In compiling the monthly data, which covers some 33,000 firms nationwide with five or more full-time employees each, the labour ministry is supposed to collect samples from all the firms that employ 500 or more workers.
But it turned out that the data sampling failed to cover two thirds of some 1,400 businesses in Tokyo for an unspecified period of time, they said.
Domestic media reported the sampling error extends back for 15 years, but officials stopped short of confirming the period, saying that the matter was still under investigation.
"We will decide what to do with the past data as soon as the investigation is completed," Jiro Yashiki, a councillor in charge of compiling the data at the ministry, told Reuters.
The prime minister and many investors closely follow wage data for clues on progress of his "Abenomics" policy, which aims to create a sustained cycle of higher wages and stronger consumer spending, thus helping the central bank to achieve its 2 percent inflation target.
Despite nearly 6 years of massive monetary stimulus, Japan's wages have been slow to grow, keeping core consumer inflation at around the half the Bank of Japan's target.
The latest case was reported to labour minister Takumi Nemoto on Dec. 20, a day before the ministry issued revised wage data for October, Yashiki said. The ministry withheld the fact that there was a problem with the data sampling.
Data on Wednesday showed inflation-adjusted real wages rose 1.1 percent in the year to November and nominal pay grew 2.0 percent, with regular pay posting the biggest gain in more than 21 years.
The data for October and November was compiled by conducting data sampling restoration, Yashiki of the labour ministry said.
"Assuming that average wages were calculated with a fewer number of big firms, workers' pay may have been underestimated over the past period," said Koya Miyamae, senior economist at SMBC Nikko Securities.
The latest error added to last year's sample changes that helped push up average pay calculations, making it more difficult to read the actual strength of wages, Miyamae said.
"Analysing erroneous data means policymakers are debating issues such as sales tax hike (scheduled for October) without grasping problems surrounding consumption and income, making it difficult to judge whether appropriate policy is adopted."
Analysts say the ministry probably had no intention of manipulating the data to artificially make wages appear high or low, but that the latest case undermined credibility of Japanese indicators.
"Economists were already complaining that the sample size was too small, and now you have this problem. The labour ministry loses some trust because of this," said Norio Miyagawa, senior economist at Mizuho Securities.
"You have to rely on several data points, and not data from just one source, to measure wages and the success of the government's policies," he said, adding that he still thought overall wages remain sluggish.
(Reporting by Tetsushi Kajimoto; additional reporting by Stanley White and Ritsuko Shimizu; Editing by Kim Coghill)
By Tetsushi Kajimoto
