The DoubleThink » data

The US Census – Mother of all Surveys

Dimitri — Wed, 23 Jun 2010 21:56:38 +0000

Imagine this – a Census worker walks on dirt road deep in the woods somewhere in North Michigan. After half an hour she reaches a bungalow, the 1^st sign of life she’s seen in a while. Just as she pulls out her new 2010 Census form she sees a bag hanging from the doorknob. In it she finds a Census from 2000 … .

This is the story Dr Robert Groves used at the beginning of his presentation at the ARF Audience Measurement Conference in NY yesterday. He runs the US Census Bureau, the government bureau in charge of surveying every household in the US once every 10 years. The Census is used to obtain the vital demographic statistics that are necessary for government planning. Dr Groves’ story and subsequent presentation showed that it is probably one of the biggest logistical exercises in the world.

The scale of the operation is pretty incredible. The government sends out the questionnaires to 140 million households and looks for a 100% completion rate. In this year’s Census 72% of the US households mailed their survey back. That is the same response rate they had when they did the last census 10 years ago. Any direct marketer would dream of having response rates at the same level of 10 years ago so this was a phenomenal achievement. A 1% drop in response rate would have cost the Government $80 million. This is because if you don’t respond by mail, they will send around someone to your house to ask you to fill out the questionnaire. They have a small army of 600k census workers following up on the 47 million households who didn’t respond. This in-person follow-up visit costs the government on average $57. (This includes the costs of handling the 300 assaults on census workers committed by anti-government minded non-responders!). The cost of a response by mail is only 42 cents so with these economics it became crucial for Dr Groves’ team to drive up the response rate.

They did this in a number of ways. First they made the questionnaire shorter which always drives up response rate. Then they added 2 touches to their communications cadence. They sent out a pre-announcement card before the questionnaire was sent and they sent out an extra reminder after the 1^st questionnaire was sent. This apparently consumed the entire printing capacity of the US!! They also paid a lot of attention to language. The questionnaire came in 6 languages and they had completion guides for an additional 59 languages! They also activated 250 thousand grass roots organizations who would work in the local communities to ask people to fill out the questionnaire. This according to Dr Groves was one of the main contributors to the high response rate.

Then they did a marketing campaign which cost them $350 million, an investment they could easily show the positive ROI of (all they needed to achieve break even was a 4% increase in response rate …). This was a fascinating campaign, perhaps the biggest multicultural campaign ever. It included a K-12 campaign designed at informing children of immigrant families about the Census. These children are often the 1^st to learn English in the household and can therefore be the main influencers. The campaign was run by a consortium of ethnic agencies who had to buy media in 28 languages, including Farsi, Armenian, Uruguayan etc. I saw Khmer videos, Indonesian outdoor ads and native Hawaiian posters. I have never seen localization at that scale.

Equally impressive was the use of real time tracking, modeling and optimization. Dr Groves’ team implemented daily tracking studies and performed statistical analysis every night to determine the effect of their media investment on awareness and behavioral intent. This allowed them to identify pockets of the population that were lagging behind. 18-24 year olds were an example. They tend to think the Census is not for them – that it is only for old people … . The team saw this very quickly and acted immediately by sending out targeted messages and by activating grass roots organizations. Dr Groves said the government saved 100’s of millions of dollars through real time tracking and optimization.

At the end of the presentation we got a sneak peak at some early results. The big trends shouldn’t come as a surprise. The population will grow from 309m to 439m in 2050. 20% of all households will be over 65 years old in 2050. There will be significant growth in the Hispanic population from 8% in 1980 to 30% in 2050. The Asian population will grow significantly as well. Apparently 47% of the national fertility in the next decades will come from minorities. We could also see the emergence of a bi-modal income distribution with an increasing gap between 90th and 10th income percentile – $138k vs $12k.

The US Census is a remarkable project. In terms of scale it is maybe only rivaled by the Indian Census which, as I found out on my trip to Bangalore last week is scheduled for the first time in 2011. Imagine surveying a billion people, a large portion of which live in remote, underdeveloped rural areas. I wouldn’t want to be in charge of that job!

Next Generation Geo Marketing

Dimitri — Wed, 09 Jun 2010 12:00:06 +0000

A friend just sent me this beautiful visualization of where people are taking pictures in London. You can clearly see the high traffic areas around the tourist hotspots. It’s a great use of the type of data people are generating by going about their everyday lives. It also shows how this data could become really practical for marketers trying to find out who is where at what moment in time. I am sure the picture would change if we were to look at certain times of day, days of week, seasons or even profiles of the people that are taking pictures. The view would be very different for people who live in London vs those who don’t for example.

This type of location based behavioral data could give marketers very valuable information about when they need to be where in order to surround their target audience. It seems this data would be able to provide much more detailed information than what is currently available. While this particular application only looks at where people are taking photographs, other location based services like Foursquare generate tons of data about what people are doing where. Foursquare already automatically shows their users which locations are trending (ie becoming popular) real time.

The application of location based behavioral data will go well beyond location based targeting on mobile devices. I believe it will soon become the basis for the next generation of geo marketing tools. Retailers for example could use it to determine where to locate their stores and media planners could use it for planning their outdoor advertising. We could even use this new type of data to profile zip codes based on the activities of the people who live there. These zip code profiles could be used in the same way marketers are now using socio demographic profiles of zip codes for targeting. The current profiles are pretty static since they are based on more traditional sources like the Census. The behavioral location based profiles could be much more dynamic and a lot more detailed. A company like Foursquare, if it wanted to, could actually already very easily make these zip profiles available to marketing professionals. This would make them a real competitor to established 3^rd party data vendors like Acxiom and Experian.

The Value of Stories

Dimitri — Fri, 07 May 2010 21:01:18 +0000

What is the actual value of a story? Are people prepared to pay more for something if there is a story attached to it? It turns out that they are. That is the outcome of a very original experiment by writer / NYT columnist Rob Walker. I saw Rob speak about his Significant Objects project, a project which I think might have generated a very significant dataset.

Here is what Rob did. He purchased a bunch of random objects on Ebay, most of them for less than $4. He then distributed the objects to his friends, fellow writers and artists and asked them to write a short story about the object they were given. He then put the objects back up for sale on ebay with the story in the description to see if people would pay more for the object now there is a story attached to it.

It turned out they did. When he sold the first 100 objects, he essentially sold $120 worth of objects for $3,612 – a 2,776% significance markup as he calls it. The object in the pictire above was the highest seller of the 1st wave of products he sold – $193 for a statue he initially bought for $3. He even sold a 25c plastic banana for $75.

So people were prepared to pay a lot more than the initial value of the storyless object. In fact the difference was so big that Rob concluded that the real value was in the story – the object was merely the vehicle for the story.

That’s very interesting and it obviously gives us a view in how brands work. But what I find most interesting about the experiment is the dataset it created that showed the exact monetary value of stories. I would love to analyze that data to see what really drove the value of the story. Was it the object, the synergy between the story and the object, the reputation of the writer or are there any attributes in the story itself that give us clues about what creates its value.

The data is available on the website. I might play around with it if I find the time. Here is the video of Rob Walker’s presentation at PSFK.

Your voice reveals your subconscious

Dimitri — Tue, 06 Apr 2010 16:27:25 +0000

I met a slightly creepy yet really interesting company yesterday. Voiceprism has a technology that analyzes sound waves generated by the human voice. This gives them the ability to listen to an individual for 10 minutes, establish a baseline voice pattern and then detect the deviations from the normal frequencies. These deviations have proven to be indicative of emotions such as excitement, stress, delight, anger, …

This reminded me of the biometrics Innerscope looks at to gauge emotional engagement. The voice could be an additional data point that can give us clues about what is happening in the subconscious. The main differentiator for this technology is its ability to listen in on thousands of phone conversations. This means it can be used at scale in a cost effective way. I haven’t seen another technology that has the potential to capture data from the subconscious for potentially thousands of consumers.

They talked to me about their work with insurance companies where they analyze claims calls and flag insurance claims that should be investigated further based on the stress levels of the callers. They were basically doing automated lie detection. This apparently helped insurance companies dramatically reduce fraud levels.

I think there could be other interesting applications, especially in quantitative research where it is often hard to quantify true emotional engagement through analyzing what people are saying. How they are saying it can now apparently give us valuable clues as well. Click here to see a video of how the technology works.

TV’s emotional impact

Dimitri — Fri, 12 Mar 2010 20:34:26 +0000

In my previous post I listed some empirical evidence that shows that TV advertising is becoming increasingly effective. One of the reasons is that TV is a superior medium for driving emotional engagement. The role of emotional engagement in driving purchasing behavior has been hotly debated in the last couple of years. This is as a result of advances in neuroscience that have demonstrated that the decisions people make are often a result of what happens in their sub-consciousness. As a result the traditional linear AIDA model, where the role of communications and advertising is to move customers down a linear path from awareness to interest to desire to action by grabbing their attention, has been severely challenged.

Robert Heath has published extensively on this topic and has shown that the role of emotions in the decision making process are very important. In a 2009 Admap article about how TV builds brands he said :”Early advertising models that addressed the role of emotion in advertising reflected the thinking of the time, which was that ‘conscious thinking’ leads to ‘feeling’, which leads to ‘attitude change’, which, in due course, leads to a purchase decision. They were not to know at the time, but they got it seriously wrong. Feelings and emotions are processed much more quickly than thoughts.” Heath has used brain scanning technology to show that we usually make decisions up to 1/5th of a second before we are aware of them. He says that “we always form an attitude about a decision through emotional and subconscious rational processing before we start to consciously and actively ‘think’ about it. So our conscious thinking tends either to support the decision or counter argue it. There is empirical evidence […] showing that the presence of brands inhibits processing of product attributes and encourages consumers to use shortcut to validate their brand choices.” This is illustrated in the diagram below.

If people use brands as subconscious short cuts, then brands need to be built through communicating at the subconscious level.

This is where TV is especially powerful. Contrary to other media such as print that require conscious processing of information (ie reading), TV more than any other medium can work at the subconscious level by evoking feelings through the uses of visuals and sound. This is why TV is such a powerful medium for creating brands. It does this through emotional communication. And Binet and Field have demonstrated, in the study of the IPA databank I mentioned in my previous post, that emotional campaigns consistently outperform rational campaigns in terms of driving sales, share, price elasticity, loyalty and penetration.

Google Analytics Intelligence

Dimitri — Mon, 01 Feb 2010 12:00:17 +0000

This is a bit of a geeky post so apologies in advance. But I have been meaning to write about how fantastic I think Google Analytics Intelligence is. If you are not sure what I am talking about have a look at the video below.

I love it for a number of reasons. First of all it finally embeds very basic analytics best practices in web analytics software. It has bothered me for years that some of the most sophisticated web analytics packages did not even have the most basic significance tests in their standard reports. I have often seen people draw conclusions on the back of results that were statistically insignificant. GA Intelligence will really help avoid this.

Second, it automates some of the mundane analytical tasks. A few rather pessimistic analytics professionals have asked me whether this was the beginning of the end for analytics as a profession – the start of the complete automation of analytics. I don’t see that happen anytime soon! If the basic, repetitive tasks are being automated then that would free up time for analytics professionals to use their skills much more efficiently on more advanced work. There is plenty of that work to be done.

The most important reason why I love GA Intelligence is the seamlessness with which they have incorporated the significance tests in their reports. It is a fantastic example of how to distribute the ability to do more complex statistical analysis among a broader non-technical audience. And the alert system is very elegant as well.

I love it!

What should I pay for a piece of data?

Dimitri — Mon, 18 Jan 2010 13:00:30 +0000

People have been buying and selling data about consumers for a long time. Companies like Axciom have been doing this for years in the direct marketing business. But recently a new breed of companies has been popping up who are acquiring and selling data about consumers in the online advertising space. Companies like BlueKai and Datran are the modern day, digital equivalent of the Axcioms. The good old direct marketing techniques from the 80’s and 90’s are now also being used for targeting online advertising ads. Therefore data collected about consumers can now be used for smarter targeting across all direct and digital channels. This broader playing field will dramatically grow the size of the data reselling business in the next few years.

So with so many companies buying and selling data about consumers, what really determines the value (and therefore the price) of a data point? I think there are 3 drivers.

1. Predictive Power

The 1^st driver is Predictive Power of the data point. Let’s say for example that I am a manufacturer of drills and that I am trying to purchase data points that will help me identify whether a consumer is interested in buying a drill. And let’s assume that I can choose between the following 2 sets of data points.

Set 1	Set 2
Number of hours spent on DIY per week	Number of vacations taken per year
The number of hammers owned	Interest in water sports
Size of the house owned	Age

Most people would agree that the data points in set 1 are more valuable for a drill manufacturer than those in set 2. This is because of their natural correlation with someone’s likelihood to purchase drills. This example is very straightforward. If you had to determine the Predictive Power of a 100 different data points however, you would have to build statistical models that predict the likelihood of someone buying a drill based on all 100 data points. Those that enter the model have a high Predictive Power which can be quantified by the lift they generate in the models. Whether you build statistical models or not, the principle is that data points with a high Predictive Power will improve our prediction of whether a consumer will be interested in buying a drill and, as a drill manufacturer, I am prepared to pay a higher price for them.

2. Recency

The 2^nd driver is Recency. This is really a special case of Predictive Power but I want to call it out separately as it has become an increasingly important driver. In a digital world people often reveal real time what their intentions are. Knowing whether a person has searched for drills on Google, whether they have clicked on a banner for drills or whether they have seen a drill related video online can be very powerful. These data points generally outperform the more traditional data points that are listed in the example above because they are direct indications of a consumer’s interests and needs at a certain point in time. For these self disclosed data points, Recency is very important. When someone searches for a drill on Google then that is very valuable information if I can target that person immediately. However, if I know that someone searched for a drill 3 months ago then that single observation in itself is a lot less valuable. The predictive power of self disclosed data points starts to decline minutes after the observed event. Because of the disproportionately high value of very recent data we anticipate most of the future innovation to focus on capturing multiple events real time and shortening the cycles between observed events and the ability to use that knowledge for targeting. This is already happening on advertising exchanges through the introduction of Real Time Buying.

3. Exclusivity

The final driver is Exclusivity. Let’s use the same example and let’s assume that I can only buy the data points in set 1. Let’s also assume that I have built a statistical model and have determined that the general predictive power of the number of hammers a consumer owns is far more predictive than the other 2 data points. I would be prepared to pay a relatively high price for data on hammer ownership. Now consider an alternative scenario where one additional data point is available: the number of nails a person uses per year. Let’s assume that the general predictive power of nails consumption is almost as high as that of hammer ownership. The availability of nails consumption will have an effect on the price I am prepared to pay for hammer ownership. It’s the basic laws of supply and demand.

In the next few years the buying and selling of data will undoubtedly become a lot more streamlined. When that happens, the market drivers described above will increasingly determine the price companies are willing to pay for information about their consumers. Consumers on the other hand will get a much more transparent view of the value they are generating by allowing companies to collect their data. Who knows, maybe they’ll even be able to claim their share of the pie.

Army Swimming in Sensors and Drowning in Data

Dimitri — Tue, 12 Jan 2010 12:00:23 +0000

On Sunday I wrote a small post about how UK company Prozone captures all English Premier League soccer matches on video and then uses that footage to create a database of every shot, pass, tackle and assist made by every player (according to a comment on that post Opta Sports are powering the Guardian Chalkboard – sorry I got that wrong). That’s a lot of data. However, as I found out in the New York Times this morning, this is nothing compared to the volume of footage captured by the US Airforce spy drones. Apparently it would take one person 24 years to watch all the surveillance footage captured in Afghanistan in 2009. According to the article, the Army has a group of young soldiers watching every second of this footage live in order to detect suspicious activities.

The Army expects the volume of footage to grow exponentially in the near future as new drones are being deployed that can film at up to 65 different angles. Processing all that footage will soon become unmanageable. The Airforce top intelligence official LT. Gen. A. Deptula said the army could soon be “swimming in sensors and drowning in data”. So they are looking for new analytical techniques that can help reduce the volume of footage that needs to be analyzed by the human eye. Now where do you think they started their quest for this technology? You got it – the sports networks. Apparently Army officials have been shadowing the ESPN broadcast trucks outside football stadiums to learn how the TV network is tagging and retrieving all the highlights in their NFL footage. It sounds like Opta Sports and Prozone have just found a second market.

One Army official called out the limitations of automated recognition technologies for the Army. He said that when it comes to the use of force “You need somebody who’s trained and is accountable in recognizing that that is a woman, that is a child and that is someone who’s carrying a weapon. And the best tools for that are still the eyeball and the human brain.” But even if technology can identify the 10% of footage that needs to be analyzed by humans it would still make a big difference.

Landon Donovan Visualized

Dimitri — Sun, 10 Jan 2010 16:09:45 +0000

Yesterday Landon Donovan tore up the right flank of the pitch at the Emirates Stadium and helped his new club Everton secure a draw against my beloved Arsenal. This morning one of the soccer blogs I follow showed this great visualization of Donovan’s impact on the game.

by Guardian Chalkboards

The graph shows Donovan’s completed passes in blue. If you roll across the dots you can see when he passed to which team mate.

The graph is built with the Guardian’s interactive chalkboard which has apparently been around for almost a year. I have to say that I am ashamed that I only found out about this fantastic little tool today. OK, I have been living in the US now for 6 years and I have started to refer to the beautiful game as soccer (mainly because I discovered the other beautiful game of American football). And coverage of soccer in the US media is spotty at best. But that should still not be an excuse for not knowing about this amazing interactive visualization tool.

It uses data collected by UK company Prozone who record all Premier League matches and use that footage to code all data on passes, shots, tackles etc for every player on the pitch. The guardian then build this very easy to use application that allows every aspiring Arsene Wenger (who was one of the early Prozone adopters) to analyze the patterns of the game. Here are a couple of great examples.

What is probably most impressive about this little application is the ease with which one can share chalkboards with others. This makes it a potentially powerful social media tool. And with most of the coverage of soccer in the US happening on social media, it could be a great tool to educate the US audience on the tactical aspects of the game. Nothing gets America more excited about sports than stats and analysis!

Listening is easy

Dimitri — Wed, 25 Nov 2009 19:41:27 +0000

It seems like every digital marketer and agency is talking about the power of listening to your consumers in the digital space. This is fuelled by an entire industry of online listening technology providers that have popped up over the last few years. They all promise the ability to mine the incredible wealth of information available on blogs and social networks to generate insights about what consumers talk about, what’s on their minds and what language they use to talk about the brand or issues related to the brand. While the technological possibilities seem limitless I have still only seen a handful of clients use the information to actually improve their communications. Earlier this week, Emily Steel from the Wall Street Journal wrote about how Ogilvy helped IBM and Harrah’s do this.

I would just like to spend 2 minutes on the Harrah’s case study for 2 reasons. First I would like to give the credit for the work to Jim Dravillas and his team. They have really done tremendously inventive work for Harrah’s over the years. Sorry Jim – in terms of recognition I realize the doublethink isn’t quite the Wall Street Journal!

The other thing I want to point out is how the Harrah’s example shows that action is more important than listening. The Harrah’s analysis was surprisingly simple. While we have used online listening technology in the past, in this case Jim and his team simply read the top 50 tripadvisor.com ratings and categorized the topics they saw using common sense. My grandmother could have done this (no disrespect to Jim and the team – or my grandmother). What was brilliant about this case however was that the team came up with clear and simple recommendations upon which they acted by changing the communications. It’s this last step – taking action – which is always the hardest. Listening is easy.