase of sharp changes in the economic environment, the result of the compilation may diverge somewhat from the real perception of business conditions, as the previous term's value, which does not appropriately reflect the change during the period, is used without modification. Given the fact that in recent years, there has been a dramatic economic change, such as the one brought about by the Lehman shock, it is worth examining if there are imputation methods which produce a higher degree of accuracy.
Against this backdrop, this paper has compared the degrees of statistical accuracy of the current method of missing value imputation of the Tankan and its alternative methods, conducting simulations based on the data for the period from 2004 to 2010. As a result, it was found that for all of the major survey items (i.e., fixed investment, sales and current profits), "there are alternative methods whose degree of accuracy is higher than or about the same as the imputation method now in use."
This paper was written based on the report made by the authors at the Fifth Joint EC-OECD Workshop on Recent Developments in Business and Consumer Surveys jointly sponsored by the European Commission (EC) and the Organization for Economic Cooperation and Development (OECD) held in Brussels on November 17-18, 2011.
The authors would like to acknowledge and express gratitude for the valuable comments received from Messrs. Kosuke Aoki (the University of Tokyo), Kiyohito Utsunomiya (Kansai University) and Hiroshi Saigo (Waseda University) as well as Messrs. Eiji Maeda, Chihiro Sakuraba, Toshitaka Sekine, Koichiro Kamada, Shuji Kobayakawa, Seisaku Kameda and others of the Bank of Japan in preparing this paper. They would also like to thank the staff members of the Economic Statistics Division, Research and Statistics Department, with whom they have discussed in depth the missing value imputation methods and simulations examined in this paper.
The opinions expressed in this paper are those of the authors and not the official views of the Bank of Japan or its Research and Statistics Department. Whatever errors there may be are those of the authors.
- *1 Research and Statistics Department
E-mail : email@example.com
- *2 Research and Statistics Department
E-mail : firstname.lastname@example.org
Papers in the Bank of Japan Working Paper Series are circulated in order to stimulate discussion and comments. Views expressed are those of authors and do not necessarily reflect those of the Bank.
If you have any comment or question on the working paper series, please contact each author.
When making a copy or reproduction of the content for commercial purposes, please contact the Public Relations Department (email@example.com) at the Bank in advance to request permission. When making a copy or reproduction, the source, Bank of Japan Working Paper Series, should explicitly be credited.