This website uses cookies

Read our Privacy policy and Terms of use for more information.

We Said Nielsen's Math Hadn't Caught Up. It Was Ours.

Yesterday ADOTAT told readers that Nielsen's holiday numbers did not add up. They add up. We read a plus as a minus, built a section on it, and put it in your inbox. That was our mistake, and this is the correction and the apology.

Unmeasured exists to check other people's numbers. The same rule applies when the number is ours, so here is exactly what we got wrong, how it happened, and what Nielsen told us.

What we got wrong

On November 17 we published "Nielsen says holiday ads work. Its math hasn't caught up." The piece reported that Nielsen's holiday sales lift for cosmetics on connected TV was minus 22.9 percent. We set that beside Nielsen's 3.2x headline, ran a section saying the figures did not add up, called the cosmetics row the most important line in the release, and closed by telling members the math was not settled.

The figure is plus 22.9 percent. There is no negative number anywhere in Nielsen's data. The contradiction we described does not exist.

How it happened

Nielsen's written answers arrived that morning with a stray dash in front of two holiday figures. We read the dash on the cosmetics line as a minus sign. We read the identical dash on the department store line, one row down, as punctuation. Same dash, two readings, and we never stopped to ask why.

It gets worse, and you should know the worse part. We asked Nielsen about the apparent decline that morning. Nielsen answered at 1:06 PM Eastern and said plainly that the lift was plus 22.9 percent, not minus. We published at 3:52 PM Eastern, two hours and forty-six minutes later, without having taken that answer on board. The correction was sitting in our inbox when we hit send.

Nielsen wrote again that evening, asked us to fix it, and apologized for the extra dash. That apology runs the wrong way. A stray dash is a typo. Publishing after you have been told it is a typo belongs to us.

Frank Kowalczyk: "Its math hasn't caught up." Read that headline again. Nielsen caught it in under three hours. The outfit that wrote the headline needed the afternoon and a second email. Somebody was behind, and it wasn't the fellows in New York.

What Nielsen's numbers actually say

These are the sales lift figures a Nielsen spokesperson gave us, corrected. They apply to connected TV, and the baseline is non-holiday campaigns.

Cosmetics: 7.2 percent non-holiday, 22.9 percent holiday

Department stores: 11.2 percent non-holiday, 27.3 percent holiday

Restaurants: 6.9 percent non-holiday, 10.6 percent holiday

Fast food: 4.4 percent non-holiday, 4.8 percent holiday

Now do the division we should have done before publishing. 22.9 divided by 7.2 is 3.2. 27.3 divided by 11.2 is 2.4. 10.6 divided by 6.9 is 1.5. Those are Nielsen's headline multiples, to the decimal. The figures we said contradicted the 3.2x are the figures that produce it.

The second thing Nielsen corrected

Our piece also said Nielsen had described the lifts as both modeled and measured, and that it had not explained the difference. It had, in the same 1:06 PM note, and we missed that too.

Nielsen's explanation: the lift is a modeled value calculated on top of observed sales data, and "measured lift" and "observed lift" are other names for the same thing. Predictive Sales Lift is a separate model that sits on top of those results and relates campaign properties to measured lifts, so it can estimate return without measuring against real sales. Nielsen says Predictive Sales Lift was not used for any of these figures. The two answers we presented as a conflict were one answer in two vocabularies.

What we changed, and what we owe

The article has been corrected on the web. The negative number, the section built on it and the question it generated are gone, and the headline no longer says the math has not caught up. The modeled-and-measured section now carries Nielsen's explanation, and the open question built on that supposed conflict has been removed. Dated update notes at the bottom of the piece say what changed and why. The email already in inboxes cannot be changed, which is why this one exists.

The questions that remain in the piece are ordinary ones about sample sizes and method. They are questions, and none of them rests on the error.

To Nielsen and its communications team: we are sorry. You answered every question we sent, on our deadline, in writing, and you flagged the error before we published. We printed it anyway. To readers: we are sorry too. Some of you forwarded that piece. Please forward this one to the same people.

We get things wrong, because we are human. What we can promise is that when we do, you will hear it from us, at the same volume as the original, with the receipts attached.

We spent a newsletter asking Nielsen to show its work. This is ours.

How we reported this. This correction draws on ADOTAT's November 17 article as emailed, the corrected web version, and our written correspondence with a Nielsen spokesperson on November 16 and 17, whose answers were given on background and attributable to a Nielsen spokesperson. We have not seen the underlying sales lift studies, so the figures are Nielsen's, and the division above is ours. Nielsen did not review this note before it was sent. Frank Kowalczyk is a fictional commentator. Unmeasured is reader-supported and has no paid placement in editorial.