Tag Archives: version

Introducing the Needle, the New Hampshire Version

Introducing the Needle, the New Hampshire Version

Before 8 p.m. Eastern on Tuesday, we expect returns from the New Hampshire Democratic primary. If you choose to follow the results live tonight, we hope you’ll do it with us: We’ll have results, analysis from our reporters and interactive maps.

We’ll also have what we call a live forecast, known to many readers simply as “The Needle.” Below, we’ve provided answers to some common questions about the needle and our plans to use it during the primary.

It’s the visual representation of The Times’s live election night forecast. It combines initial election returns with other sources of data to estimate the final result of an election before all the votes have been counted.

Before the election, we use polling data, census data and prior election results to come up with a baseline estimate for how every area will vote. As the results come in, we compare the actual returns with our baseline. We then adjust our expectations for the vote that’s left: If a candidate is running ahead of our pre-election expectations in the areas that are counted, we might assume that he or she is also poised to do better in the demographically similar areas that remain.

The needle gives many readers the piece of information they want more than anything else on election night: It tells them who is on track to win the election. And when the needle doesn’t know, it tells them that, too.

The needle works because incomplete election results are often deeply unrepresentative. In some races, predominantly rural places tend to be the first ones to report votes. The raw vote count can be misleading evidence about who is actually in a better position to win, even after most votes have been reported.

Sophisticated election analysts and decision desks know this. Most casual followers of an election do not. The needle is a shortcut to most of the things we know about political geography.

In 2016, the needle helped us realize fairly early on election night that Donald J. Trump was on track to win. We showed him with an edge in Pennsylvania, even when Hillary Clinton had a double-digit lead in tallied votes.

Using similar methods in 2014, we were able to tell readers that Mark Warner, a Virginia Democrat, was probably on his way back to the Senate despite trailing all the way through 99 percent of precinct reports. In Alabama’s special Senate election in 2017, the needle showed before many analysts did that the Democrat, Doug Jones, had a path to victory.

For those who wonder whether the world really needs the election needle, we realize the actual results will emerge soon enough. But we also think that the millions of people who follow election night results online ought to have the context to understand them as well as experts do.

The New Hampshire primary isn’t the easiest contest to forecast, but it’s not the hardest either.

The state reports its election results by township, offering a more granular look than the typical returns by county. (The Iowa results, which were reported by precinct, gave us a good sense of where we might expect the candidates to fare better or worse.)

Pete Buttigieg will probably excel in the affluent suburbs of southeastern New Hampshire, as he did in the suburbs of Des Moines. He might also hope to fare well in older parts of rural and small-town New Hampshire, but Bernie Sanders will find great strength along the border of Vermont, his home state.

Mr. Sanders will also probably do well in many of the state’s larger communities, like Concord and Portsmouth, and in most of the state’s college towns. But he might not be as strong in the elite bastion of Hanover, home to Dartmouth, where Elizabeth Warren, Mr. Buttigieg or even Amy Klobuchar might have a shot.

The Iowa Democratic caucus was not the smoothest-run electoral contest we’ve ever seen, and that was reflected in the needle’s performance.

In the broadest sense, the needle fared all right: It consistently showed Mr. Buttigieg favored to win the most state delegate equivalency results, while showing Mr. Sanders favored to win the most votes. This is the current result, irregularities in the existing data notwithstanding.

But we believe the needle was too confident before full returns were reported. Most important, it did not handle the depth of one candidate’s strength in the satellite caucuses. These caucuses, first used on a large scale this year, were held for Iowans in places like hospitals, out-of-state military installations and overseas, and they broke overwhelmingly for Mr. Sanders.

The model should have controlled for varying vote methods, as it usually does but did not here. This is very important in anticipating the chance of big swings in election results. (The early vote, for instance, is often unlike the Election Day vote.)

Making matters worse, the Iowa Democratic Party calculated state delegate equivalents for satellite caucuses differently than we expected, and differently than we believe is described in the state’s formal delegate selection plan. This decision made the race closer than we expected, and it is possible that the state party will ultimately agree with our interpretation of the rules. Even if our interpretation was wrong, the model as it was still did not rule out a Sanders victory.

Once all of the votes have been counted, the needle will — by definition — match up perfectly with the final result.

But before all votes have been counted, the needle may suggest that one candidate has a pretty good chance of winning, only to have someone else eventually win. A 65 percent chance of winning means losing quite often. Likely to happen does not mean certainly will happen. We don’t quite think of this as “wrong.” (We know that some people do. This year, the needle will also describe probabilities with a simple phrase; it would describe the odds of someone with a 65 percent chance of winning as “likelier than not.”)

The needle can also be wrong in regular elections by reacting too quickly to unrepresentative returns, or by failing to anticipate them.

If, by random chance, a candidate’s very best or worst precincts are the first ones to report, the needle may be misleading.

That problem briefly confused the needle in the 2017 Alabama Senate race between Mr. Jones and Roy Moore. For a few minutes, the needle projected Republican turnout to be abysmal — but this was because the first counties to complete their count included some of the weakest Republican turnout of the night. The magnitude of our error at that point was huge — nearly seven percentage points. But our model quickly realized that Republican turnout was going to be higher in other counties.

On the other hand, if the results are reported in a perfectly representative way, the needle may not offer much insight beyond raw vote totals.

No. Calling is for humans.

The needle reflects an estimate of the size and the makeup of the outstanding vote. And it may fail to anticipate returns that don’t fit the patterns we’ve seen until a certain point.

For this reason and others, the needle is not used to make definitive statements about when a race should be considered over.


Source : Link