Category Archives: Consumer Psychology

When Are Crowds Not So Wise?

the-wisdom-of-crowdsSince James Surowiecki published his book “The Wisdom of Crowds”, the common wisdom is – well – that we are commonly wise. In other words, if we average the knowledge of many people, we’ll be smarter than any of us would be individually. And that is true – to an extent. But new research suggests that there are group decision dynamics at play where bigger (crowds) may not always be better.

A recent study by Iain Couzin and Albert Kao at Princeton suggests that in real world situations, where information is more complex and spotty, the benefits of crowd wisdom peaks in groups of 5 to 20 participants and then decreases after that. The difference comes in how the group processes the information available to them.

In Surowiecki’s book, he uses the famous example of Sir Francis Galton’s 1907 observation of a contest where villagers were asked to guess the weight of an ox. While no individual correctly guessed the weight, the average of all the guesses came in just one pound short of the correct number. But this example has one unique characteristic that would be rare in the real world – every guesser had access to the same information. They could all see the ox and make their guess. Unless you’re guessing the number of jellybeans in a jar, this is almost never the case in actual decision scenarios.

Couzin and Kao say this information “patchiness” is the reason why accuracy tends to diminish as the crowd gets bigger. In most situations, there is commonly understood and known information, which the researchers refer to as “correlated information.” But there is also information that only some of the members of the group have, which is “uncorrelated information.” To make matters even more complex, the nature of uncorrelated information will be unique to each individual member. In real life, this would be our own experience, expertise and beliefs.  To use a technical term, the correlated information would be the “signal” and the uncorrelated information would be the “noise.” The irony here is that this noise is actually beneficial to the decision process.

In big groups, the collected “noise” gets so noisy it becomes difficult to manage and so it tends to get ignored. It drowns itself out. The collective focuses instead on the correlated information. In engineering terms this higher signal-to-noise ratio would seem to be ideal, but in decision-making, it turns out a certain amount of noise is a good thing. By focusing just on the commonly known information, the bigger crowd over-simplifies the situation.

Smaller groups, in contrast, tend to be more random in their make up. The differences in experiences, knowledge, beliefs and attitudes, even if not directly correlated to the question at hand, have a better chance of being preserved. They don’t get “averaged” out like they would in a bigger group. And this “noise” leads to better decisions if the situation involves imperfect information. Call it the averaging of intuition, or hunches. In a big group, the power of human intuition gets sacrificed in favor of the commonly knowable. But in a small group, it’s preserved.

In the world of corporate strategy, this has some interesting implications. Business decisions are almost always complex and involve imperfectly distributed information. This research seems to indicate that we should carefully consider our decision-making units. There is a wisdom of crowds benefit as long as the crowd doesn’t get too big. We need to find a balance where we have the advantage of different viewpoints and experiences, but this aggregate “noise” doesn’t become unmanageable.

Five Years Later – An Answer to Lance’s Question (kind of)

112309-woman-internetIt never ceases to amaze me how writing can take you down the most unexpected paths, if you let it. Over 5 years ago now, I wrote a post called “Chasing Digital Fluff – Who Cares about What’s Hot?” It was a rant, and it was aimed at marketer’s preoccupation with what the latest bright shiny object was. At the time, it was social. My point was that true loyalty needs stabilization in habits to emerge. If you’re constantly chasing the latest thing, your audience will be in a constant state of churn. You’d be practicing “drive-by” marketing. If you want to find stability, target what your audience finds useful.

This post caused my friend Lance Loveday to ask a very valid question…”What about entertainment?” Do we develop loyalty to things that are entertaining? So, I started with a series of posts on the Psychology of Entertainment. What types of things do we find entertaining? How do we react to stories, or humor, or violence? And how do audiences build around entertainment? As I explored the research on the topic, I came to the conclusion that entertainment is a by-product of several human needs – the need to bond socially, the need to be special, our appreciation for others whom we believe to be special, a quest for social status and artificially stimulated tweaks to our oldest instincts – to survive and to procreate. In other words, after a long and exhausting journey, I concluded that entertainment lives in our phenotype, not our genotype. Entertainment serves no direct evolutionary purpose, but it lives in the shadows of many things that do.

So, what does this mean for stability of an audience for entertainment? Here, there is good news, and bad news. The good news is that the raw elements of entertainment haven’t really changed that much in the last several thousand years. We can still be entertained by a story that the ancient Romans might have told. Shakespeare still plays well to a modern audience. Dickens is my favorite author and it’s been 144 years since his last novel was published. We haven’t lost our evolved tastes for the basic building blocks of entertainment. But, on the bad news side, we do have a pretty fickle history when it comes to the platforms we use to consume our entertainment.

This then introduces a conundrum for the marketer. Typically, our marketing channels are linked to platforms, not content. And technology has made this an increasingly difficult challenge. While we may connect to, and develop a loyalty for, specific entertainment content, it’s hard for marketers to know which platform we may consume that content on. Take Dickens for example. Even if you, the marketer, knows there’s a high likelihood that I may enjoy something by Dickens in the next year, you won’t know if I’ll read a book on my iPad, pick up an actual book or watch a movie on any one of several screens. I’m loyal to Dickens, but I’m agnostic as to which platform I use to connect with his work. As long as marketing is tied to entertainment channels, and not entertainment content, we are restricted to targeting our audience in an ad hoc and transitory manner. This is one reason why brands have rushed to use product placement and other types of embedded advertising, where the message is set free from the fickleness of platform delivery challenges. If you happen to be a fan of American Idol, you’re going to see the Coke and Ford brands displayed prominently whether you watch on TV, your laptop, your tablet or your smartphone.

It’s interesting to reflect on the evolution of electronic media advertising and how it’s come full circle in this one regard. In the beginning, brands sponsored specific shows. Advertising messages were embedded in the content. Soon, however, networks, which controlled the only consumption choice available, realized it was far more profitable to decouple advertising from the content and run it in freestanding blocks during breaks in their programming. This decoupling was fine as long as there was no fragmentation in the channels available to consume the content, but obviously this is no longer the case. We now watch TV on our schedule, at our convenience, through the device of our choice. Content has been decoupled from the platform, leaving the owners of those platforms scrambling to evolve their revenue models.

So – we’re back to the beginning. If we want to stabilize our audience to allow for longer-term relationship building, what are our options? Obviously, entertainment offers some significant challenges in this regard, due mainly to the fragmentation of platforms we use to consume that content. If we use usefulness as a measure, the main factor in determining loyalty is frequency and stability. If you provide a platform that becomes a habit, as Google has, then you’ll have a fairly stable audience. It won’t destabilize until there is a significant enough resetting of user expectations, forcing the audience to abandon habits (always very tough to do) and start searching for another useful tool that is a better match for the reset expectations. If this happens, you’ll be continually following your audience through multiple technology adoption curves. Still, it seems that usefulness offers a better shot at a stable audience than entertainment.

But there’s still one factor we haven’t explored – what part does social connection play? Obviously, this is a huge question that the revenue models of Facebook, Twitter, Snapchat and others will depend on. So, with entertainment and usefulness explored ad nauseum, in the series of posts, I’ll start tracking down the Psychology of Social connection.

Letting the Foxes into Journalism’s Hen(Hedgehog) House

First published March 27, 2014 in Mediapost’s Search Insider

fanhI am rooting for Nate Silver and fivethirtyeight.com, his latest attempt to introduce a little data-driven veracity into the murky and anecdotal world of journalism. But I may be one of the few, at least if we take the current backlash as a non-scientific, non-quantitative sample:

I have long been a fan of Nate Silver, but so far I don’t think this is working. – Tyler Cowen, Marginal Revolution

Nate Silver’s new venture may become yet another outlet for misinformation when it comes to the issue of human-caused climate change, Michael Mann, director of the Earth System Science Center at Pennsylvania State University.

Here’s hoping that Nate Silver and company up their game, soon. – Paul Krugman, NY Times

Krugman also states:

You can’t be an effective fox just by letting the data speak for itself — because it never does. You use data to inform your analysis, you let it tell you that your pet hypothesis is wrong, but data are never a substitute for hard thinking.

Now..Nate Silver doesn’t disagree with this. In fact, he says pretty much the same thing in his book, The Signal and the Noise:

The numbers have no way of speaking for themselves. We speak for them. We imbue them with meaning.

But he goes on,

Like Caesar, we may construe them in self-serving ways that are detached from their objective reality.

And it’s this construal that Silver is hoping to nip in the bud with FiveThirtyEight. In essence, he wants to do it by being a Fox, to borrow from Isaiah Berlin’s analogy.

‘The fox knows many things, but the hedgehog knows one big thing.’ We take a pluralistic approach and we hope to contribute to your understanding of the news in a variety of ways.

Silver thinks the media’s preoccupation with punditry is a dangerous thing. Pundits, whether they’re coming from the right or left, are Hedgehogs. They get paid for their expertise on “one big thing.” And the more controversial their stand, the more attention they get. This can lead to a dangerous spiral, as researcher Philip Tetlock found out:

What experts think matters far less than how they think. If we want realistic odds on what will happen next, coupled to a willingness to admit mistakes, we are better off turning to experts who embody the intellectual traits of Isaiah Berlin’s prototypical fox—those who “know many little things,” draw from an eclectic array of traditions, and accept ambiguity and contradiction as inevitable features of life.

Tetlock was researching how expertise correlated with the ability to make good predictions. What he found was that it was actually an inverse relationship. The higher the degree of expertise, the more likely the person in question was a hedgehog. Media pundits are usually extreme versions of hedgehogs, which not only have one worldview, but also love to talk about it. Nate Silver believes that to get an objective view of world events, you need to be a fox, first, but secondly; you should be a fox that’s good at sifting through data:

Conventional news organizations on the whole are lacking in data journalism skills, in my view. Some of this is a matter of self-selection. Students who enter college with the intent to major in journalism or communications have above-average test scores in reading and writing, but below-average scores in mathematics.

So, all this makes sense. The problem in Silver’s approach is that journalism is the way it is because that’s the way humans want it. While I applaud Silver’s determination to change it, he may be trying to push water up hill. Pundits exist not just because the media keeps pushing them in front of us – they exist because we keep listening. Humans like opinions and anecdotes. We’re not hardwired to process data and objectively rationalize. We connect with stories and we’re drawn to decisive opinion leaders. Silver will have to find some middle ground here, and that seems to be where the problems arise. The minute writers add commentary to data; they have to impose an ideological viewpoint. It’s impossible not to. And when you do that, you introduce a degree of abstraction.

The backlash against Fivethirtyeight.com generally falls into two camps: Foxes like Silver that have no problem with the approach but disagree with the specific data put forward and Hedgehogs that just don’t like the entire concept. The first camp may come onside as Silver and his team work out the inevitable hiccups in their approach. The second, which, it should be noted, have a large number of pundits in their midst, will never become fans of Silver and his foxlike approach.

In the end though, it really doesn’t matter what columnists and journalists think. It’s up to the consumers of news media. We’ll decide what we like better – hedgehogs or foxes.

The Bug in Google’s Flu Trend Data

First published March 20, 2014 in Mediapost’s Search Insider

Last year, Google Flu Trends blew it. Even Google admitted it. It over predicted the occurrence of flu by a factor of almost 2:1.  Which is a good thing for the health care system, because if Google’s predictions had have been right, we would have had the worst flu season in 10 years.

Here’s how Google Flu Trends works. It monitors a set of approximately 50 million flu related terms for query volume. It then compares this against data collected from health care providers where Influenza-like Illnesses (ILI) are mentioned during a doctor’s visit. Since the tracking service was first introduced there has been a remarkably close correlation between the two, with Google’s predictions typically coming within 1 to 2 percent of the number of doctor’s visits where the flu bug is actually mentioned. The advantage of Google Flu Trends is that it is available about 2 weeks prior to the ILI data, giving a much needed head start for responsiveness during the height of flu season.

FluBut last year, Google’s estimates overshot actual ILI data by a substantial margin, effectively doubling the size of the predicted flu season.

Correlation is not Causation

This highlights a typical trap with big data – we tend to start following the numbers without remembering what is generating the numbers. Google measures what’s on people’s minds. ILI data measures what people are actually going to the doctor about. The two are highly correlated, but one doesn’t not necessarily cause the other. In 2013, for instance, Google speculated that increased media coverage might be the cause for the overinflated predictions. More news coverage would have spiked interest, but not actual occurrences of the flu.

Allowing for the Human Variable

In the case of Google Flu Trends, because it’s using a human behavior as a signal – in this case online searching for information – it’s particularly susceptible to network effects and information cascades. The problem with this is that these social signals are difficult to rope into an algorithm. Once they reach a tipping point, they can break out on their own with no sign of a rational foundation. Because Google tracks the human generated network effect data and not the underlying foundational data, it is vulnerable to these weird variables in human behavior.

Predicting the Unexpected

A recent article in Scientific American pointed out another issue with an over reliance on data models –  Google Flu Trends completely missed the non-seasonal H1N1 pandemic in 2009. Why? Algorithmically, Google wasn’t expecting it. In trying to eliminate noise from the model, they actually eliminated signal coming during an unexpected time. Models don’t do very well at predicting the unexpected.

Big Data Hubris

The author of the Scientific American piece, associate editor Larry Greenemeier, nailed another common symptom of our emerging crush on data analytics – big data hubris. We somehow think the quantitative black box will eliminate the need for more mundane data collection – say – actually tracking doctor’s visits for the flu. As I mentioned before, the biggest problem with this is that the more we rely on data, which often takes the form of arm’s length correlated data, the further we get from exploring causality. We start focusing on “what” and forget to ask “why.”

We should absolutely use all the data we have available. The fact is, Google Flu Trends is a very valuable tool for health care management. It provides a lot of answers to very pertinent questions. We just have to remember that it’s not the only answer.

The Psychology of Usefulness – Part Five: A Recap

0824_lifestyle_luddite_630x420In the past five posts, I’ve been looking at how we choose to accept new technologies. As part of that, we’ve had a fairly exhaustive review of the various versions of the Technology Acceptance Models proposed by Fred Davis, Richard Bagozzi and, most prolifically, Viswanath Venkatesh.

Before forging ahead, I’d like to provide a brief recap of primary thoughts behind the models.

In the first post, I explored the different between autotelic and exotelic activities. The first we do for the sheer enjoyment of the activity itself – our reward is inherent in the doing. Exotelic activities are the things we do because we have to. There is little to no reward in them. Generally, when we’re judging usefulness, it’s to complete an exotelic activity. In judging usefulness, the emotion most commonly invoked is an aversion to risk – so it carries a negative emotional valency, although a relatively mild one – typically invoking anxiety or concern rather than outright fear or dread. The degree of emotional valency is generally quite low – it’s more a calculation of the resources required vs the usefulness expected.

Next, in the second post, I explained why I believe that our judgement of usefulness is based on a fairly heuristic calculation of the brain. It’s similar to the same mechanisms we use when foraging for food. Because of that belief, I’ve borrowed heavily from the previous work done by Pirolli and Card on Information Foraging and also Eric Charnov’s work on Optimal Foraging and his Marginal Value Theorem.

Because there’s little emotional engagement, we also tend to make useful resources habits if the frequency is high enough.This is the ground I covered in posts Three and Four. First, using Google as an example, I looked at how habits are created and maintained. Then, in the next post, I looked at the factors that might disrupt a habit, forcing us to look for a viable alternative. The more the brain has to be involved in judging usefulness – the less loyal we tend to be.

Also, we will only seek new technologies if: a) our current technology no longer meets our expectations, which are often reset because; b) we’ve become aware of a new, superior technology.

Then, we have to decide whether or not to accept a new technology. There have been several attempts to create a model that can predict the acceptance of a new technology. Most relied on the same foundational assumptions:

  • Some mix of internal and external motivators will result in the creation of an attitude.
  • Depending on the valence of the attitude (either negative or positive) we may form an intention to use the technology.
  • Once this intention is formed, it leads to usage.

All the modifications to the model (5 revisions at last count) focused in the first two of these three assumptions, offering alternatives for the motivators that create the attitude. Some versions removed the attitude step completely and moved directly to intention. But none of them changed the assumed progression from intention to usage.

The useful parts of these models that I wanted to carry forward are:

  • Intentions are formed by a heuristic balancing of negative and positive factors in the adopter’s mind, often labeled Perceived Usefulness and Perceived Ease of Use.
  • External factors, such as the opinions of others, impact our decisions to adopt a technology.
  • The cognitive process involved roughly corresponds to a Bayesian analysis, where we set a “prior” – our original attitude, and update it based on new information gathered through the decision process.

The potentially flawed assumptions I would like to leave behind are:

  • The process is typically a linear one, moving from the left of the model (attitude) to the right of the model (usage).
  • There are no mediating factors between the intention and usage boxes in any of the models.

In 2007, one of the original authors of the first TAM, Bagozzi, said it was time for a paradigm shift in the thinking about technology acceptance. He brought in an entirely new context in which to think about the acceptance of technology – the striving for and achievement of goals. This created a more holistic view of the decision process, where the acceptance of technology wasn’t artificially isolated, but was part of a much broader frame where that acceptance was contingent on a hierarchy of goals and sub goals. What I particularly liked was the addition of “desire” as a step, and also the introduction of self regulation as a mediating factor. Bagozzi was the first to indicate that the process was possibly more recursive, an iterative cycle rather than a linear path.

Bagozzi’s inclusion of goal setting and achievement builds a context for adoption. This aligns with Everett Roger’s extensive work in innovation adoption, in which he said,

An important factor regarding the adoption rate of an innovation is its compatibility with the values, beliefs, and past experiences of individuals in the social system.

While the acceptance of a technology may be a personal decision, it is almost always set within a broader social context. All the versions   of Technology Acceptance Models I looked at included some type of social mediation in the acceptance process. But it was more of a factor in the creation of an initial attitude and a mediating factor in the progression from attitude to intention to behavior. In other words, if the acceptance of a technology made you socially unpopular, you would probably change your mind and reject it.

But when we choose to achieve a goal, there is a cognitive process that happens which creates a framework for acceptance. The goal becomes the primary evaluative topic and the technology generally becomes secondary to it. Bagozzi recognized that the two are interlinked and have to be evaluated together. We choose a goal, divide this into sub-goals and then seek how to execute against these goals.

Let’s use a personal example to see how goals and technology acceptance are intrinsically linked. Let’s say our goal is to get healthier. This breaks down into several sub-goals: Beginning a regular exercise routine, eating better, losing weight, drinking less, etc. Each of these then can be further divided into more specific goals. Let’s take eating better. It could involve tracking our calories, paying more attention to nutrition labels, including more fresh fruits and vegetables in our diet, cutting out sugar and avoiding processed foods. At this point, we may decide to use a tool like Livestrong‘s MyPlate Calorie tracker. If you were to use one of the various versions of TAM to predict the acceptance of this new technology, you would artificially divorce the act of acceptance from the broader goal hierarchy that precedes it. According to TAM, your acceptance of MyPlate would be determined by your evaluation of the ease of use and the expected usefulness. While undoubtedly important, these two factors are completely dependent on the mental scaffolding you’ve built around the idea of getting healthier. There are a myriad of factors that live beyond these that would have some impact on your eventual acceptance or rejection of the technology in question. For example, perhaps you decide that calorie counting is not the best path to eating better and so any tool that counts calories gets rejected out of hand. Or, perhaps you fall off the wagon with your eating plan and reject the tool not because it’s not useful, or easy to use, but simply because counting calories constantly reminds you how weak your will power is.

So, with the past posts recapped, next post we’ll forge forward with a proposed new Technology Acceptance Model.

The Psychology of Usefulness – The Acceptance of Technology – Part Four

After Venkatesh and Davis released the TAM 2 model, Venkatesh further expanded the variables that went into our calculation of Perceived Usefulness in TAM 3:

TAM3

Venkatesh, V. and Bala, H. “TAM 3: Advancing the Technology Acceptance Model with a Focus on Interventions,”

Venkatesh divided the determinants of Perceived Ease of Use into two categories: Anchor determinants and Adjustment determinants. Anchor determinants where the user’s baseline and came from their general beliefs about computer and usage. In Bayesian terms, this would be the user’s “prior.” It would create the foundational attitude towards the technology in question.

Anchor determinants included:

Computer self efficacy – How proficient the user is with the current technology paradigm (i.e. how comfortable they are with computers)

Perception of External Control – How much organizational support there is for the system to be accepted?

Computer Anxiety – Is there apprehension or fear involved with using a computer?

Computer Playfulness – Spontaneity in computer interactions.

Then, Venkatesh added the Adjustment determinants. These factors come from direct experience with the technology in question and are used to “adjust” the user’s attitude towards the technology. Again, this is a very Bayesian cognitive process.

Adjusting Determinants included:

Perceived Enjoyment – Is using the system enjoyable?

Objective Usability – Is the effort actually required what it was perceived to be (resulting in either positive or negative reinforcement)

Over time some of the Anchor factors will diminish in importance (Playfulness, Anxiety) and adjustments will become stronger.

So now, in TAM 3, we have essentially the same process of acceptance, but with much more granularity in the definition of the determinants that go into the Perceptions of Ease of Use. However, with the division of determinants into the categories of “Anchor” and “Adjustment” Venkatesh starts to hit at the iterative nature of this process of acceptance. We create a baseline belief or attitude and then this becomes updated, either though external forces (the Subjective Norm) or our own internal experiences (the Adjusting Determinants). While the model indicates a linear decision path it’s now appearing likely that it’s a more recursive path.

As an interesting aside, Brown, Massey, Motoya-Weiss and Burkman (2002) said that they found variance in the importance of Perceived Ease of Use. Remember, in the original TAM model, Davis indicated that while Perceived Ease of Use does have impact over the original attitude towards a technology, he found that Perceived Usefulness is a more powerful indicator of usage intention. Brown et al found that this varies depending on whether the acceptance of a technology is mandatory or voluntary. If acceptance is mandatory, they found that Perceived Ease of Use may actually have a more important impact on system acceptance.

After TAM 3, there was a further attempt to round up the competing theories that all contributed to the evolution of TAM. This was the hopefully named Unified Theorty of Acceptance and Use of Technology (or UTAUT) – proposed in 2003 by Venkatesh, Morris, Davis and Davis:

UTAUT

Venkatesh, V., Morris, M.G., Davis, F.D., and Davis, G.B. “User Acceptance of Information Technology: Toward a Unified View,” MIS Quarterly, 27, 2003, 425-478

So, what began as an attempt to simplify our understanding of the acceptance of technology became a rather unwieldy beast. Bagozzi, who worked with Davis on the original TAM model, finally had step back in and comment:

The exposition of UTAUT is a well-meaning and thoughtful presentation. But in the end we are left with a model with 41 independent variables for predicting intentions and at least eight independent variables for predicting behavior. Even here, arguments can be made that important independent variables have been left out, because few of the included predictors are fundamental, generic or universal and future research is likely to uncover new predictors not subsumable under the existing predictors. The IS field risks being overwhelmed, confused and misled but the growing piecemeal evidence behind decision making and action in regard to technology adoption/acceptance/rejection.

Bu the biggest criticism of Venkatesh’s models from Bagozzi (2007)  pointed out the same fundamental flaw that I mentioned – the assumption that intention leads to usage. Here are Bagozzi’s main concerns with the evolution of TAM:

Is Behavior the End Goal? – Bagozzi says:

The models…fail to consider that many actions are taken not so much as ends in and of themselves but rather as means to more fundamental ends or goals.

TAM ends at behavior. Actually, it ends at intent, as it assumes that intent always leads to behavior, but we’ll come back here in a moment. Bagozzi’s point is that behavior is dependent on a broader context, and there could be an end-goal that will impact acceptance that is completely ignored in the model. For example, lets say that the specific technology to be accepted is a tool to analyze conversion data in online campaigns. The end goal is to improve ROI from all online campaigns. But, to reach this end goal, there are a number of contributory goals, including, but not limited to:

  • More efficient budget allocations
  • Improving Landing Page performance
  • Better tracking of performance data
  • Improve click throughs on online ads
  • Improve the online conversion path

The tool in question is a subset of the third item in the list. Because the decision to accept this tool is contingent on meeting a number of broader goals, it obviously becomes a critical factor in that acceptance. But Bagozzi’s point is the only place this is accounted for, presumably, is in the anticipated belief upstream in the model. Once again, TAM falls victim to its own quest for parsimony. Bagozzi argues a better approach is to understand that this is a process, and as such will include goal striving. And that is a recursive process:

In goal striving, intention formation is succeeded by planning (e.g, when, where and how to act instrumentally), overcoming obstacles, resisting temptations, monitoring progress to goal achievement, readjusting actions, maintaining effort and willpower, and reassessing and even changing goals and means. These processes fill the gaps between intention and behavior and between behavior and goal attainment and are crucial for the successful adoption and use of technology.

Finally! An acknowledgement that it’s not a straight line from intention to behavior!

The Gap between Attitude and Intention – First of all, Bagozzi disagreed with the elimination of Attitude as a preliminary step to Intention. Further, he says that even with Attitude back in place, there may be very compelling reasons why a person may agree that the technology is acceptable, but still choice not to accept it. To fill in the gap, he proposes borrowing from the Belief-Desire model. Here, even if the Belief is in place, there also needs to be Desire before the intention to act is formed. So, to recap, a User may have all the right determinants in place to decide that the technology in question is perceived to be useful (PU) and that it will be sufficiently easy to learn to use (PEU) but still not have any desire to accept the technology. Perhaps the decision to accept was made by her boss, who is an asshole and she’s resisting on principle.

In Bagozzi’s paper, he goes on at some length to outline the limitations in any of the proposed TAM models. In addition to the above points, he also suggests that things like the group, cultural and social aspects of technology acceptance are not adequately dealt with in the model, as well as emotions, self-regulation and other mediators that are common in our pursuit of goals. In short, he says that technology acceptance is too reliant on context to lend itself to general models and the decision path itself is much more complex than the models would indicate. He advocates a new foundation, based on his work of goal setting and striving:

37_10_1108_S1548-6435_2011_0000008005

Bagozzi 2007

This lays out the decision making core process, not specific to technology acceptance, but applicable to the striving towards any goal. But, in this model, there is the opportunity to “plug in” factors specific to the acceptance of a technology within a specific context. For example, inputs into Goal Desire could include any number of things: first of all, the goal itself (taking into account the entire goal hierarchy that leads to the focal goal – the acceptance of a specific technology), anticipated and anticipatory emotions, relative advantage, job fit, attitudes toward success, failure, outcome expectancies, and, of course, Perceived Usefulness and Perceived Ease of Use. The exact balance will be contingent on circumstance.

The arrow leading up to Action Desire represents mediating factors such as group norms, subjective norms, social identity, effort expectancy and attitudes towards an act.

A key addition to Bagozzi’s proposed model  is self regulation. Somewhere between desire and intention, humans have the ability to reflect on their desires and decide if they are comfortable intending to act on that desire. Specifically, in the case of technology adoption, we have to decide if the behaviors we would undertake would sit well with our moral and belief framework. Let’s say, for instance, that the technology adoption being considered would increase efficiency dramatically, allowing the company to decrease head count. You may be aware of the human cost of the decision to adopt and this may cause you to regulate your desire (increased efficiency) against it.

Below is an expanded version of Bagozzi’s model with all the inputs shown.

decisioncore

Bagozzi’s proposed model with inputs shown (Bagozzi 2007)

What is interesting in Bagozzi’s model is the chain of decision making which separates Goals desire from Behavioral desire. This is a further exploration of the transition from Attitude to Action, which was so deterministic in Venkatesh’s models. Bagozzi sets it in what, to me, is a more palatable framework. We have goals, which likely include broad goals and sub goals in some type of hierarchy. Our desire to reach these goals then have to be translated into intentions, where the actual execution required is begun to be planned out. This helps us understand the required behaviors, which then leads to behavioral desires. These desires then get translated into intentions. But throughout this chain, there are a number of mediating factors, both internal and external, that can cause reflection, resetting, outright abandonment or modification of both desires and intentions. There is no single arrow pointing to the right. There is, instead, an iterative process that allows for looping back to any one of the previous stages.

This post has become much longer and much more technical than I originally intended, so I think we’ll wrap this up and start fresh next time with a recap of the various models of Technology Acceptance, and my attempt to build a model that allows for iterative reflection and adjustment.

The Psychology of Usefulness: The Acceptance of Technology – Part Three

In Part Two of this series, I looked at Davis and Bagozzi’s Technology Acceptance Model, first proposed in 1989.

Technology_Acceptance_Model

As I said, while the model was elegant and parsimonious, it seems to simplify the realities of technology acceptance decisions too much. In 2000, Venkatesh and Davis tried to deal with this in TAM 2 – the second version of the Technology Acceptance Model.

TAM2

In this version, they added several determinants of Perceived Usefulness and demoted Perceived Ease of Use to being just one of the factors that impacted Perceived Usefulness.  Impacting this mental calculation were two mediating factors: Experience and Voluntariness. This rebalancing of factors provides some interesting insights into the mental process we go through when making a decision whether we’ll accept a new technology or not.

Let’s begin with the determinants of Perceived Usefulness in the order they appear in Venkatesh and Davis’s model:

Subjective Norm: TAM 2 resurrects one of the key components of the original Theory of Reasoned Action model – the opinions of others in your social environment.

Image: Venkatesh and Davis also included another social factor in their list of determinants – how would the acceptance of this technology impact your status in your social network? Notice that our calculation of the image enhancement potential has the Subjective Norm as an input. It’s a Bayesian prediction – we start with our perceived social image status (the prior) and adjust it based on new information, in this case the acceptance of a new technology.

Job Relevance: How applicable is the technology to the job you have to do?

Output Quality: How will this technology impact your ability to perform your job well?

Result Demonstrability: How easy is it to show the benefits of accepting the technology?

It’s interesting to note how these factors split: the first two (subjective norm and image) being related to social networks, the next two (Job Relevance and Output Quality) being part of a mental calculation of benefit and the last one, Demonstrability, bridging the two categories: How easy will it be to show others that I made the right decision?

According to the TAM 2 model, we use these factors, which combine practical task performance considerations and social status aspirations, into a rough calculation of the perceived usefulness of a technology. After this is done, we start balancing that with how easy we perceive the new technology to be to use. Venkatesh and Davis commented on this and felt that Perceived Ease of Use has a variable influence in two areas, the forming of an attitude towards the technology and a behavioral intention to use the technology. The first is pretty straight forward. Our attitude is our mental frame regarding the technology. Again, to use a Bayesian term, it’s our prior. If the attitude is positive, it’s very probably that we’ll form a behavioral intention to use the technology. But there are a few mediating factors at this point, so let’s take a closer look at the creation of Behavioral Intention..

In forming our intention, Perceived Ease of Use is just one of the determinants we use in our “Usefulness” calculation, according to the model. And it depends on a few things. It depends on efficacy – how comfortable we judge ourselves to be with the technology in question. It also depends on what resources we feel we will have access to to help us up the learning curve. But, in the forming of our attitude (and thereby our intention), Venkatesh and Davis felt that Perceived Usefulness will typically be more important than Perceived Ease of Use. If we feel a technology will bring a big enough reward, we will be willing to put up with a significant degree of pain. At least, we will in what we intend to do. It’s like making a New Year’s Resolution to lose weight. At the time we form the intention, the pain involved is sometime in the future, so we go forward with the best of intentions.

As we move forward from Attitude to Intention, this transition if further mediated in the model by our subjective norm – the cognitive context we place the decision in. Into this subjective norm falls our experience (our own evaluation of our efficacy), the attitudes of others towards the technology and also the “Voluntariness” of the acceptance. Obviously, our intention to use will be stronger if it’s a non-negotiable corporate mandate, as opposed to a low priority choice we have the latitude to make.

What is missing from the TAM 2 model is the link between Perceived Ease of Use and actual Usage. Just like a New Year’s Resolution, intentions don’t always become actions. Venkatesh and Davis said Perceived Ease of Use is a moving, iteratively updated calculation. As we gain hands-on experience, we update our original estimate of Ease of Use, either positively or negatively. If it’s positive, it’s more likely that Intention will become Usage. If negatively, the technology may fail to become accepted. In fact, I would say this feedback loop is an ongoing process that may repeat several times in the space between Intention and Usage. The model, with a single arrow going in one direction from Intention to Usage, belies the complexity of what is happening here.

Venkatesh and Davis wanted to create a more realistic model, expanding the front end of the model to account for determinants going into the creation of Intention. They also wanted to provide a model of the decision process that better represented how we balance Perceived Usefulness and Perceived Ease of Use. I think they made some significant gains here. But the model is still a linear one – going in one direction only. What they missed is the iterative nature of acceptance decisions, especially in the gap between Intention and Behavior.

In Part Four, we’ll look at TAM 3 and see how Venkatesh further modified his model to bring it closer to the real world.

So, Six Seconds is the Secret, Huh?

First published February 13, 2014 in Mediapost’s Search Insider

oreo-superbowl-blackout-adApparently, the new official time limit for customer engagement is 6 seconds, according to a recent post on Real Time Marketing. How did we come up with 6? Well, in the world of social media engagement it seemed like a good number and no one has called bull shit on it yet, so 6 it is

Marketers love to talk about time – just in time, real time, right time. At the root of all this “time talk” is the realization that customers really don’t have any time for us, so we have to somehow jam our messages into the tiny little cracks that may appear in the wall of willful ignorance they carefully build against marketing. The marketer’s goal is to erode their defenses by looking for any weakness that may appear.

Look at the supposed poster child for Real Time Marketing – the Oreo coup staged during the black out in the 2013 Super Bowl. Because the messaging was surprising and clever, and because, let’s face it, we weren’t doing much of anything else anyway, Oreo managed to gain a foothold in our collective consciousness for a few precious seconds. So, marketers being marketers, we all stumbled over ourselves to proclaim a new channel and launch a series of new micro-attacks on consumers. That’s where the 6 seconds came from. Apparently, that’s the secret to storming the walls. Five seconds and you’re golden. Seven seconds and you’re dead.

Oreo surprised us, and it wasn’t because the message was 6 seconds long. It was because we weren’t expecting a highly relevant, highly timely message. Humans are built to respond to things that don’t fit within our expected patterns. The whole approach of marketing is to constantly blanket us with untimely, irrelevant messages. Marketers, to be fair, try to deliver the right message at the right time to the right person, but it’s really hard to do that. So, we overcompensate by delivering lots of messages all the time to everyone, hoping to get lucky. Not to take anything away from the cleverness and nimbleness of the Oreo campaign, but they got lucky. We were surprised and we let our defenses down long enough to be amused and entertained. Real time marketing wasn’t a brilliant new channel; it was a shot in the dark – literally.

And there’s no six-second gold standard of engagement. If you can deliver the right message at the right time to the right person, you can spend hours talking to your prospective customer.  It’s only when you’re trying to interrupt someone with something irrelevant that you have to hopefully shoehorn it into their consciousness. Think of it like a Maslow’s hierarchy of advertising effectiveness.  At it’s best advertising should be useful. This sits at the top of the pyramid. After usefulness comes relevance – even if I don’t find the ad useful to me right now, at least you’re talking to the right person. After relevance comes entertainment – I’ll willingly give you a few seconds of my time if I find your message amusing or emotionally engaging.  I may not buy, but I’ll spend some time with you. After entertainment comes the category the majority of advertising falls into – a total waste of my time.  Not useful, irrelevant, not emotionally engaging. And making an ad that falls into this category 5 seconds long, no matter what channel it’s delivered through, won’t change that. You may fool me once, but next time, I’m still going to ignore you.

There was something important happening during the Oreo campaign at the 2013 Super Bowl, but it had nothing to do with some new magic formula, some recently discovered loophole in our cognitive defenses. It was a sign of what may, hopefully, emerge as trend in advertising – nimble, responsive marketing that establishes a true feedback loop with prospects. What may have happened when the lights went out in New Orleans is that we may have found a new, very potent way to make sense of our market and establish a truly interactive, responsive dialogue with them. If this is the case, we may have just found a way climb a rung or two on the Advertising Effectiveness Hierarchy.

The Psychology of Usefulness: The Acceptance of Technology – Part Two

In my last post, I talked about how the Theory of Reasoned Action and the original Technology Acceptance Model tried to predict both intention and usage of new technologies. As a quick recap, let’s look again at Davis and Bagozzi’s original model.

Technology_Acceptance_Model

In aiming for the simplest model possible, there was significant conflation applied to the front end of the model – with just one box representing external variables, which then led to two similarly conflated boxes: Perceived Usefulness and Perceived Ease of Use. While this simplification was admirable in the quest for parsimony, in real world situations it seemed like it went too far in this direction. There was a lot happening between the three boxes at the front of the model that demanded closer examination.

Davis indicated that there was an interesting relationship between Perceived Usefulness and Perceived Ease of Use. One of the mechanisms at play that has to be understood is self-efficacy. In understanding adoption of technology, self-efficacy is a key factor. Essential, it means that the easier a system is to use, the greater the user’s sense of efficacy. They believe they have control over what they are doing. And control, especially on a work context, is a strong motivational driver. There is an extensive body of work exploring the psychological importance of control. If we feel we’re in control, we also feel empowered to mitigate risk. The concept of self-efficacy helps to highlight the importance of the Perceived Ease of Use box. But what about the other box: Perceived Usefulness?

Davis, in his accompanying notes and research, indicated that Perceived Usefulness is a stronger indicator of intention than Perceived Ease of Use. In other words, we are willing to put up with some pain to learn a new technology if we feel it will offer a significant improvement in our ability to complete a task. This balancing equation requires two heuristic evaluations on the part of the user: the allocation of cognitive resources required to gain proficiency and the expected usefulness of the tool once proficiency is gained. This is exactly the same equation used in Charnov’s Marginal Value Theorem, applied in a different context. In optimal foraging, we (and all animals who forage) balance expenditure of resources required to reach a food patch against the expected food value to be derived from that patch. In technology adoption, we balance expenditure of resources required to master a new technology against the increased usefulness that technology offers.

In this heuristic evaluation, there are four key marketing lessons for anyone who’s business model relies on the adoption of new technology:

1)   Lessen the intimidation of the learning curve. Persuade the user (and this is a key point that I’ll return to in in point 4) that this is a reasonable investment of resources. Build a sense of perceived ease of use. Provide visible links to intuitive learning resources. Often, marketers overplay the feature benefits of their products to show how powerful they are. But, as they’re doing this, they fail to realize that this upsets the balance between perceived usefulness and perceived ease of use.

2)   Provide clear examples of perceived usefulness in terms that are immediately relevant. Remember, this is the key factor in the equation the prospect is trying to balance. The more salient you can make the perceived usefulness, the more likely the user is to adopt it, even if a learning curve is present. Ideally, get that usefulness across with very specific, industry relevant examples that allow the user to visualize usage of the technology.

3)   Remember that the user is balancing the two factors. Ease of use is great, but it can’t come at the expense of overall usefulness. In fact, in calculating the right balance (which should be done with extensive testing feedback from target customers) it should offer a significant gain in usefulness (as measured against any incumbent technologies) with a relatively manageable investment of resources.

4)   Remember that you’re talking to a user. When trying to strike the right balance, remember that you’ll probably be talking to different people as the decision progresses. For the user, the right balance between perceived usefulness and perceived ease of use must be struck. But at some point, you’ll be talking to a buyer, not a user, before the sale actually is closed. This would be one of those external variables that fall outside the scope of the Technology Adoption Model. This switching of roles from “doers” to “buyers” is dealt with extensively in my book, The BuyerSphere Project.

In the next post, I’ll talk about how the Technology Acceptance Model has been modified over the past 2 decades so it better reflects real world decision making.

The Psychology of Usefulness: The Acceptance of Technology – Part One

oldpeopletech7_317161In the last post, I talked about what it takes to break a habit built around an online tool, website or application. In today’s post, I want to talk about what happens when we decide to replace that functional aid, whatever it might be.

So, as I said last time, the biggest factor contributing to the breakdown of habit is the resetting of our expectation of what is an acceptable outcome. If our current tools no longer meet this expectation, then we start shopping for a new alternative. In marketing terms, this would be the triggering of need.

Now, this breakdown of expectation can play out in one of two ways. First, if we’re not aware of an alternative solution, we may just feel an accumulation of frustration and dissatisfaction with our current tools. This build up of frustration can create a foundation for further “usefulness foraging” but generally isn’t enough by itself to trigger action. This lends support to my hypothesis that we’re borrowing the evolved Marginal Value algorithm to help us judge the usefulness of our current tools. To put it in biological terms we’re more familiar with, “A bird in the hand is worth two in the bush.” You don’t leave a food patch unless: A) you are reasonably sure there’s another, more promising, patch that can be reached with acceptable effort or B) you have completely exhausted the food available in the patch you’re in. I believe the same is true for usefulness. We don’t throw out what we have until we either know there’s an acceptable alternative that promises a worthwhile increase in usefulness or our current tool is completely useless. Until then, we put up with the frustration.

The Technology Acceptance Model

Let’s say that we have decided that it’s worth the effort to find an alternative. What are the mechanisms we use to find the best alternative? Fred Davis and Richard Bagozzi tackled that question in 1989 and came up with the first version of their Technology Acceptance Model. It took the Theory of Reasoned Action, developed by Martin Fishbein and Icek Ajzen, put forward a decade earlier (1975, 1980) and tried to apply it to the adoption of a new technology. They also relied on the work Everett Rogers did in the diffusion of technology.

First of all, like all models, the TAM had to make some assumptions to simplify real world decisions down to a theoretical model. And, in doing so, it has required a number of revisions to try to bring it closer to what technology adoption decisions look like in the real world.

Let’s start with the foundation of the Theory of Reasoned Action. In it’s simplest form, the TRA says that voluntary behavior is predicted by an individual’s attitude towards that behavior and how they think others would think of them if they performed that behavior.

TRA

So, let’s take the theory for a test drive – if you believe that exercising will increase your health and you also believe that others in your social circle will applaud you for exercising, you’ll exercise. With this example, I think you begin to see where the original TRA may run into problems. Even with the best of intentions, we may not actually make it to the gym. Fishbein and Ajzen’s goal was to create an elegant, parsimonious model that would reliably predict both behaviors and intentions, creating a distinction between the two. Were they successful?

In a meta-analysis of TRA, Sheppard et al (1988) found that attitude was a fairly accurate predictor of intention. If you believe going to the gym is a good thing, you will probably intend to go to the gym. The model didn’t do quite as good a job in predicting behavior. Even if you did intend to go to the gym, would you actually go?

The successful progression from intention to behavior seemed to be reliant on several real world factors, including the time between intention and action (the longer the time interval, the more the degree of erosion of intention) and also lack of control. For example, in the gym example, what if your gym suddenly increased it’s membership fees, or a sudden snowstorm made it difficult to drive there.

Also, if you were choosing from a set of clear alternatives and had to choose one, TRA did a pretty good job of predicting behaviors. But if alternatives were undetermined, or there were other variables to consider, then the predictive accuracy of TRA dropped significantly.

Let me offer an example of how TRA might not work very well in a real world setting. In my book, The BuyerSphere Project, I spent a lot of time looking at the decision process in B2B buying scenarios. If we used the TRA model, we could say that if a buyer had to choose between 4 different software programs for their company, we could use their attitudes towards each of the respective programs as well as the aggregated (and weighted  – because not every opinion should carry the same weight) attitudes towards these programs of the buyer’s co-workers, peers and bosses to determine their intention. And once we have their intention, that should lead to behavior.

But in this scenario, let’s look at some of the simplifying assumptions we’ve had to make to try to cram a real world scenario into the Fishbein Azjen model:

  • We assume a purchase will have to be made from one of the four alternatives. In a real world situation, the company may well decide to stick with what they have
  • We assume the four choices will remain static and we won’t get a new candidate out of left field
  • We assume that attitudes towards each of the alternatives will remain static through the behavioral interval and won’t change. This almost never happens in B2B buying scenarios
  • We assume the buyer – or rational agent – will be in full control of their behaviors and the ultimate decision. Again, this is rarely the case in B2B buying decisions.
  • We assume that there won’t be some mitigating factor that arises in between intention and behavior – for example a spending freeze or a change in requirements.

As you can see, in trying to create a parsimonious model, Fishbein and Azjen ran into a common trap – they had to simplify to the point where it failed to work consistently in the real world.

But, in this review by Alice Darnell, she pointed out Sheppard’s main criticism of the TRA model:

Sheppard et al. (1988) also addressed the model’s main limitation, which is that it fails to account for behavioural outcomes which are only partly under the individual’s volitional control.

I’ve added bolding to the word volitional on purpose. I’ve highlighted many external factors that may lie beyond the volitional control of the individual, but I think the biggest limitation of the TRA lies in its name: Theory of Reasoned Action. It assumes that reason drives our intentions and behaviors. It doesn’t account for emotion.

Applying Reasoned Action to Technology Acceptance

Now, let’s see how Rogers and Bagozzi took Fishbein and Azien’s foundational work and applied it to the acceptance of new technologies.

In their first model (1989) they took attitudes and subjective norms (the attitudes of others) and adapted them for a more applied activity, the use of a new technological tool. They came up with two attitude drivers: Perceived Usefulness and Perceived Ease of Use. If you think back to Charnov’s Marginal Value Theorem, this is exactly the same risk/reward mechanism at work here.  In foraging, it would be yield of food over perceived required effort. In Technology Acceptance, Perceived Usefulness is the reward and Perceived Ease of Use is the risk to be calculated. In the mental calculation, Rogers and Bagozzi assume the user would do a quick mental calculation, using their own knowledge and the knowledge of others to come up with a Usefulness/Ease value that would create their attitude towards using.  This then becomes their Behavioral Intention to Use – which should lead to Actual System Use.

tam

The TAM model was clean and parsimonious. There was just one problem. It didn’t do a very good job of predicting usage in real world situations. There seemed to be much more at work here in actual decisions to accept technologies. In the next post, we’ll look at how the TAM model was modified to bring it closer to real behaviors.