Facebook Reports More Measurement Glitches

By Published on .

Facebook CEO Mark Zuckerberg speaks during the Oculus Connect 3 event in San Jose, Calif., in October.
Facebook CEO Mark Zuckerberg speaks during the Oculus Connect 3 event in San Jose, Calif., in October. Credit: David Paul Morris/Bloomberg

Facebook on Wednesday said that it has discovered more measurement problems that could have given marketers the wrong picture of their performance with consumers on the platform.

The social network, which revealed last month that it had been overstating the average time people spent with videos, said in a new blog post that it had given marketers and publishers faulty numbers in a few other spots, inflating how many people visited their Pages and how long they spent reading Instant Articles.

Facebook said none of its findings affected paid advertisement metrics. Still, publishers and brands rely on numbers from Facebook to understand their audience and how to plan content strategies.

Facebook said a "bug" in its Page Insights since last May was responsible for overstated reach measurements, calculating the number of people reached over a week and a month without de-duplicating them. That meant a brand or publisher saw a raw total of visitors to their content over a month, inadvertantly potentially counting repeat visitors more than once.

"The de-duplicated 7-day summary in the overview dashboard will be 33% lower on average and 28-day will be 55% lower," the blog post said.

Other reach data, such as per-post reach and historical figures, were not affected, according to Facebook.

Facebook also said it is changing how it calculates the reach of marketers' pages. "For paid ads reports, we've moved to a stricter definition that only counts reach once the post enters the person's screen ("viewable impressions")," the blog post said. "With the stricter definition, we estimate that reported reach will be 20% lower on average."

An issue with measuring video completions understated those results. "We are now updating how we read the video length to address this issue," the blog post said. "This may result in roughly a 35% increase in the count of 'video watches at 100%.' For example, if 'video watches at 100%' were 1%, they would now be 1.35%."

But the time spent reading Instant Articles, with which publishers let readers consume whole articles on Facebook without clicking through to their own sites, was overstated by 7% to 8% since last August, the company said. "This was caused by a calculation error: we were calculating the average across a histogram of time spent, instead of reflecting the total time spent reading an article divided by its total views," the blog post said. "We have now fixed this issue."

The company's revelation last month had already rattled many marketers, who began to wonder what numbers they could trust when platforms, and not just Facebook, provide their own analysis.

On Wednesday, Facebook said it will improve how it works with third-party measurement companies and set up a "Measurement Council" to work with advertisers on the subject.

Most Popular
In this article: