Everyone has been reporting on Nielsen’s latest numbers suggesting that ESPN has lost over 600,000 subscribers.

Citing things like the rise of cord cutting, a decline in quality programming, and the NFL’s decreasing rankings, there were plenty of reasons to believe that these numbers were accurate.

However, Nielsen quickly retracted these numbers after ESPN issued a statement suggesting that these numbers were inaccurate:

“The Nielsen numbers represent a dramatic, unexplainable variation over prior months’ reporting, affecting all cable networks. We have raised this issue with Nielsen in light of their demonstrated failures over the years to accurately provide subscriber data. The data does not track our internal analysis nor does it take into account new DMVPD entrants into the market.”

Nielsen was quick to respond, pulling the data:

Nielsen is investigating a larger than usual change in the November 2016 Cable Network Coverage Universe Estimates (versus the prior month). We take the accuracy of our data very seriously and are conducting a thorough analysis to determine whether or not there is an issue with these estimates. In the meantime, we have removed the November 2016 Cable Network Coverage Universe Estimates file from the Answers portal and ask clients not to use the numbers that were posted Friday. We are working closely with clients and will alert them on the findings of our internal review.

While there have certainly been declining numbers, people were too quick to jump on the inaccurate numbers as the narrative has recently been focused heavily on ESPN’s decline. There’s no doubt that these numbers were reflective of some level of truth, as ratings for the sports network have been declining, but such a drastic number should have created some pause among the sports media.

Nielsen is mostly accurate, but the way they measure ratings is over only a small subset of homes. Statistically speaking, they are able to deduct accurate ratings with a large enough sample size (they claim to be in the home of tens of thousands of US television viewers). But with this, there can occasionally be months with outlying data.

This is what appears to have happened in October 2016.