Iceberg ... Meet Titanic

In the rare event we find people who still keep a home phone, we must then ask, "And who answers those phones?" Phones are about as useful as a stick-shift in an F-16. Thanks to scams and misrepresentations: "This is the IRS," "About your credit card," "You've just qualified for a free vacation" ad nausem, patience is exhausted. In fact, in NuVoodoo's recent tracking that spanned from fall 2015 to late summer of this year, landline population has been dropping; by this past September almost 55% of 25-44 adults had no landline, while 25-54 residences were only slightly higher.

Among the balance of those who do have landlines, 35% screen calls while only 20% or fewer will actually answer. So for months, pollsters -- even more reputable ones -- continued to cast models showing "Clinton up by seven" or "Trump up by two." Across the months and weeks, ardent partisans tracked their candidates with euphoria or morosity. We now know almost none of them were correct. Only two major exceptions had it right, got it right! The outliers were IBD (Investors Business Daily) and an unlikely bedfellow, the L.A. Times/USC Dornsife poll. From early-on IBD and the Times' poll showed something quite contradictory; turns out all the others were wrong! So let us be among the first (several of our clients have heard us stake this claim) polling as we've known it is as dead as Edgar Alan Poe.

The billion-dollar question suddenly becomes "Why did the two outliers get it so right when everyone else was so inaccurate?" ADG does not specialize in data modeling but through Arbitron-then-Nielsen experience; it's clear declining response rates have changed the game. The L.A. Times/USC polling took a different road: they tracked the same three thousand people through their "Daybreak" polling whereby remaining in the same data pool all through the election season, right to the end. So instead of the traditional spray-and-pray polling using automated devices or people who sound like them, the Times and USC stayed within their 3,000 person model ... and began to find a very different picture forming. By 2am Eastern the following morning, regardless of party persuasion, much of the country was in disbelief asking, "How could the polls have been this wrong?"

The media got it wrong and the pundits missed the mark. Countless political hacks ducked for cover. Why? Because the data was simply flawed, which brings us to the steep hill Nielsen must climb to overcome the same malaise as the polling industry; not because Nielsen isn't populated by extremely bright dedicated people, but because what once was, isn't! With landline penetration waning, even more ominous is the stark reality that the percentage who'll actually answer their phone has dropped yet another 4% since a year ago.

Now try to imagine doing a 20-minute perceptual questionnaire or a standard music test that depends on listeners with landlines. Do we really believe most Americans will respond with obligation to "just a few more questions" 10 minutes into a questionnaire? At the same time, the demand for more and more research productivity is seeing an increasingly available pool of on-line respondents. Out from under pressure of a timed call, online respondents can move at their pace and become more engaged in the process.

We're at the proverbial fork-in-the-road. As programmers we need to break the rules. Ownerships and agencies must demand better data modeling and retrieval. Select marketing entities like DMR are going where few have been and seeing significant results.

The election really wasn't a stunning upset -- at least in an empirical context. USC Dornsife/ L.A. Times had it right from the jump. They dared to break the mold.