Accuracy Improvement Measures for Energy Consumption Statistics of Japan

         
Author Name KAINOU Kazunari  (Fellow, RIETI)
Creation Date/NO. April 2013 13-J-022
Download / Links

Abstract

Energy Consumption Statistics of Japan that started in 2005 by Ministry of Economy, Trade and Industry, Agency of Natural Resources and Energy is a large scale general official statistics in Japan in order to identify energy consumption by industrial classification and region. But the author evaluated the statistic's accuracy using supply side statistics from 2006 to 2010 that concluded that the statistics includes 20 per cent of error in electricity, town gas and heavy fuel oil A, and that the statistics does not have enough accuracy and not adequate to use as a base statistics of General Energy Statistics.

In order to identify the root cause of the problem, the author made error cause analysis observing the energy consumption data distribution directly using questionnaire's answer sheets. The analysis showed that three digit or four digit errors and imperfect answers are frequently and unstably seen in the answer due to the statistics mainly covers small scale companies in the tertiary industry and manufacturing industry. But the present statistics applies only conventional erroneous data exclusion system such that box plot measures, so error data frequently and unstably passed the error data exclusion system and incomplete data are mostly discarded by mistake.

Based on the root cause analysis, the author improved and re-designed the erroneous data exclusion system for the statistics and proved that less than 5 per cent errors in average from 2005 to 2009 and maximum 7 per cent errors level of estimation accuracy is achievable for major energy sources such as electricity, town gas and heavy fuel oil A.

But, the new erroneous data exclusion system has side effects that only a small number of energy types are able to estimate and energy transformation efficiency such as auto power generation or steam heat generation are not directly able to estimate, so further improvement of statistics answer recovery ratio and the erroneous data exclusion system are deemed to be necessary from now on.