Tag Archives: statistics

J.D. Power’s 2013 Smartphone Satisfaction Survey

Apple easily won the survey, for the ninth time in a row. Last year, them and HTC were the only ones above the industry average; this year, Apple was alone in that regard. Nokia improved a lot, thanks to their Windows phone, and it, Samsung, Motorola and HTC were virtually tied for second place. Way down at the bottom, LG and Blackberry.

2013 U.S. Wireless Smartphone Satisfaction Study by J.D. Power


The survey asks people who have had their smartphone for less than a year to rate it based on performance, physical design, features, and ease of operation.

See also:

From J.D. Power, via iMore

Divorce And Drinking

Scientists had a lot of data on some 20,000 people in a Norwegian county, so they did a study looking at how alcohol consumption affected their marriages. The percentage of subjects that got divorced:

  • 5.8% of couples in which both people were light drinkers
  • 13.1% of couples in which the husband was a heavy drinker, but the wife wasn’t
  • 17.2% of couples in which both people were heavy drinkers
  • 26.8% of couples in the wife was a heavy drinker, but the husband wasn’t

Moral of the story: don’t drink that much, but however much you drink, make sure your spouse is on the same page. Ditto goes for smoking. Also, be generous, committed, and good in bed. And it wouldn’t hurt to live in Utah.

Photo by Tetra Pak

Photo by Tetra Pak


See also:

From Alcoholism, via LA Times and Neatorama

Nate Silver Predicted Most Of The Oscar Winners, Too

Nate Silver is now a famous statistician, after correctly calling the presidential election in every state for 2012, and all but one state for 2008. For his next trick, he’s applied his calculus on the movies: on the Friday morning before the Oscars, he published predictions for the main categories and got four of the six right:

  • Best picture: correctly called Argo, with Zero Dark Thirty being a distant second
  • Best director: incorrectly called Spielberg (Lincoln) as a narrow favorite over the actual winner, Ang Lee (Life of Pi). In the write-up, he explains that this was a shaky prediction to make, because his top two choices — Ben Affleck for Argo and Kathryn Bigelow for Zero Dark Thirty — were actually not even nominated.
  • Best actor: correctly called Daniel Day-Lewis (Lincoln), with Bradley Cooper being a distant second (Silver Linings Playbook)
  • Best actress: correctly called Jennifer Lawrence (Silver Linings Playbook), with Jessica Chastain (Zero Dark Thirty) not being far behind
  • Best supporting actor: this was a pretty big mistake, as he predicted the winner would be Tommy Lee Jones (Lincoln), with Philip Seymour Hoffman (The Master) as a second possibility not that far behind. But he forecast that the actual winner, Cristoph Waltz (Django Unchained), was third most likely to win — though very closely behind Hoffman.
  • Best supporting actress: correctly called Anne Hathaway (Les Misérables), with Sally Field (Lincoln) being a distant second.

Not bad at all. He made the predictions by looking at historical data for 16 other awards shows, like the Golden Globes, SAG Awards and BAFTAs. Now if only he could bring that kind of forecasting accuracy to economics.

Nate Silver


See also:

From The New York Times, via Slashdot

Nate Silver’s Election Forecast Was Deadly Accurate Again

Nate Silver is a statistician that gained some fame after developing a system for predicting the performance of baseball players. That field is now known as sabermetrics, and is the basis for Moneyball — both the book and the movie. In 2008, Silver turned his attention to politics and started a blog called Five Thirty Eight (which is the total number of electoral votes), where he developed a system that, using all polling data available, correctly called not only the presidential election, but how 49 out of the 50 states voted, and how all of the 35 senate races turned out. (Indiana was the one state he missed, which went for Obama by 1%.)

For the 2012 election, the algorithms were even better: he got the all 50 states completely right.

Nate Silver's election forecast on the morning of the election


CNN's election map the day after the election


On the morning of the election, his blog — which was bought by the New York Times in 2010 — predicted, with a confidence level of 90.9%, that Obama would win the election. To anyone following this and other polling data, like Princeton’s Election Consortium‘s, the results on election night amounted to a complete lack of surprise. But of course, all pundits on the news coverage of the election ignored these predictions and instead spent the entire night making their own, much less rigorous forecasts, based on personal knowledge and gut instincts instead of mountains of polling data and algorithms. Then again, how else would a dozen news networks fill as many hours of programming, except by pretending that the future is unknowable?

The always insightful xkcd hit the nail on the head:

The comic’s hover text reads:

As of this writing, the only thing that’s ‘razor-thin’ or ‘too close to call’ is the gap between the consensus poll forecast and the result.

(Updated November 9th to reflect that Florida was called for Obama.)

From Five Thirty Eight and xkcd

We’re Drinking And Eating More, But At Least Smoking Less

Scientific American has a pretty interesting interactive graphic (there’s a non-interactive version below) that shows the trend, over the past 15 years, in five categories: heavy drinking, binge drinking, smoking, obesity and exercising. The top three causes of death are heart disease, cancer, respiratory diseases and too much eating, drinking, smoking and laziness are all causes of all of them, so this is important stuff.

However, it’s not easy to tell how much the figures changed using that graph. For example, binge drinkers went up from 14.1% of people to 15.1% and exercisers went up from 72.1% to 76%, which seems like a bigger deal, but in reality they both grew by about the same percentage. So instead of falling into the same trap, here, the stats are presented like stocks and ordered by the magnitude of the change, followed by the issues with the numbers:

  • Obesity: +74%
  • Heavy drinking: +69%
  • Tobacco use: -24%
  • Binge drinking: +7%
  • Exercise: +5%

So, obesity and drinking are way up, tobacco use is down a pretty healthy amount, and binge drinking and exercise are slightly up. Since exercise hasn’t changed that much, the giant increase in obesity can only be blamed on our diet, which makes sense given all the cheap food. That means we’re eating and drinking a lot more than we used to 15 years ago. And, thanks to the prolonged public education campaign, we’re smoking a good bit less.


And now, for the problems with the numbers

For binge drinking and exercising, the questions used are pretty ridiculous. We’ve talked about the binge drinking definition issue before: having five beers in five hours, five beers in one hour, and fifteen beers in five hours are all counted as binges. And the question asked about exercise is if you’ve done physical activity in the last month; if you helped someone move last week, that would count as exercise. Smoking is defined as “current smokers”, and there’s no category for the many people that are casual smokers and would not identify themselves as “current smokers”.

Heavy drinking is defined as having more than two drinks per man per day, but other researchers define it as more than three per day, including the study that showed conclusively that heavy drinkers live longer than teetotalers. Which brings us to the other problem: if drinking heavily is a habit important enough to our health to be tracked, then it seems like abstaining from alcohol should also be. The fact that it’s not, indicates that some morality factor is also present in the surveys.

Obesity is defined as having a BMI of 30 or more. The BMI is a 200-year old measure with such severe problems that it actually says about 40% of obese people are not obese. Why? Because it only uses height and weight, which is great for easy research, but awful for figuring out how fat someone is: if you have a lot of muscle and no fat, it will say you’re overweight. If you’re nothing but fat and bones, it’ll say you’re normal weight. Ideally, obesity would be defined by body fat percentage, not BMI, but that would be impossible to figure out over the phone. Bottom line: the survey says 27% of Americans are obese, but that number is probably more like 45%.

See also:

Interesting Divorce Statistics

Mental Floss dug up some data on what circumstances are correlated with either higher- or lower-than-normal divorce rates. Of course, correlation is not causation, so just because something is related to a higher divorce rate, doesn’t mean that it causes divorce. Still, they’re interesting to consider:

  • Smoking: couples where one person smokes are almost twice as likely to get divorced. It’s a little higher if the smoker is the wife. And even when both spouses are smokers, the divorce rate is still higher than non-smokers; however, this may have a common cause, since poorer people smoke more and are also more likely to get divorced.
  • Jobs: some professions have lower divorce rates than normal. Optometrists, shuttle car drivers, transit cops, farmers, nuclear engineers and clergy. Again, this is probably just due to the kind of people that choose those professions, and the spouses that marry them — the jobs themselves don’t really make them less divorce-prone. Massage therapists and mathematicians were among the most likely to get divorced.
  • Generosity: couples that split chores evenly tend to have a lower divorce rate. This was also corroborated by the National Marriage Project, which showed that generosity was the 3rd biggest predictor of a happy marriage.
  • Ideology: conservative states have a higher rate of divorce than progressive ones. This is probably due a few factors, including the average level of education and marriage age being lower in conservative states: people there get married younger, then probably grow apart as they age, and don’t have coping skills like good communication techniques to work problems out. It’s worth noting that while some liberal states do have the lowest rates of divorce, others have the highest likelihood of marriages ending in divorce.
  • Influence: people who know more divorced couples tend to get divorced more. The social effect is very powerful and its presence can be seen in a lot of other problems, such as smoking and obesity: people tend to be like their friends and family. When it comes to people, birds of a feather don’t necessarily flock together, but rather those who flock together start growing the same feathers.
  • Beginnings: couples who met during high school or college are more likely to stay together; those who met in bars are more likely to get divorced. This likely has to do with shared history, values and other things the spouses should have in common.
  • Kids: couples who have more daughters have a higher rate of divorce; couples who have more sons have a lower rate.
  • Wife: 73% of divorces are initiated by the wife.


See also:


From Mental Floss, via Neatorama

J.D. Power’s 2012 Smartphone Satisfaction Survey

They rated manufacturers, not devices, which makes it easy for Apple to win since they only make one phone, and it also happens to be pretty awesome. Consumers rated Apple significantly above the other manufacturers, and only it and HTC were rated above the industry average. The other big Android manufacturer, Samsung, was rated just slightly below average, followed closely by Motorola; LG rounded out the Android makers. The rest were all the other (non-iOS, non-Android), failed smartphone operating systems that are still drawing their last breaths: Blackberry ranked about the same as LG, followed by Nokia and, at the very bottom, the now-defunct HP/Palm webOS phones.


Next time your friends want to get an Android phone, point to the iPhone’s place at the top of the user satisfaction survey, their record sales numbers, and their crazy high 89% customer retention rate.

See also:

From JD Power, via iMore

The CDC Thinks 5 Drinks = Binge Drinking

Definition of “binge“, from Merriam-Webster:

a: a drunken revel : spree b: an unrestrained and often excessive indulgence <a buying binge> c: an act of excessive or compulsive consumption (as of food)

A couple of weeks ago, the CDC released a supposedly alarming report saying that 17% of Americans went on at least one drinking binge in the month before. The report is based on a survey which took place in 2010 and measured three drinking parameters:

  • prevalence: the percent of people in a group that binge drink
  • frequency: the number of times a month they go on a binge
  • intensity: the number of drinks per binge

One of the main issues people have with the study is that binge drinking is defined as five or more drinks in a sitting for a guy — four for a woman. That definition was clearly thought up by someone who has never consumed alcohol and is at least somewhat ridiculous, because five beers over the stretch of a football game won’t even get most people legally intoxicated, much less on par with “a drunken revel.” (The blood alcohol concentration for a 180lb man after drinking five beers in 3.5 hours is about 0.05%).  The main issue is the terms of the definition, which uses a vague “drinks per sitting” measure instead of the more precise “drinks per hour.” For example, five beers in a one-hour sitting is probably binge drinking; five beers in a five-hour sitting… not so much. And the survey treats both of those events as if they were the same.

Thanks to raw data though, the CDC’s poor definition is not quite as important: the actual number of drinks consumed are present, even if the length of the drinking sessions is missing.  The prevalence and frequency statistics, however, are just given in terms of binge drinking (e.g., percentage that binge drink), so due to the definition issues, for the purposes of the below, take the “binge” part with a grain of salt. Armed with that, here are the most interesting numbers:

  • Twice as many men as women binge drink: 23% vs 11%, and 9 drinks in a sitting vs 6
  • As people get older, fewer of them go on binges: from almost 30% for younger people to 3% for retirees
  • As people get older, their binges get less intense: from 9 drinks for younger people to 6 for older ones
  • The 3% of old people that do binge though, do it more often than any other age group: more than five times a month, vs four times a month for the rest, who probably binge every weekend
  • A higher percentage of whites and hispanics binge drink, but otherwise the races are pretty similar
  • A higher percentage of well-educated people binge drink, but they do so less often, and with less drinks per sitting: 7 drinks, 3 times a month for college grads vs 9 drinks, 5.5 times a month for high school dropouts. But only 14% of dropouts binge, as opposed to 18% of college grads.
  • Same story for rich people — although the number of drinks (around 8) per sitting is similar for all income ranges, more rich people binge, but less often: 20% of those making over 75k$/year binge 4 times a month, vs 16% of those making under 25k$/year, who binge 5 times a month
  • More of the population binges in the north, and less in the south; the rest of the country is mixed
  • Wisconsin has the highest population of bingers, at 25.6%; Nebraska and D.C. followed, with 22.3% and 21.9%, respectively.
  • Utah and West Virginia have the lowest population, at 10.9% each. Arkansas was next, with 11.8%
  • Bingers in Wisconsin also drink most: 9 drinks per sitting. Hawaii and West Virginia followed, with 8.7 each. So not many people drink in West Virginia, but those that do, are probably from Wisconsin.

The CDC’s recommendations all revolve around reducing the supply of alcohol by making it more expensive and less available: raising prices via sin taxes and selling it in less places and during fewer hours. The reason given for these recommendation is that binge drinking causes around 40,000 deaths a year. To add perspective to that figure, according to another CDC report:

  • about the same number of people kill themselves
  • twice as many die from Alzheimers
  • three times as many die from accidents (118,043)
  • 14 times as many die from cancer
  • 15 times as many die from heart disease (595,444)

In that report, binge drinking itself is not considered a cause of death, because it’s secondary: it may lead to accidents, but is not the immediate reason someone dies. From CDC, via NPR

What Killed People In 2010

For every year, the CDC compiles a list of the 15 leading causes of death. They just released their report (PDF) for the causes of death in 2010 — no word on why it took them a whole year to crunch some numbers. All in all, almost 2.5 million Americans died that year. In the report, the causes of death are given in medicalese (as defined by the ICD-10) so the below has the layman’s terms in parentheses:

  1. Diseases of heart (a.k.a. heart disease) — 595,444 deaths
  2. Malignant Neoplasms (a.k.a. cancer) — 573,855 deaths
  3. Chronic Lower Respiratory Diseases (COPD, generally caused by smoking) — 137,789 deaths
  4. Cerebrovascular Diseases (these cause strokes) — 129,180 deaths
  5. Accidents (unintentional injuries) — 118,043 deaths
  6. Alzheimer’s disease — 83,308 deaths
  7. Diabetes mellitus (a.k.a. simply diabetes, includes both type 1 and 2) — 68,905 deaths
  8. Nephritis, nephrotic syndrome and nephrosis (kidney diseases) — 50,472 deaths
  9. Influenza (a.k.a the flu) and pneumonia — 50,003 deaths
  10. Intentional self-farm (suicide) — 37,793 deaths
  11. Septicemia (infection in the blood) — 34,843 deaths
  12. Chronic liver disease and cirrhosis — 31,802 deaths
  13. Essential hypertension and hypertensive renal disease (high blood pressure) — 26,577 deaths
  14. Parkinson’s disease — 21,963 deaths
  15. Pneumonitis due to solids and liquids (lung inflammation from things that don’t belong in the lungs being there: water, food, smoke, animal dander, etc) — 17,001 deaths
  16. All other causes — 488,954 deaths


Accidents cause about 5% of deaths


The interesting this to note is that contrary to what police dramas would have you believe, for the first time in the list’s 45 year existence, murder has not made the top 15. It was usually at the bottom of the list anyway, but this year it dropped off completely. As for what actually killed people, heart disease and cancer are responsible for about half of deaths. After that, the numbers drop off pretty steeply, by about 75% — from almost 600,000 to almost 150,000.

The causes of the next close group of killers are smoking (via COPD), being old, and being careless. The latter two may be related (along with being young), and it’s worth mentioning that besides COPD, smoking can also kill through lung cancer, which is covered in the first group. All of these put together, however, don’t even get close to adding up to the deaths caused by heart disease. To put the its death toll in perspective, it would take about two hundred 9/11s to kill as many people as heart disease does every year. Maybe some of the billions spent on terrorism prevention would be more effective in preventative healthcare.


Michael J. Fox has Parkinson's. Also, Back To The Future, Part 2 supposedly took place in 2015, but we still have no flying cars.


But let’s say you wanted to avoid dying: what can this list teach us? Right off the bat, not being fat, lazy, a smoker, heavy drinker or unhealthy eater will get rid of about half the list — including a lot of cancers. Being careful — to not get infected, to not inhale your food, to not fall off the roof — takes care of almost everything else. If you don’t kill yourself, that’s another one. So if you’re level-headed and generally live a healthy lifestyle, what’s left is the stuff you can’t do anything about: being old, Parkinson’s and some cancers — all of which have no known cause.

From CDC (PDF), via NPR

Correlation Is Not Causation

Business Week has a series of graphs which use statistics that lead to ridiculous conclusions. Their point: just because two things appear to be related, it doesn’t meant they actually are; and even if they are, it doesn’t mean that one caused the other.

From Business Week, via Neatorama