WSJ Graphs Showing Increases in Students Studying Statistics

A recent WSJ special on Big Data had a section call Help Wanted!  The first paragraph reads:

“Corporate executives face a daunting obstacle when it comes to reaping the benefits of big data: Who’s going to tell them what it all means?”

This is a real problem.  As firms collect more and more data, they will need people who can productively make sense of it.

Fortunately, more people are getting trained in statistics (and analytics) to meet this demand.  The WSJ ran an article called “Data Crunchers Now the Cool Kids on Campus” (the statistics are from that article.)

And, IBM, Ohio State, and others announced plans for an advanced analytics center in Columbus, OH to help train, do research, and help firms with analytics.

There is a lot of opportunity for people who can work with and make sense of data.


Economist Feb 9 2013 Print Edition

The Economist recently published an interesting article on how advances in security cameras are allowing stores to better track and respond to consumers.

It seems like it wasn’t too long ago that an average 3-year-old could easily beat a computer at determining someone’s gender by looking at a picture.  Now, these cameras are getting better at determining gender as well as age of the shoppers in the store.

This advance really allows in-store retailers to catch-up with their on-line counterparts in terms of understanding customers:  where do they spend their time, how long do they stay, what do they look at, and what do they ultimately buy?

The article provides an example of a retailer who determined when the number of shoppers peaked (it wasn’t when sales peaked) and built their staff schedules around that to generate more sales.

There is a lot more retailers can do with this information.  It will be interesting to see how this evolves over the next few years.

BusinessWeek recently highlighted Peter Kooman of Optimizely in its Innovator column.

This article talks about a statistical technique, A/B testing, that is becoming a more mainstream.

A/B testing is a way to test two versions of the same website and see which one works better.  That is, a random group of visitors to a website are shown page A, and the other group is shown page B.  Then, you track what each visitor does.  An A/B test is a statistical test that you apply to show which website did better.  “Better” could be measured by more views, visitors staying longer, visitors deciding to buy, or visitors deciding to buy a certain item.

Right away, you then start using the better of the two websites.

Mr. Kooman is quoted as saying that they were able to increase donations on the Bush Clinton Haiti Relief website by 10% by placing a picture of the disaster next to the “Donate Now” button.  They have also used this technique to determine which headlines will get more clicks for a news agency.

You can think about A/B testing like hypothesis testing from your statistics classes.

High-traffic websites are a great place to use a massive number statistical tests to test ideas.  It is very easy to create a test– just create two versions of the same page.  So, instead of laboring over coming up with the best design, best layout, best price, and so on, you can simply test different ideas and let your customers determine which one is better.

And, with a website, you can implement the better idea right away.

In yesterday’s post, we discussed a short definition of Analytics (“the ability to collect, analyze, and act on data.”).

Of course, this is a broad definition.  To determine your appropriate analytics strategy, it is important to understand the different categories that make up the field of analytics.

A team from IBM published one of the better definitions of the three categories of analytics in the Analytics Magazine.  The article is worth a read.  Other organizations and universities, are also converging on the same three categories.  These three categories, from INFORMS, are:

Descriptive analytics

  • Prepares and analyzes historical data
  • Identifies patterns from samples for reporting of trends

Predictive analytics

  • Predicts future probabilities and trends
  • Finds relationships in data that may not be readily apparent with descriptive analysis

Prescriptive analytics

  • Evaluates and determines new ways to operate
  • Targets business objectives
  • Balances all constraints


As a side note, the article in Analytics Magazine also is careful to point out that the above three categories apply to structured data.  There is a branch of analytics that applies to unstructured data like analyzing consumer sentiment using Facebook and Twitter or analyzing written text for medical diagnosis like IBM’s Watson (from Jeopardy  fame).


Price Analytics (also known as Revenue Management) is on the front page of today’s Wall Street Journal (you need a subscription to view online).

Revenue management has long been practiced by the airline industry and by some hotels.  But, I’m guessing that the reason this article was on the front page is that these practices are being applied to everyday items.   The article mentions that items like consumer electronics, shoes, clothing, jewelry, detergent, and razors may change many times per day.

As more shoppers go on-line to buy, prices become more obvious.  And, some shopping sites will list the lowest priced items first (which helps drives sales).

This is an interesting trend.  Revenue management transformed the airline industry.  It could do the same for retail.  It also means that pricing has become more strategic and dynamic for retailers.  They will need to to deploy sophisticated analytics to maximize revenues and profits.


On a side note, the article was a little vague on the direction of pricing and I think they fell into an innumeracy trap.

The gist of the article was about getting lower prices to get to the top of rankings.  However, the article mentions:  “So far, shoppers are winning the price game about as often as they lose— with about half of price changes going down, and half going upward…”

The innumeracy problem is that they don’t mention how much prices decrease or increase (if the average increase is 5 cents and the decreases average 10 cents, that tells you something) nor do they mention low long the prices are at their current levels (if the price increases only last for 10 minutes and the decreases last 10 hours, that tells you something).

So, before we judge how this impacts the shopper, we should know these answers (which I’m sure someone knows but are never going to reveal).




For a Web Application that relies on recurring revenue from customers (that is, customers who continue to pay monthly for service), it is especially important to keep your customers.  (This, of course, applies to many types of businesses.

The CEO of a Treehouse wrote a short article on how they used analytics and statistics to better predict when a customer might cancel.  They could then use this information to reach out to the customer to see if there was a problem they could fix to keep the customer.

This article mentioned a few of the tools they used, but did not provide too many details.  I think it was interesting that the article and comments talked about the false positives.  I think this is important.  We often forget that these tools can never be 100% accurate.  We need to understand these issues when making decisions with models.

Of course, this type of analytics applies to many different types of firms.