For a better browsing experience, please upgrade your browser.


Dear Mona,

Is there any state or region that really has worse drivers? Everywhere I’ve lived I have heard people bemoan the driving ability of others, normally from a neighboring state.

Lisa, 31, Asheville, North Carolina


This is a tricky one. I want to try and answer your question using three types of historic data that could indicate where America’s worst drivers are: The number of car crashes in each state (especially those where the driver was negligent in some way), how much insurance companies pay out, and how much insurance companies charge drivers. All three measures vary a lot across the country and no state is consistently at the bottom, but drivers in Texas don’t do very well on any of them. By contrast, drivers in Iowa, Indiana and Vermont are pretty good across the board.

First, collisions. There were 5.6 million motor vehicle crashes in the United States in 2012, of which 4 million involved solely property damage, 1.6 million involved a personal injury, and 30,800 resulted in a fatality, according to the National Highway Traffic Safety Administration (NHTSA). Texas accounted for 3,021 of those fatal crashes, more than any other state, while Washington, D.C., had 14, fewer than anywhere else.

But a state with more fatal car crashes is not necessarily a state with more bad drivers — it could simply be a state that has more drivers or a state with worse driving conditions. So we need to factor in the total miles traveled in each state and focus on the characteristics of the drivers who were involved in fatal collisions.

North Dakota has the highest number of such drivers for every billion miles traveled. Over the course of the 9.1 billion miles traveled in the state in 2011 (the latest data available), 218 drivers were involved in 147 crashes. That produces a figure of 23.8 drivers involved in fatal collisions for every billion miles traveled — far higher than the national average of 15.5.

You’re interested in the behavior of those individuals. We know, for example, whether a driver was distracted at the time of a fatal accident. Of the 45,670 drivers involved in fatal road accidents nationwide in 2012, 3,758 were recorded as being distracted at the time (although it’s worth bearing in mind that in 8,991 cases, it was either not known or not recorded whether the driver was distracted). There’s detail on what those distractions were: 397 of those drivers were distracted by their cellphones, 39 were eating or drinking, and 17 drivers were simply “lost in thought/day dreaming.”

In both Ohio and North Dakota, just 1 percent of drivers involved in fatal accidents were recorded as distracted, compared to 10 percent nationally. But those percentages need to be treated with plenty of caution — it might sound pretty impressive that none of the drivers involved in fatal accidents in D.C. were recorded as being distracted at the time, but that’s based on only 10 drivers for which we have information.

Bear in mind, though, that not all drivers who got in fatal crashes in a given state are licensed there — in New York, for example, only 87 percent did.


The database also shows whether drivers were involved in previous crashes. For 88 percent of drivers nationally, it was their first crash. That figure varies by state, though. At the high end, in Idaho, 98 percent of drivers hadn’t been involved in any previous collisions, while in New Jersey, at the low end, that figure was 78 percent.


Another way to put those fatal crashes in perspective is to determine whether the driver was speeding at the time. The latest data available for speeding-related fatalities is from 2009, when the NHTSA recorded 33,808 total traffic fatalities, 31 percent of which occurred while a driver was speeding. In Mississippi, just 15 percent of traffic fatalities occurred while a driver was speeding, while in Pennsylvania, the share was 50 percent.


Because it’s an irresponsible behavior, speeding is a good indicator of who’s a bad driver — so, too, is drunken driving. Thirty-one percent of all traffic fatalities in 2012 occurred while a driver was alcohol-impaired. In Montana though, 44 percent of traffic fatalities that year involved a driver who was alcohol-impaired, while it Utah, that figure was 16 percent.


Those numbers probably aren’t news to insurance providers, who base their prices on a multitude of indicators, including driver behavior in accidents that weren’t fatal. So, average premiums in each state could reflect insurance companies’ overall assessment of who is likely to cost them in the future.

According to the latest figures from the National Association of Insurance Commissioners (NAIC), high-risk drivers are to be found in New Jersey, where at $1,302, car insurance is the most expensive in the country. Nationally, the average combined premium (collision, comprehensive, etc.) was $912 in 2011. On that same logic, Idahoans, whose car insurance is on average less than half that, are the best drivers in the country.


Bad drivers can affect good drivers’ premiums, too. According to the Insurance Research Council, 12.6 percent of drivers on American roads were uninsured in 2012. That fraction is highest in Oklahoma, where 1 in 4 motorists doesn’t have insurance.

Not insuring a vehicle certainly makes you “bad” in terms of being irresponsible, but I don’t think that’s what your question is driving at. So, rather than looking at prices, we can use NAIC data on the losses that were incurred by insurance providers in each state.

The sums are vast. For all collisions (and not just fatal ones), insurance companies paid out $26.4 billion in 2010. That figure is unequally distributed among states, but so is the number of insured drivers. To make the comparisons fairer, I’ve divided insurance companies’ losses in each state by the number of insured registered drivers there (which I’ve estimated using the number of licensed drivers and the percentage of drivers who are insured).

Yet again, Idahoans come out as America’s best drivers, costing insurers on average $83 each for collisions in 2010. New Jerseyans still don’t come off so good, costing insurers $160 apiece for collisions, but they’re still far behind the most expensive state, Louisiana, where it was $195.


I’m sorry there’s no easy answer here, Lisa. The number of car crashes, even fatal ones, just isn’t a clear-cut way to understand who is and who isn’t a bad driver. But I can say that insurance providers think that you North Carolinians deserve low prices compared to the national average — perhaps because each of your insured drivers only cost them $128 in collision losses in 2010.

Hope the numbers help,


Have a question you would like answered here? Send it to or @DataLab538.

Comments Add Comment

Who doesn’t love puppies? More than 43 million American households own pet dogs, and a whopping 74 percent of Americans “like dogs a lot.” (Only 41 percent of people said the same of cats.) Yet despite this — or perhaps because of it — cities and states across the country are cracking down on pet stores’ abilities to sell dogs. Although major chains such as Petco and PetSmart refuse to sell dogs and instead promote adoptions through shelters, Petland franchises and many local shops across the country continue the practice.

The main issue here isn’t whether pet stores should be able to sell dogs — it’s about how these stores get the dogs they sell. The Humane Society of the United States estimates that there are about 10,000 licensed and unlicensed large-scale commercial breeders in the country, known controversially as “puppy mills,” that churn out more than 1 million puppies per year. Meanwhile, 3 million cats and dogs are euthanized in shelters each year.

On top of this, investigations and exposés have publicized that these mills often prioritize quantity over quality. Those reports say that these breeders create poor conditions for dogs and produce puppies with health issues that often don’t present themselves until the dogs are brought home by their final owner.

The Humane Society collects complaints from owners who have bought puppies with health issues and think the dog was bred at a puppy mill. It published a report on more than 2,400 complaints it received between 2007 and 2011 — 65 percent of these involved puppies bought at a pet store, and the remainder were bought through a breeder or broker.


The most alarming statistic is that 15 percent of these complaints were for puppy deaths, most commonly caused by parvovirus and heart and liver defects. Other common complaints included intestinal parasites, respiratory issues, seizures (due to suspected neurological diseases) and skeletal disorders.

These issues are well-known, and many stores renounce puppy mills and claim that they only get their dogs from small-scale, licensed breeders. But the Humane Society has conducted multiple investigations that conclude the problem is more rampant than it appears.

  • In 2012, an investigation of 12 Chicagoland pet stores confirmed, via Illinois Department of Agriculture health records, that seven of the stores received dogs from puppy mills and one received dogs from unlicensed breeders. All 12 of these stores either told the Humane Society that they only receive dogs from small-scale, licensed breeders or did not respond to the Humane Society’s questioning.
  • A similar 2011 inspection of New York City pet stores, in which the Humane Society visited 11 stores in person and assessed more than 1,300 interstate shipping documents, found that every store visited in-person received dogs from puppy mills despite claims — both to the public and to the Humane Society — that they obtained dogs exclusively from reputable breeders. Shipping documents showed that more than 100 other pet stores were found to procure dogs from puppy mills.
  • Perhaps the most damning was a 2008 probe into Petland’s puppy sources. Investigators visited 21 of its approximately 140 stores nationwide, 35 large-scale breeders linked to the chain, and they reviewed more than 100 federal and state inspection records. They found that despite company claims, Petland bought dogs directly from puppy mills, as well as brokers and Internet auctions that traced back to puppy mills. A 2010 Animal Planet investigation corroborated these findings.

New laws, as well as protests encouraging the closure of puppy mills, show these sales are becoming an increasingly pressing issue. But even as reforms take place, it’s not clear that pet stores are willing to give up their practices. Maryland passed a law in 2012 that merely required pet stores to “post conspicuously on each dog’s cage” certain information regarding its source. The Humane Society visited 12 pet stores throughout Maryland last year and found nine were noncompliant. Three still acquired their dogs from licensed brokers linking back to “puppy mill states,” such as Arkansas, Missouri and Illinois, but information on the specific breeders was unavailable.

Maybe the best solution for any concerned animal lover is the one that 85 percent of Americans have already discovered: Don’t buy a dog from a pet store.

Comments Add Comment

SENATEUPDATEElectoral junkies — ourselves included — have been fascinated by the prospect Republicans may lose races in red Georgia and redder Kansas. We shouldn’t lose sight of the fact, however, that Republicans don’t need to win either of those races to win the Senate.

Republicans need to pick up six seats.

According to FiveThirtyEight’s latest Senate forecast, Republicans have at least a 78 percent chance of picking off seven seats. In declining order of likelihood: Montana, West Virginia, South Dakota, Louisiana, Arkansas, Alaska and Colorado. If the GOP loses Georgia and Kansas, they’re down to five.

That’s why Iowa is key. If Republicans win it, then they can afford to lose Georgia and Kansas and win the majority without pulling off an unexpected victory in New Hampshire or North Carolina. It’s no mistake that Republican Joni Ernst’s chance of winning in Iowa, 66 percent, is nearly the same as the 64 percent chance Republicans have at taking back the Senate.

On Thursday, Quinnipiac University released a survey, and — for the second week in a row — found Ernst ahead of Democrat Bruce Braley by 2 percentage points (48 percent to 46 percent). Along with a Gravis Marketing survey showing Ernst up 6 points, her average lead in the 13 nonpartisan sponsored polls conducted over the past month has been just 2.2 percentage points. It’s even smaller in the our projection (which accounts for pollster accuracy, house effects and state fundamentals) at 1.3 percentage points.

But Ernst’s advantage doesn’t come only from the polling average. It also comes from how consistent the polls have been. Including Thursday’s polling, Ernst has been ahead in 10 of the past 13 independent surveys. She has been tied in two and losing in just one. This is similar to the polling in Colorado and in the 2012 presidential race in Ohio. The leading candidate’s advantage is small, but the polls are all clustered tightly around the average.

The FiveThirtyEight model takes into account this lack of deviation. The model projects Ernst to win with more certainty because the polls are in close agreement.

Are the polls wrong? They could be. But Democrats shouldn’t count on it.

There are no signs in the early voting numbers that Democrats are cooking up something that pollsters aren’t picking up. Both sides are turning out about the same number of their voters, while Democrats are leading among independents. The Democrats’ edge among independents is reflected in the Quinnipiac poll, which found Braley leading among people who have already voted by 21 percentage points.

Still, however you measure it, this is a close election. Ernst is a 66 percent favorite — a far cry from a lock. And Braley has 12 days left to make something happen. Democrats shouldn’t count on a polling error in Iowa, but an error is always possible; Ernst’s lead is small, even if it is consistent.

But if Ernst does win, Iowa gives Republicans more room for letdowns elsewhere, and that could prove crucial in determining who controls the Senate.

Check out FiveThirtyEight’s latest Senate forecast.


Comments Add Comment

The numbers are the numbers. There’s nothing sacred about 51 percent or 72 percent or 95 percent. But certain probabilities, I’ve found, are harder to translate into the right words. For most of 2014, Republicans’ probability of taking over the Senate has been somewhere in the neighborhood of 60 percent, according to the FiveThirtyEight forecast. The gambler in me says that’s not quite close enough to describe as a “tossup”; you’d make a lot of money over the long run betting on a coin toss weighted 60-40 to your side. But it still represents a highly doubtful outcome. A 60 percent chance of an outcome occurring means there’s a 40 percent chance of it failing to occur. As 60-40 underdogs, Democrats’ chances of keeping the Senate would be about as good as Ted Williams’s chances of getting a base hit in 1941.

Over the past week or two, the FiveThirtyEight forecast has drifted slightly more toward Republicans. As of Wednesday night, the GOP’s chances of a Senate takeover were up to 66 percent, its highest figure on the year.

Sixty-six percent might seem a lot different than 60 percent; it tends to read as “2-to-1 favorites” rather than “just slightly better than a coin flip.” But it isn’t much of a change, really; Democrats still have a 34 percent chance of prevailing. The difference between a 40 percent chance and a 34 percent chance is one additional “hit” for every 17 attempts. Essentially, Democrats have fallen from Williams’s chances of getting a hit in 1941 to Tony Gwynn’s in 1989.

With that said, it’s been hard to find good news for Democrats in the Senate polls lately. Colorado has broken against its incumbent, Mark Udall. Sen. Mark Pryor’s odds of holding his seat in Arkansas have become longer. Democratic incumbents are still favored in North Carolina and New Hampshire, but those races have tightened.

Sometimes, Democrats have had to settle for an absence of bad news. Joni Ernst, a Republican, is the slight favorite in Iowa. But her lead is small and steady at 1 or 2 percentage points; it’s not expanding like Republican Cory Gardner’s in Colorado. Greg Orman, an independent in Kansas who could caucus with the Democrats if he wins, no longer holds a consistent lead over the Republican incumbent, Pat Roberts. But Roberts hasn’t pulled ahead either.

The one real exception has been in Georgia. Contrary to some media accounts, it was never out of reach for the Democrat, Michelle Nunn. But Nunn has led Republican David Perdue in the past two surveys by the highly rated pollster SurveyUSA and been very close in just about every other poll. Her chances are as good as at any point in the election cycle.

There are a few complications. The recent polls in Georgia have been a slightly Democratic-leaning bunch, according to our model’s house effects adjustment. But far more important is that the race will require a runoff if neither candidate wins 50 percent of the vote Nov. 4.

The dynamics of a potential runoff are a bit unclear but probably somewhat unfavorable to Nunn:

  • Georgia’s Senate race went to a runoff in 2008, and Democratic candidate Jim Martin performed drastically worse than he did on the November ballot. But 2008 was a presidential year. There might not be such a disproportionate drop in Democratic turnout in an off-year election.
  • The runoff, if the outcome of the Senate was still undecided, could turn into a referendum on party control. Voters might place more emphasis on the party identification of the candidates and less on the qualities of the individual candidates. That probably helps Perdue because Georgia is Republican-leaning (although becoming less so).
  • A third-party candidate, Libertarian Amanda Swafford, would not appear on the runoff ballot. As a group, voters who hold libertarian positions are more Republican-leaning than Democratic-leaning, which could help Perdue. However, Swafford’s positions are split pretty much down the middle between the major parties. Although fiscally conservative, she supports same-sex marriage, marijuana legalization and a reduced U.S. military presence. She’s more of a Gary Johnson Libertarian than a Rand Paul type.

So, there’s a good deal of uncertainty. But the FiveThirtyEight forecast now has Nunn with a 40 percent chance of winning — just slightly worse than a coin flip.

Technically speaking, that projection doesn’t distinguish the Nov. 4 ballot from the runoff. If pollsters were testing a two-way matchup without Stafford on the ballot and trying to evaluate how turnout might differ in a runoff, we might do something more sophisticated, like running separate simulations of the Nov 4. and runoff ballots. Unfortunately, almost none of the pollsters are doing that, so 40 percent has to serve as our stand-in for Nunn’s overall chances of winning the Senate seat.

Nunn does have some chance — about 20 percent, according to our model — of winning an outright majority Nov. 4. (Perdue has about a 35 percent chance of doing so.) Otherwise, Democrats would have to take their chances in a runoff.

Here’s the thing: Taking their chances in a runoff would be a lot better for Democrats than having no chance at all. Nunn’s chances of winning Georgia — 40 percent — are better than Democrats’ chances of keeping the Senate (34 percent). That means Georgia may be more a necessity than a luxury.

From the set of simulations the FiveThirtyEight model ran Wednesday night, I pulled out those cases in which Democrats held onto the Senate with 50 seats (including any independents who might caucus with them). This is the bare minimum Democrats would require, as Vice President Joe Biden’s vote would break the 50-50 tie in their favor.

In the chart below, I’ve listed how often Democrats won the competitive states in those cases where they held control of the Senate 50-50. This isn’t quite the Democrats’ “path of least resistance.” There’s a lot of resistance: Holding the Senate will require Democrats to win some states where they’re underdogs. It’s more like their path of last resort. And more often than not — 57 percent of the time — that path included a Democratic win in Georgia.

silver-datalab-mustwinSome of the other figures in the chart are intriguing. Democrats have a 97 percent chance of winning New Hampshire conditional upon the outcome in the Senate being 50-50, for example. That might sound awfully high; Jeanne Shaheen’s overall chances of winning the state are about 80 percent. However, most of the time Shaheen loses New Hampshire, Democrats will have had a poor night nationally and Shaheen’s loss would be superfluous. (At least for now, if Democrats lose the Senate, their margin of defeat will be quite important in 2016.) In fact, in the simulations where Republican Scott Brown won New Hampshire, the GOP finishes with an average of 54 Senate seats, winning the majority with ease.

Georgia has more chance of being a tipping-point state that would determine majority control. It would be too much to call it a “must-win” state for Democrats; they do have some other options. But they have a better chance of winning Georgia than they do of salvaging their incumbents in Colorado, Louisiana, Alaska and Arkansas — or holding the seats in South Dakota, West Virginia and Montana, where Democratic incumbents are retiring. Losing all seven of those seats, but winning two from among Georgia, Kansas and Iowa, is the Democrats’ best chance.

Without Georgia as an option, Democratic chances of keeping the Senate would be down to about 25 percent. Twenty-five percent chances come in fairly often, too — they come in 25 percent of the time! — but Democrats would be downgraded from Gwynn to Alfredo Griffin.

Comments Add Comment

SENATEUPDATEColorado is a tipping point state in presidential elections — lately, it’s been the tipping point state. Arrange all the states by President Obama’s margin of victory in 2012, starting with his biggest wins, and add up each state’s electoral votes. The state that gets you to 270 is the tipping point state. In 2012 and 2008, that was Colorado.

Colorado’s place at the center of the political universe seems to be secure this year. The Senate race between Democratic Sen. Mark Udall and Republican Cory Gardner has been one of the most competitive for most of campaign season.

That’s why the strength of Gardner in recent polls has been so important in our Senate forecast, which currently gives Republicans a 66 percent of securing a majority. Gardner’s chance of winning has risen to 80 percent. He has led in all but one of 15 nonpartisan sponsored surveys released since mid-September, and by an average of 3.4 percentage points. That includes a newly released Ipsos poll that has Gardner ahead 47 percent to 45 percent, and a new Suffolk University poll that found Gardner up 46 percent to 39 percent. Gardner was only ahead by 1 percentage point in the previous Suffolk survey.

Yet we’re beginning to hear arguments not to trust the polls in Colorado. Skeptics point to polls there in 2010 and 2012 that underestimated Democratic strength. Others argue that the polls aren’t including enough Latinos, and that there’s too many white respondents. Or that the Latino vote is too Republican-leaning. Or that maybe Colorado’s vote-by-mail system will complicate things, making it harder to know who will vote this year.

Pay no attention to those public polls. Instead, Democrats are pushing two recent surveys by the well-respected Democratic pollsters Joel Benenson and Mark Mellman, both showing Udall leading by 3 percentage points.

I don’t put much stock in any of these arguments. Let’s take them one at a time:

It’s true that polling in Colorado has overestimated GOP support. The FiveThirtyEight projections in the 2010 Senate race and 2012 presidential race, mostly based on the polls, were off toward the Democrat by 2.7 and 2.9 percentage points, respectively.

But Gardner’s lead is nearing a point at which he could afford a polling error of that magnitude. He’s up by 2.4 percentage points in our latest projection. Udall would need every bit of those past errors to pull out the victory.

You’d also have to think the mistakes pollsters made in the past will happen again this year. But pollsters adjust their methodologies. Furthermore, two elections in a row doesn’t mean the polls “always” underestimate the Democratic candidate in Colorado. In Udall’s last campaign, in 2008, he was projected to win by 11.1 percentage points by FiveThirtyEight. He won by 10.3 points. In other words, the FiveThirtyEight estimate — again, based mostly on the polls — slightly overestimated Democratic strength.

Second, there’s isn’t much evidence public pollsters are overestimating the share of Latinos will make up of voters. The recent Benenson and Mellman surveys projected the electorate to be 84 percent white and 9 percent Latino, and 79 percent white and 9 percent Latino, respectively.

What have public pollsters found? The past five polls released in Colorado to list demographic breakdowns have used samples, on average, 79.7 percent white and 11.3 percent Latino. In other words, the public polls are less white and more Latino than the two recent Democratic-leaning surveys. These five public polls found Gardner ahead by an average of 4.4 percentage points.

So, we don’t have any compelling reason to think the public polls are underestimating Latino turnout, but maybe they’re reaching disproportionately Republican-leaning Latinos?

Not really. The Benenson poll, which showed Udall up 3 percentage points overall and released a breakdown of the vote by race, had Udall winning by 22 percentage points among Latinos. The past five public polls with a Latino crosstab found Udall leading by 28 percentage points among Latinos, on average. The Suffolk poll released Wednesday had Udall winning Latinos by 73 percentage points.

Picking apart a poll’s subsamples, which have big margins of sampling error, is usually a waste of time. But for what it’s worth, these public polls have had Latino subsamples more Democratic-leaning than the left-leaning polls.

The real reason the public polls differ from the Benenson and Mellman surveys is because the public polls have Gardner doing better among white voters. Gardner leads among whites in those five public polls by 10.6 percentage points, on average. In the Benenson survey, Gardner and Udall were tied among white voters. Whites make up most of the electorate, and so that lead is worth 8.4 percentage points in the overall sample — pretty much the entire difference between the public polls and the Benenson survey.

Finally, it’s possible that all-mail voting will throw off the pollsters, but there isn’t any evidence it will. Oregon has had all-mail voting since 2000, and Josh Katz of The New York Times has found it had the seventh-lowest error rate in Senate polling out of the 36 states he studied. Washington switched to all-mail voting for the 2012 election, and the polling average in the 2012 Senate race was off by 0.5 percentage points.

Pollsters haven’t had problems with vote-by-mail states.

The FiveThirtyEight forecast projects Gardner as an 80 percent favorite. The polls aren’t too far off from our model’s “fundamentals” projection, which accounts for the state’s partisan history, the national political environment, candidate quality and candidate fundraising. Gardner leads in our fundamentals projection by 0.5 percentage points.

The polls and the fundamentals tell the same basic story: An unpopular incumbent is losing in a purple state against a decent candidate in a slightly Republican-leaning year. That’s a pretty believable story.

Check out FiveThirtyEight’s latest Senate forecast.


Comments Add Comment

A study released this week finding that three Ebola-infected air travelers could be leaving the three affected West African countries each month based its estimate on some big assumptions.

The researchers assumed that people infected with Ebola would take commercial flights at the same rate as everyone else in their countries. They assumed the official counts of the people infected were accurate. And they assumed those counts would remain static.

Tweak those assumptions, as the researchers did in the appendix to their Lancet article, and the numbers change significantly.

Take the case of Guinea. The researchers estimated it would take, on average, 2.7 months for one Ebola-infected person to leave the country — if air travel were at the levels it was at the end of last year. Factor in the restrictions to air travel because of Ebola fears, and it would take eight months. In other words, there would be about a 50 percent chance of someone infected with Ebola flying out of Guinea between the start of September and the end of the year.


But according to the appendix, if the real case burden is 10 times the reported level, and air travel isn’t restricted, then about 15 people would fly out during that period. Switch instead to an exponential growth rate in number of Ebola cases, and 45 infected people would leave the country by plane between September and December.


One assumption the researchers didn’t change in the appendix is that all people in each country have an equal chance of flying — whether or not they have Ebola. “I don’t know how realistic that is,” Andreas Handel, an assistant professor in the department of epidemiology and biostatistics at the University of Georgia’s College of Public Health, said in an email. He pointed out that the ability to fly internationally is likely to be correlated with better sanitation and access to health services. “All of these uncertainties make me very hesitant to attach too much meaning to any specific number,” Handel said.

Kamran Khan, lead author of the Lancet study, said he and his colleagues had no data to assume otherwise. “We highlight this as an important limitation in our manuscript,” Khan, an infectious diseases specialist at the University of Toronto, said in an email. He and his co-authors point out in the paper that visiting health-care workers have both a higher chance of infection and of being able to fly internationally than do most residents of Guinea, Liberia and Sierra Leone.

To make their estimates, the Lancet authors combined flight data from the International Air Transport Association, World Health Organization estimates of Ebola cases, and World Bank population data.

Although the estimate of three infected travelers per month is what made headlines, the paper also includes other interesting data on Ebola and air travel, such as the projected share of traveler volume from the three affected countries to various worldwide destinations (2 percent to the U.S.).


And the report isn’t alone in making big assumptions to project the spread of the Ebola epidemic. Researchers from the U.S. Centers for Disease Control and Prevention, for instance, made many assumptions in projecting 1.4 million worldwide Ebola cases by late January. Calculations during the outbreak, with limited and incomplete data, will necessarily be approximations.

Whether the true number is three infected air travelers per month or not, Khan said, a travel ban isn’t a solution. “The most proactive and preventative way to decrease the risk of international spread would be to decrease the number of new infections in West Africa,” he said. “Travel bans wouldn’t address the root cause of the problem and could compromise the international community’s ability to manage the outbreak in West Africa, which ultimately should be the top priority.” Exit screening travelers for Ebola symptoms is more sensible, he and his co-authors wrote in the paper.

The paper’s estimates also don’t translate into a significant chance of catching Ebola while flying, even at the upper end of the range, according to other researchers. “The chance of sitting next to an infected person would be greatest on a flight originating in Liberia, Sierra Leone and Guinea, but even there it would be very unlikely,” Benjamin Neuman, a virologist at the University of Reading in Britain, said in an email.

“I didn’t try to crunch the numbers, but I would venture a guess that currently the overall risk of getting and dying from Influenza while on a flight from NYC to London are probably higher than getting and dying from Ebola during such a flight,” Handel said. “That could of course change if things get worse and we don’t get the situation under control soon.”

Comments Add Comment

Sohan Murthy at the LinkedIn blog published a map Wednesday that shows which skills and jobs are most disproportionately represented in American cities based on data from LinkedIn members.

Here’s LinkedIn’s map:

The map doesn’t show the most common job in each city; that’s another metric and would probably make for a pretty boring map, all things considered, because the plurality of Americans work in sales. Instead, it shows which jobs are observed at a disproportional rate in different metropolitan areas among LinkedIn members (which may not represent the general public).

Murthy points out a few things about the map: You can see America’s major energy hubs along the Gulf of Mexico and around Bismarck, North Dakota. And many cities fit their stereotype: D.C. with its public policy obsession, New York with its finance types, and Nashville, Tennessee, with the music business.

But it’s also fun to compare the prevalence of jobs within cities. “Mining and commodities” is bigger in Las Vegas than “travel” or “restaurant and catering,” which leads me to think that what happens in Vegas doesn’t stay in Vegas — it’s exported to industrial consumers of raw materials. I thought writing and publishing would be more disproportionately popular in New York than, say, general finance, but that’s just the bubble I’m in.

The map also happens to pick out a few other unexpected but sensible findings. Look at the gray dots, and you see a map of major U.S. military installations. It wasn’t shocking to see Seattle’s a hub for tech, but I was surprised that Boise, Idaho, was. As it tuns out, some of the largest employers in the Boise area are circuit designers.

There’s also a ton of data from European cities. Check out the whole treatment by Murthy on LinkedIn.

Comments Add Comment

The Denver Broncos, who moved into the top slot in our Elo ratings last week, widened their advantage over the rest of the league in Week 7. Denver looked dominant in beating the San Francisco 49ers, who had been the No. 3 Elo team. The No. 2 Seattle Seahawks lost — greatly imperiling their playoff chances — as did the No. 5 San Diego Shargers. The No. 4 team, the New England Patriots, barely won against an awful opponent, the New York Jets.


So, it’s time to pose the question Aaron Schatz asked at Football Outsiders. Is this Denver team merely the best in the league — or one of the best teams ever?

The Broncos’ Elo rating is 1683, which translates to being favored by a touchdown in a neutral-site game against an average opponent. It’s also a fairly typical rating for the best team in the league at this point in the season. On average since 1970, the highest-rated team through Week 7 had an Elo rating of 1686, almost exactly matching Denver’s this year.

Let’s pick some nits with the Broncos: They’ve lost once this year. They had a bye week. They fizzled in the second half against the Indianapolis Colts. And they were crushed in last year’s Super Bowl.

Is this unfair? Except perhaps for the Super Bowl part, of course it is. Denver’s loss came on the road at Seattle and may have literally been the result of a coin flip. Beating Indianapolis by any margin looks more impressive given how well the Colts have played since. It’s not the Broncos’ fault that the schedulers gave them an early-season bye (and thereby one less opportunity to improve their Elo rating). More sophisticated statistical treatments — from Football Outsiders’ DVOA (Defense-adjusted Value Over Average) to Benjamin Morris’s assessment of Peyton Manning’s clutchness — tend to have a higher opinion of the Broncos than Elo does.

But if we’re judging whether the Broncos are a historically great team, we have the right — nay, the duty — to be very, very picky.

This is not just an academic question. In the table below, I’ve taken the teams that had the highest Elo rating through Week 7 in each season since 1970. Eleven of the 25 teams that rated higher than the Broncos — or 44 percent — would go on to win the Super Bowl. But just two of the 19 (11 percent) that rated below the Broncos did.


A lot of this is because even small advantages can compound over time. It’s not as though we can say with all that much confidence whether this year’s Broncos are better than, say, the 1998 edition of the team under John Elway. (Elo would have the 1998 Broncos favored, but by only 1.5 points). But a team that’s ever-so-slightly better is more likely to make the playoffs as a division champion rather than a wild card, more likely to have playoff games at home and to get a first-round bye, and more likely to win those playoff games once they occur. A team that’s a 70 percent favorite in each playoff game, and that needs to win three games to win the Super Bowl because it gets a bye, has a 34 percent chance of winning a championship. A team that’s a 60 percent favorite and that has to win four times has just a 13 percent chance.

Elo puts the Broncos’ chances of winning the Super Bowl at 18 percent, somewhere between the two groups. Their Thursday-night home game against San Diego will be highly informative. A win for Denver would put it in a dominant position in the AFC West — it’d be a game-and-a-half ahead of the Chargers and at least two games ahead of the Kansas City Chiefs (with a tiebreaker advantage against each division rival). A loss to the Chargers would instead place Denver half a game behind San Diego with a tiebreaker disadvantage (and with the remaining game between the teams to be played in San Diego).

The Broncos will probably make the playoffs as a wild card even if they lose their division, but the Super Bowl is much harder to win under those circumstances. The Broncos would have to play an extra game, and they’d have to play on the road. In our simulations, the Broncos won the Super Bowl 23 percent of the time as division champs but just 4 percent of the time as a wild card team. There’s not only playoff leverage riding on this game; there’s a lot of Super Bowl leverage at stake.

Here are the current playoff odds for the other NFL teams:


The Colts, who were briefly underdogs to make the playoffs after starting off their year 0-2 (Seahawks fans can take solace in that), are now 95 percent favorites to do so after having won every game since. Some of this is because the Colts are very good, but just as important is their extraordinarily weak division. The Houston Texans project to a record of 7-9, the Tennessee Titans 6-10, and the Jacksonville Jaguars (despite finally winning last week) 4-12. In our simulations, the Colts won the AFC South 82 percent of the time they finished with a 9-7 record and 62 percent of the time they went 8-8. Even a 7-9 record would often be enough to get them in. Andrew Luck is both lucky and good.

Otherwise, the AFC playoff chase is more scrambled than last week. The Cincinnati Bengals are winless in their past three games (they managed a tie against the Carolina Panthers) and down to a 54 percent chance of making the playoffs, in danger of being lapped by Kansas City (41 percent) and the Pittsburgh Steelers (33 percent). Their matchup this week, at home against the division rival Baltimore Ravens, is another high-stakes game.

The NFC, almost inevitably, will feature a ferocious battle for the wild card positions. Dallas, Philadelphia, Green Bay, Detroit, Arizona, San Francisco and Seattle all have somewhere between a 22 and 32 percent chance of winning a wild card. Three of those teams (barring a miraculous run by a team like the Chicago Bears) will win their divisions, but that leaves four teams fighting for two slots.

The Arizona Cardinals, incidentally, have a chance to become the first team to play a Super Bowl in its home stadium. Although Arizona is probably weaker than either San Francisco or Seattle, it has a leg up in the NFC West by virtue of its 5-1 start. Elo gives the Cardinals a 15 percent chance of making the Super Bowl and an 8 percent chance of winning it.

Elo point spreads

Record against point spread: 49-50-3 (8-7 in Week 7)

Straight-up record: 74-31-1 (10-5 in Week 7)


Even if we thought you should bet based on these odds — and we don’t — it wouldn’t be in cases where bookmakers have a team as a 2.5-point favorite and Elo has the same team favored by, say, two points. There are a lot of games like that this week and fewer differences between Vegas and Elo than in past weeks. The biggest split of the week is in an unwatchable game: Elo has the Titans favored by three points at home against Houston, while Vegas has the Texans as one-point favorites.

The Washington Redskins are also a perpetual punching bag for Elo, and the system would have you take the Dallas Cowboys against them (even as 9.5-point favorites, as Vegas has the Cowboys). As Washington was on the verge of losing to the Titans last week — it pulled the game out but didn’t cover the point spread — The Wall Street Journal’s Neil King tweeted that Washington was struggling to beat “the third worst team in the league,” presumably referring to Tennessee. Actually, Elo has Washington rated as the third-worst team! Washington went 11-5 against the point spread in 2012 and perhaps is still getting too much credit for it; it’s 7-16 against the spread since then. The team’s straight-up record, 5-18, is even worse.

Comments Add Comment

SENATEUPDATERepublicans are 66 percent favorites to win a Senate majority, according to the latest FiveThirtyEight Senate forecast. New polls confirmed GOP advantages in Arkansas and Colorado, and a generic ballot survey from AP-GfK found Republicans up 8 percentage points — an unusually large lead.

But — as I’ve written before — there’s a good chance we won’t know who controls the next Senate when the sun rises on Nov. 5. We may have to wait, as we did for the 2000 presidential election. But not because of hanging chads and the Supreme Court. With a possible runoff in Georgia, a probable runoff in Louisiana and the specter of independent Greg Orman in Kansas taking his sweet old time deciding whom to caucus with — not to mention potential recounts — the 2014 midterms may remain unresolved until you’re shopping for Christmas presents (or even making New Year’s resolutions).

So we’re introducing a new metric as we enter the campaign’s final stretch: The probability that Republicans or Democrats will have a majority of Senate seats on or near Nov. 4. We’ll call this a regulation-time win.

Here’s how it works. In each simulation, the FiveThirtyEight model counts up all the seats except:

  • Louisiana, where a runoff is very likely;
  • Georgia, if neither candidate gets above 50 percent (the threshold to avoid a runoff);
  • Any independent win (Kansas or South Dakota);
  • Any race with a margin of victory within 0.5 percentage points (which could necessitate a recount).

If the GOP has at least 51 seats without these cases, that’s a regulation-time win. If Democrats have at least 50 seats without these cases (with Vice President Joe Biden breaking the tie in the Senate), that’s a regulation-time win.

As of Tuesday’s model run, there is a 53 percent chance we’ll be able to project Senate control near Nov. 4; 47 percent of the time, the midterms go to overtime. It’s basically a coin flip.

Republicans have a 38 percent chance of securing the majority near Nov. 4. Democrats have only a 15 percent chance.

It’s possible this calculation underestimates the chances we’ll know the winner within a day or two of Election Day. In 2008, it took two weeks for Sen. Mark Begich to be declared the winner in Alaska. His margin of victory was small but not that close. He won by just over 1 percentage point. It’s not hard to imagine that even a 2 percentage point or 3 percentage point victory could take a while to finalize in Alaska. Republican Dan Sullivan is projected to win Alaska by about 4 percentage points.

If we assume we won’t know Alaska right away, the chance of a regulation-time win for either party falls to 36 percent — 64 percent of the time, we’re left waiting (without a call in Alaska, the GOP wins in regulation 24 percent of the time, and the Democrats win and avoid overtime in 12 percent of simulations).

The GOP’s most likely path to regulation-time victory includes sweeping Alaska, Arkansas, Montana, South Dakota and West Virginia. FiveThirtyEight currently has Republicans with at least a 75 percent chance of winning in each of these states.

The GOP would also need to win three of the following five states: Colorado, Georgia, Iowa, Kansas and Kentucky. FiveThirtyEight has Republicans as 74 percent favorites in Colorado, 65 percent in Iowa and 78 percent in Kentucky. Results in these states are correlated, but not enough to make Republicans favorites to sweep all three. FiveThirtyEight also forecasts that Georgia is more likely than not to go to a runoff, and that Orman has a 54 percent chance of winning in Kansas.

Democrats have a narrow path to regulation-time victory. That’s because Republicans have a good shot at sweeping Alaska, Arkansas, Montana, South Dakota and West Virginia. For a regulation-time win, Democrats would need to win one of those states, or be forced to sweep the remaining battleground races. They would need to win three of the following four contests: Colorado, Georgia (with a majority), Iowa and Kentucky. Democrats are favored to win in zero of these states. (And this scenario gives them New Hampshire and North Carolina, neither of which is a sure thing.)

Fortunately for political reporters who have scheduled vacations in November or December, it’s still more likely than not that someone will win in regulation time — but just barely.

Check out FiveThirtyEight’s latest Senate forecast.


Comments Add Comment

Christophe de Margerie, chief executive of French oil giant Total, was killed Monday night when his private jet hit a snowplow while taking off in Moscow. News reports say the plow driver was drunk. Three crew members also died in the accident.

De Margerie was a colorful figure — his impressive mustache made him unmistakable at industry events — and fond remembrances poured in from both business and political leaders when news of the tragedy became public. He was remembered for being forthright and accessible, not common qualities in an industry where CEOs are often guarded by a phalanx of public-relations staffers.

De Margerie isn’t the first prominent executive to die aboard a private jet, though such incidents are rare. Earlier this year, Lewis Katz, the co-owner of the Philadelphia Inquirer, died when his Gulfstream IV exploded on the runway outside of Boston. Others, such as Micron CEO Steve Appleton, have been killed while piloting their own planes. And of course, there’s a long list of celebrities, from former Alaska Sen. Ted Stevens to singer John Denver, who have died in plane crashes. Most, though not all, have been aboard private aircraft.

That made me wonder whether corporate jets are riskier than commercial airliners. Big companies routinely pay for their top executives to fly private, and many actually require it, even for personal travel. The board of directors at Exxon Mobil, for example, “requires the Chairman and CEO to use Company aircraft for both business and personal travel,” according to the company’s latest proxy statement. (I couldn’t immediately determine whether Total has a similar policy.)

The National Transportation Safety Board collects data on aircraft safety in the United States. At first glance, it looks like it’s much safer to fly commercial: “General aviation” — the umbrella term for noncommercial flights — recorded 70 accidents and 12 fatal accidents per 1 million flight hours in 2012. Commercial airlines had just 1.5 accidents per 1 million hours, and no fatalities.

But general aviation is a broad category that lumps together professionally operated corporate jets like de Margerie’s and tiny two-seaters flown mostly by amateurs (some of these planes, such as Appleton’s and Denver’s, are homemade or have experimental designs). Personal planes are by far the most dangerous, with an accident rate of 120 per 1 million hours. Corporate jets, by contrast, rival commercial aircraft in terms of safety: They had 4.7 accidents and 0.8 fatal accidents per 1 million flight hours in 2012. Moreover, four of the 11 corporate accidents in 2012 involved helicopters, not planes.


In other words, corporate jets aren’t necessarily any safer than commercial flights, but they aren’t much more dangerous, either. As Slate noted in 2008, executives have other reasons for flying private. Companies, including Exxon, usually cite security, but private jets also offer other advantages: flexible schedules, privacy and a better environment for getting work done. Of course, they’re also a nice executive perk — one not all companies decide is worth the money. Microsoft, for one, is famously stingy on business travel: Bill Gates flew coach for much of his time as Microsoft’s chief executive, though he later bought his own plane.

De Margerie, 63, is survived by his wife and three children.

Comments Add Comment

More >

Powered by VIP