Where do undecideds go?

My distinguished sparring partner in these pages (and real-life friend) Dr. David Hill is, alas, behind the times, or perhaps letting his hopes outrun the data. Last week, he argued that the president is doomed because he is under 50 percent and “if you haven’t yet decided to vote for the incumbent, you won’t.” Hill repeats a maxim I too learned decades ago — in races involving incumbents, 70 to 100 percent of undecideds break toward the challenger.

It proved a useful guide for many years; however, it no longer holds true, in part because the dynamics underlying it have changed considerably.

This observation, sometimes dubbed the “incumbent rule,” played a central role in my own career. One of my first races pitted an unknown against a House incumbent. When Congressional Quarterly listed 125 competitive races that year, ours wasn’t among them. Understandably. Six weeks out, we were behind by over 20 points — 48 percent to 25. I urged the candidate to ignore the margin, noting that the central fact was that the incumbent was under 50 and that most of those undecideds would come our way. Six weeks later, after just 10 days of carefully crafted TV ads, we won by less than 1 point and I had a business.

But what was once a law of politics no longer rules.

In the 2008 and 2010 Senate cycles, 11 incumbents went into Election Day with less than 50 percent of the vote in the poll averages computed by RealClearPolitics — five of them won; six lost. Hardly evidence of an iron law. Five of the six losers were not just garnering less than 50 percent, but were running behind their opponents. (You don’t need a special rule to know incumbents are in trouble when they’re behind.)

Moreover, all but two of those incumbents gained votes beyond the final RealClearPolitics poll average, adding a mean of 3.4 percentage points. Challengers added an average of 1.2 points. Thus, undecideds not only failed to break to challengers, they were more likely to move to incumbents.

Consider some key races. The RCP poll average showed Sen. Barbara BoxerBarbara Levy BoxerDems face hard choice for State of the Union response Billionaire Steyer to push for Dem House push Billionaire Steyer announces million for Dem House push MORE (D-Calif.) with a 5-point margin over Carly Fiorina. On Election Day, Boxer piled up nearly a 10-point margin. That would mean undecideds broke toward Boxer, with Fiorina actually losing support. Our own data had Boxer leading by 10, suggesting undecideds broke evenly, as they did in Sen. Mitch McConnellAddison (Mitch) Mitchell McConnellDems confront Kelly after he calls some immigrants 'lazy' McConnell: 'Whoever gets to 60 wins' on immigration Overnight Defense: Latest on spending fight - House passes stopgap with defense money while Senate nears two-year budget deal | Pentagon planning military parade for Trump | Afghan war will cost B in 2018 MORE’s (R-Ky.) reelection battle.

In Sen. Patty MurrayPatricia (Patty) Lynn MurrayLawmakers eye retirement help for gig economy workers Overnight Regulation: Labor Department reportedly hid unfavorable report on tip-pooling rule | NY plans to sue EPA over water rule | Senators urge FTC to probe company selling fake Twitter followers Trump's vows to take on drug prices, opioids draw skepticism MORE’s (D-Wash.) 2010 contest, nearly all the undecideds broke to her, even though she was the incumbent. In 2008, Oregon challenger Jeff MerkleyJeffrey (Jeff) Alan MerkleyPentagon: War in Afghanistan will cost billion in 2018 Trump has declared war on our climate — we won’t let him win Lawmakers left with more questions than answers on Trump infrastructure plan MORE (D) added 1.1 points to his vote total from poll averages, while incumbent Gordon Smith (R) picked up almost three times as many.

The “incumbent rule” emerged from an underlying structure that is increasingly absent.

In the ’80s and ’90s, most incumbents were much better-known than their challengers. So when a voter said he was “undecided,” what he really meant is that he knew the incumbent and did not particularly want to vote for him or her. However, these “undecideds” did not yet know enough about the challenger to support that candidate. By Election Day, these voters picked up at least a scintilla of information about the challenger, enough to justify casting a ballot against the disliked incumbent.

As a result, in Senate races between 1994 and 2004, on average 70 percent of undecideds broke to the challenger, 30 percent to the incumbent.

Today, things are different. As the amount of money spent in competitive races increased exponentially, fewer challengers are unknowns prior to Election Day. So now, when voters say they are undecided, often they mean they know something about both candidates and like (or frequently dislike) them about equally. That can lead to undecideds breaking even, or to late shifts in favor of incumbents.

Being below 50 percent is a sign of vulnerability for an incumbent, but it is no longer an indication of impending political demise.

Mellman is president of The Mellman Group and has worked for Democratic candidates and causes since 1982. Current clients include the Majority Leader of the Senate and the Democratic Whip in the House.