Psychologist says maths can predict chances of divorce

• Tim Radford, The Guardian, Friday 13 February 2004 11.39 GMT

A psychologist claims that a newly devised mathematical model can predict with 94% accuracy which couples will divorce - entirely on the basis of the first few minutes of a discussion about some disputed issue.  John Gottman, of the University of Washington, and two applied mathematicians analysed hundreds of videotaped conversations between couples in Professor Gottman's relationship research institute. They also analysed pulse rates and other physiological data to provide a "bitterness rating" for each conversation. 

 

The researchers were looking for what they called the "masters and disasters" of marriage. What mattered was not the dispute itself, but a couple's attitudes during the argument. "When the masters of marriage are talking about something important, they may be arguing, but they are also laughing and teasing and there are signs of affection because they have made emotional connections," Prof Gottman said. "But a lot of people don't know how to connect or how to build a sense of humour, and this means that a lot of fighting that couples engage in is a failure to make emotional connections.

"We wouldn't have known this without the mathematical model."

 

The researchers will take part in a symposium on love and marriage at the American Association for the Advancement of Science in Seattle tomorrow. On St Valentine's Day, they will produce the magic ratio of positive to negative interactions that is the mark of marital success. This ratio is 5 to 1: couples who keep their tempers and consider each other 80% of the time while arguing stand a chance of celebrating their golden wedding. Those who fall below this ratio might as well dial the lawyers, or at least the marriage guidance counsellors. The team say their model charts a "Dow Jones industrial average for marital conversation".

 

Prof Gottman has spent almost 30 years trying to discover what makes marriages work and fail. In 1999, he unveiled a systematic study of conversations between 124 couples who had been married less than nine months, and rated them for emotion, gesture and attitude. The "positive" codes were for affection, humour, joy, interest and validation. And then there were ratings for disgust, contempt, anger, fear, defensiveness, whining and sadness. At the end of three years, 17 couples had divorced.

Thread with caution when building pregnancy models..

Many retailers know that if they could really anticipate our purchasing patterns and where it leads to – that this could be very beneficial to them since they could  reach the customer quicker and more efficiently. And for many retailers “holy grail” application in the family or women’s segment is pregnancy prediction.

 

We all know that  life of any individual or family is very different in terms of priorities, habits and shopping behavior – before and after baby is born. At very least no one should argue that it should be different.  So, to be able to time such “earth-shattering” event where old world is gone and a new star is born – and then “help” that individual or family by paddling your own products ahead of competitors - can really get you large share of their wallets on purpose of serving their needs better than competitor. What is wrong with that? Well, few things can go wrong here, mostly in “privacy” department, and some “smarties” who went ahead of themselves eventually learned their lessons and they had to move a few steps back.

Lets’s start with conceptually outlying how you could build pregnancy predictive model, before putting a few warning signs, kind of “proceed with caution” or “danger ahead”.

The first thing you need to do is to put the "carrot on the hook" for any female customers who would be willing to share their pregnancy secret with you (first or second trimester preferable) for some hefty promotional discounts. Once you have a critical mass of newly pregnant customers – it is just a matter of capturing their purchasing history, so that you are able then to differentiate between them the rest (non-pregnant segmented) in the form of robust and accurate predictive model. Once, such model is in place it is a matter of implementing it, monitoring it and measuring value it generates.

All sound well and good - here is reality..

Once upon time there was one very clever man, in very clever marketing department of one forward-thinking retail company. And that man created a very smart data-mining model who could predict if woman customer is pregnant. Soon after mailing list followed to its likely pregnant female customers. As the story goes there were some very impressed customers who were amazed with “how did they know”? But they were some who were not so impressed, and they asked different questions of “how did they dare to know”? There were also some who felt wrongly “impregnated” like the father who stormed marketing department accusing them of leading his teenage daughter into getting pregnant - so they can sell to her their new range of baby products. But then, a few months later the same father end up sending letter of apology after discovery that his daughter was indeed pregnant!. Not to say that he was being completely stunned by how this retailer knew something he did not - even though his daughter lived with him. 

The biggest problem was that many customers felt spied on, feeling that their privacy was compromised, so they started cutting ties with this retailer and doing everything they could to hide their purchasing behavior. This prompted retailer to adjust accordingly their model execution. And the only remedy was to blur the fact that they had such probabilistic knowledge. This resulted in promotions where baby-products coupons were masked with other vouchers, and therefore it was no longer obvious that marketers had such knowledge, which kept customers at ease.  So, if you are competing for baby product market think carefully about how you navigate through this. Could be some stormy waters just when you think it is smooth sailing.

Goran Dragosavac

      

Text  Mining on F-word

I have a colleague who works as the analytical practitioner and recently she was involved in banking project where they were analyzing free text data collected online.

The idea was to hear who is talking out there about this company,  what are they saying, how influential are the voices, what is the sentiment, what is the critical mass ad so on. And no better words to start your exploration of negative sentiment than F-word, and then go on from there.

Next thing my colleague had done - was to use a technique called concept linking which takes selected word, in this case F-word, and produce a graphical display of the linkages between that word and other entities. And the thicker links would indicate a stronger connection between the words.

So, there she was, sitting with a senior bank manager who was probably dressed in a grey suit and tie,  using some neat technologies for linguistic exploration to find the most F–ed up areas of the business.  Isn't this just pure pragmatism! Basically - let’s see what are the customers most angry about before we see  if we can do something about it.

 

Next time someone throws expletive in your face – don’t get angry, try to learn from it!  
 

Goran Dragosavac