Big Data, Plainly Spoken (aka Numbers Rule Your World)

Who rewrites headlines? Man or machine?

As many of you know, Netflix is the most highly valued (or, overvalued) stock in the universe. Therefore, we come across lots of breathless reports about how amazing its business is. Investors are willing to pay $85 for each $1 of earnings, which is roughly in the Google range when Google went public.

 

This little snippet caught my eye when I looked at Google News last night.

 

 

Wow, 1 in 8 Americans have Netflix, that's a lot of people! With roughly 300 million people, 1 in 8 would be 38 million customers. As a non-subscriber, I would feel left out!

Then, I clicked on the link, which led to a CNN Money article.

The strange thing is the article has a different headline. I'm not sure if the Google machine created a different headline for use in Google News, or someone at CNN Money wrote a headline specifically for Google News, or what, but it's interesting they are different.

Now, 7% of Americans is not the same as 1 in 8. Seven percent of 300 million is 21 million, which is close enough to the 24 million reported in the article. Alternatively, divide 300 million by 24 million, and we have 1 in 12 or 13 Americans. That's still a lot of people but 1 in 12 ain't 1 in 8.

 

***

 The article notes that Netflix just surpassed Comcast's number of subscribers. How impressive is this feat? Perhaps the writer should look into "price elasticity": demand goes up when price goes down?

Netflix is getting roughly revenues of $30 per user per quarter, or $10 per user per month for its 24 million customers. Comcast, by contrast, gets $80 per user per month, 8 times as much, with a similar number of customers. (Comcast has a price-to-earnings ratio of $20.)

From a revenues perspective, Netflix needs to add 8 customers for every customer Comcast adds. One wishes CNN Money tells you about this while it breathlessly reports on Netflix's rampant growth.

 

Squeezing our kids' salaries and sending them into debt

I wondered (in this prior post) why many economists, including Ben Bernanke, propose that the solution to our unemployment crisis is to send our kids to college. Their argument appears to be based on the statistical observation that the unemployment rate for college graduates is markedly lower than that for high school graduates.

That sounds good but must deal with the reality that many college graduates are not finding jobs. See this chart from a survey by Adecco of recent college grads:

Only about 60% have full-time jobs (it's unclear when after graduation these surveys were conducted but based on another question about how long their job search were, I'm imagining up to a year after.). And this is likely to be an over-estimate because of non-responder bias: college grads who failed to find jobs are less likely to respond to such surveys.

***

Although I don't have the data, I suspect that the unemployment rate for college graduates, referenced by those economists, can be broken up into unemployment for new college graduates and unemployment for college graduates who have been in the work force for a while. I suspect that the second group, which accounts for a much bigger portion of college grads  by way of accumulation over time, does not suffer from serious unemployment and thus, the overall average for college graduates hides an ugly number for new college graduates.

If your kid listens to these pundits and goes to college, upon graduation, the job market he/she faces will look like the one for new college graduates, not the one for grads who have been working for a while. (In Chapter 3 of Numbers Rule Your World, I call this the "dilemma of being together".)

If the market is not clearing now for new college graduates, encouraging more kids to get college degrees will only worsen the unemployment rate for new college grads. It will also drive down starting salaries for new college graduates. It will also stick many more of our kids with heavy education debt. (This article tells us student loans just exceeded credit card loans.) It will also have a long-term impact on those graduates who eventually find jobs (think value of money and lagging behind one's cohort.)

Know your data 6: Apple think different, not!

In case you missed the news, your iPhone and iPad are recording your every move. Researchers have found a secret file that "contains the latitude and longitude of the phone's recorded coordinates along with a time stamp". The Guardian (UK) reports that it's unclear how this file is being used although PC Magazine tells us that "forensic" analysts and the police have known about this file and have been using it for some time.

As I pointed out before, I believe that such data collection for app development or marketing purpose should be opt-in. And "opt-in" does not mean burying some legalese in dozens of pages of fine print ("Terms and Conditions") which gives customers the choice of letting (insert technology company) do whatever they want or don't use their product.

In addition to the stench, those T&Cs do not provide enough disclosure. If companies want to collect tracking data, they should specify exactly what data is being collected, how it is stored, how long it is retained, why the data is being collected, whether the data is being exchanged with third parties (and who those third parties are, and what they are using the data for), whether the data is being sold for profit, etc.

P.S. This report says that Android phone does the same thing.

A cancer problem switches gender

The New York Times recently printed a valuable article discussing the PSA screening test for prostate cancer. I previously discussed the statistics of this test here.

The entire article is worth reading. Here, I summarize the key issues:

• Prostate cancer is a slow-growing cancer, which means that many men die with it, rather than die from it. This means that autopsies will identify many men with prostate cancer, many of whom did not die because of it. If these men were to have received treatment in their lifetime for prostate cancer, they would still have died the same way they died.
• Detecting prostate cancer late in life is rather useless because many men would have died from other causes before these cancers could have killed them.
• Screening tests always generate many false positives. The false positive rate depends on the proportion of those tested who actually have the ailment: there are more false positive errors when the tests are given to people who are less likely to have the ailment. Thus, the younger men taking PSA tests will suffer a higher chance of false positives.
• If you are given the knowledge that there is a chance you have prostate cancer (even if it is a small chance, say 30%), you may elect to do a biopsy to buy yourself "peace of mind". Likewise, the doctors may want to insure themselves against future negligence lawsuits, or may simply want to earn the fees associated with the surgeries.
• Since most of those being treated are false positives, the treatment will have zero benefit, and some unlucky patients will suffer side-effects.
• False positive is a laboratory term. Your test result is a positive, period, not a "false" positive. You must make the decision whether to treat it or not - and risk side-effects - before you find out if it is a false positive. (If you know it is a false positive, then you already know you don't have cancer, which means you don't need to be screened in the first place.)
• The PSA screening test is a multi-billion-dollar industry.
• These are the same issues raised during the mammography controversy. One ought to be consistent and gender-neutral in the way one approaches each.

These concepts were covered in Chapter 4 of Numbers Rule Your World.

***
While the article is great, the title is not. In the print edition, it is called "Screening prostates at any age". The article is primarily concerned with whether elder men should be screened at all.

Nice summary of coffee studies

The LA Times has a good article discussing recent findings about whether coffee is good or bad for our health.

This is the type of problem that demands a statistical model to establish a cause-effect relationship, and the Times tells readers, appropriately, that we are not quite there yet:

Definitive proof that coffee is good for the blood vessels is unlikely to emerge anytime soon, Hemmen says. Such studies would need to randomly select people to drink either a lot of coffee or a little coffee, and then researchers would have to closely monitor their coffee intake and health for decades.

What we currently have is circumstantial evidence, in other words, evidence of correlation only. Such evidence is not definitive for a number of reasons, including:

• The studies examined certain subgroups of the population, which makes generalizing to the whole population a bit of a stretch. For example, the 2008 Finnish study concerned male smokers who drink 8 or more cups a day compared to those who don't drink at all. How comfortable are you that this result applies to the larger population?
• Multiple studies with roughly similar findings do help increase our confidence in the result but having many studies does not mean that all relevant segments of the population have been covered, nor should we treat every study with equal weight. The Times did great to give us some details of each study, such as the sample size and the composition of the population under study.
• The data comes from "observational studies", meaning that people choose to drink whatever amount they choose. The challenge with this type of study is that we cannot eliminate the possibility that coffee drinkers tend to be more health conscious, and therefore, it's not the coffee that is doing the magic but other unobserved traits. Now, you may not agree that this correlation is plausible but you also can't claim it is not possible. As the article points out, research does show that there are differences in behavior between coffee drinkers and non-drinkers.

As one of the interviewees explained: "[The observed benefits] may be due to some other factors we haven't even taken into consideration." The reason why a randomized experiment, in which people are randomly assigned to drink either more or less coffee, is more reliable is that random selection ensures that if some unknown factor is affecting the results, it affects the results of both test groups evenly, not biasing the relative results.

Given the nature of these results, the Times does well not to hyperventilate, telling readers that "Such studies reveal that coffee isn't harmful, as once thought, and might even be beneficial."

 

 

The case for more false positives in anti-doping testing

Here comes the promised second installment to my recent post on anti-doping in which I argue that we should pay a lot more attention to false negatives. Here's the last paragraph: 

For me, the difficult question in the statistics of anti-doping is whether the current system is too lenient to dopers. If the risk of getting caught is low, the deterrence value of drug testing is weak. In order to catch more dopers, we have to accept a higher chance (than 0.1%) of accusing the wrong athletes. That is the price to be paid.

Note that this post has been contributed to the Statistics Forum (link), which is a new blog sponsored by the American Statistical Association and edited by Andrew Gelman, and is reprinted here. You can click on the link above or scroll below to read the full post.

FIFA throws weight behind statistical software

According to Goal.com, FIFA, the world governing body for soccer (football), has decided to use imaging technology to review goal-line decisions by referees that often lead to real goals disallowed erroneously -- and hear this, FIFA will use statistical software to predict what might have happened after the goals were reinstated, and change the match scores accordingly.

As an example, in last year's World Cup, England had a goal disallowed erroneously in the otherwise disastrous match against Germany, which ended with England losing 1-4. Based on statistics, the score will soon be revised to 2-5. Goal.com explains:

We then used a programme based on state-of-the-art technology provided by the makers of Championship Manager to simulate what would have happened had England levelled the score at 2-2 and the final score is now 5-2 to Germany.

We applaud this audacious move to embrace statistical analysis. This move now puts soccer ahead of other professional sports. April 1, 2011 will be a day to remember. Read the details here.