SYDNEY: Last September it emerged that Facebook had been overestimating the average viewing time for video ads on its platform by between 60% and 80%, but new data suggest the situation in Australia may have been worse.

Nielsen altered the measurement methodology in October last year and, according to figures obtained by Mumbrella, Australian video streams dropped by 94% over the course of a single month.

The previous metric Facebook gave advertisers for their average video view time counted only the people who watched the video for more than three seconds, the length of time Facebook equates to a "view".

It didn't count anyone who didn't watch, or watched for a shorter time than three seconds, thereby inflating the average viewing time in its analytics. Nielsen now measures streams of all length.

Mumbrella reported that Nielsen had measured total Facebook streams of 8.6bn in July 2016, 12.5bn in August and 9.9bn in September. But in the three months after changing the measurement method, numbers plummeted to 560m in October and November and 580m in December.

And it noted that media buying agencies had cut video spending on Facebook by almost a third in the final quarter, compared to the previous year (based on Standard Media Index data), while comparable spending on YouTube had increased by nearly 12%.

A spokesperson for Nielsen told Mumbrella "it was not appropriate" to compare the numbers. "Our engineering team created a solution that would count progressive streams as one stream per view. The solution was released on October 1, 2016," the spokesperson said.

The devil may be in the details but one media specialist expressed surprise at the apparent scale of the error.

"I knew they were wrong but I didn't realise how wrong," they told Mumbrella. "It throws out all Nielsen's figures before October 2016."

Announcing fourth quarter results recently, Facebook COO Sheryl Sandberg said the platform was working hard with third parties and advertisers to get its metrics right. But she added that viewing times counted for less than sales effect.

"What matters the most is the A/B test that these people saw ads on Facebook and Instagram, these people didn't, and here's the sales lift," she said.

"All of the other metrics … are proxy metrics, and those metrics are going through a platform shift that we need to work on."

Data sourced from Mumbrella, Marketing Week; additional content by Warc staff