Category Archives: biases

When governments lead us astray

Lotteries are also known as “stupidity tax”; a nod to their improbable odds. In India, lotteries are often run by state governments – its an easy way to cover for their budget deficits. What these governments don’t realize is that they are fueling an addiction.

But what are the reasons behind this addiction? In the article, I talk about the behavioral science of lotteries.

Lotteries generate many ‘near misses’ thus making people believe that she is a winner even when she has lost, thus inducing a a dopamine fueled craving. I also talk about the incorrect application of ‘regression to the mean’ mental model, and how governments make it easy for someone to rationalize their lottery addiction.

Read more on livemint site here.

Rethinking Behavioural Science Research

The past couple of years have been painful for Social Sciences, with the replicability crisis putting a dent on the credibility of multiple studies in the field – from social priming effects to power poses and will power. An effort to reproduce effects reported in more than 100 cognitive and social psychology studies in three journals, called the Reproducibility Project, has found that findings from around 60 studies do not hold up when retested. Even when effects were replicated, they were weaker than reported in the original studies.

The replicability debate has been focussed, to a large extent, on experimental design and effect sizes. It is suggested that low-power research designs (smaller sample sizes) and lower or weaker effect size studies were more likely unable to be replicated. Additionally, an inherent bias in publication favouring positive results is argued to contribute towards the replication crisis.

An often overlooked part of the discussion seems to be the social context of the experiment and it’s effect on the participants themselves. Currently, academic researchers are sticklers for controlled design, this way the effects of multiple factors on behaviour can be reduced to just one. In view of this, in most universities, the research lab, usually cubicles/ computer laboratory is a heavily controlled, isolated environment. Having a controlled physical environment, however, does not preclude the participants from coming in to the research with their own motivations, dispositions, expectations and emotions. These cannot be dismissed as irrelevant to the study at hand just because the study has been stripped of any context. On the other hand, they exert a large influence on outcomes of the study.

For example, aspects of the experimental setting can influence the participants’ reaction to stimuli presented by the experimenter. Participants in psychology studies get paid, and are motivated to play the role of ‘good’ subjects – ascribe to what they think the experimenter wants – these are termed ‘Demand Effects’.  Participants consciously try to recreate experimenters’ hypotheses using available cues. Any psychology experimenter will attest to this fact. As a student, when I conducted my research on Automatic Priming, I used the same testing protocol – picked solitary computer terminals, used a confederate to trick participants into believing they were engaged in two separate studies – one to deploy the priming intervention (‘professor’ versus ‘hooligan’) and another to study the effect it had on knowledge (IQ test). We did probe participants on what they thought the experiment was about and so on, but at the end of the day, the truth is that most participants had their own hypotheses about what we were trying to prove and played up to their hypotheses. Experimenters themselves unwittingly influence participants with their expectations – which participants want to play up to, dubbed ‘Experimenter Effects’.

Psychology is the study of human behaviour – in our anxiety to ensure that it is a strict science, we are using the same experimental models that we use to study physics to study human behaviour. It is time psychology experiments stop treating participants as passive receptors of stimuli. What we want to study are the motivations, the emotions, the beliefs and dispositions for different contexts – why try to make the participants leave those behind at home (which they won’t anyway). Our research will be richer if we simulate the real-life context that we are trying to study, rather than control for it, so the decisions and outcomes of research will be closer to home.

Research at FinalMile attempts todo just this. With our EthnoLab, we simulate real-life contexts as far as possible – we want the decisions in the Ethnolab to reflect decisions taken in real life, not create an alien context which leads to perceived ‘correct answers’. This might mean recreating the real-life environment – either physically, or virtually. The EthnoLab marries the practicality of a controlled laboratory with the ‘real-life’ness of Ethnography. As Smith and Semin (2004) put it :  “The true strength of the laboratory is not its supposed insulation of behavior from context effects, but its flexibility in allowing experimenters to construct very different types of contexts, suited to test different types of hypotheses.” Welcome to Behavioural research v2.0!

Image Credit: american.edu

Cheating ourselves to Death?

India is often referred to as the diabetes capital of the world, with around 41 million people living with diabetes in 2007, and projected to reach 68 million by 2025. In one of our engagements we were trying to understand how people living with diabetes manage this disease. One of the perplexing observations was that many people had the belief that their diabetes is under control. This conflicts with most data and expert opinion which suggests that majority of diabetes cases are uncontrolled.

We were trying to understand the source of this belief and started interviewing close family members of patients. One of the most interesting factors that we heard when we spoke to family members of these patients was that these patients “prepared” themselves before going for a blood glucose test. A week before they get their blood sugar tested, they would change their lifestyle – they would exercise, go for walks and control their diet. So when testing happens they get a more favorable result than their actual condition. It looks so irrational that people would cheat themselves into believing that their condition is better than it actually is, thereby putting themselves at risk of not getting the right treatment.

What explains this seemingly irrational behavior? Why would intelligent people who are aware of the dangers of the disease that they have, not want to know the truth and provide their physician with more accurate data for better decision making?

One of the moderators of decision making is the kind of mental models that people create in life that helps them simplify the world. While this is often great to improve efficiency of decision making, it could be deadly if used in the wrong context. A very popular example of a mental model being used in the wrong context is in the case of diarrhea. As Sendhil Mullainathan explains in this video, 35-50% of the mothers in Rural India think that they should reduce fluids if their child has diarrhea. They use an intuitive mental model of a leaky bucket – that you should not pour water into a leaky bucket if it has to stop leaking. This makes diarrhea, something that can be easily managed to the status of a deadly condition.

In the case of diabetes, the mental model that patients have is that one should not to fail a test. People look at each blood test as a test of how well they are managing their condition, thereby framing the issue as a judgment on their own capabilities. And not one that objectively measures the status of their condition and as an input into a treatment regimen that would help their doctor take better decisions.

How do we address such a condition? Breaking mental models is often a high-investment long-term game. One of the approaches that we take at Final Mile is to see how one can work with existing mental models rather than fight it. In this case, simply by encouraging people to adopt HbA1C instead of a spot test can help address this behavioral issue and get a more accurate measure of their condition. It is a simple intervention, but one that addresses the inherent risks of misdiagnosis. The other intervention is to address how doctors and counselors frame the test – it is important that patients do not see this as a test that they fail or pass but one that helps calibrate the medication for a chronic condition.

Nudging Accurate Road Crash Investigation

Around 400 people lose their lives on Indian roads every day. In order to reduce this death count and improve road safety, we need to identify the root causes and then determine solutions. Accurate data capture is critical for this.

The road incident data in India is collected through local police and recorded in FIRs (First Information Report) at the police stations. The investigation reports from the local police stations are sent to the State Governments, who in turn send their report to the Central Government. The police data is used in publishing annual reports of ‘Accidental Deaths and Suicides in India’ and ‘Road Accidents in India’ by two government organizations, National Crime Records Bureau and Ministry of Road Transport and Highways respectively.

Data collected by the police could be subjective as it depends on how the police personnel interpret and ascribe reasons for the incident during investigation at the site. A motor vehicle collision could happen due to faulty road design, pedestrian’s fault, driver’s fault, poor visibility and many other reasons, but more often than not, the driver is blamed. To address this bias and to make crash investigation more objective, an effective and efficient data collection system should be in place.

The route patrolling team (RPT) is usually the first responder for incidents on national or state highways. They record the details of the incidents – the people involved, vehicles involved, cause and type of the incident, date and time of the incident, etc. Because of their ‘first responder’ status, many a times, RPT data forms the basis for the police report. For our Road Safety projects, we designed Crash Investigation Form for the RPT (route patrolling team) with a goal of making the form more objective.

To illustrate the problem of subjectivity, let us look at some examples from our Road Safety projects.

While interacting with a truck driver involved in an incident, the driver revealed that he lost control as he was blinded for a second due to glare of headlights in the opposite lane and rammed into the median opening structure. The RPT classified the cause of this incident as ‘driver was sleepy.’

In another incident, the car driver (admittedly over-speeding) said that in order to avoid running into an auto trying to cut across at the median opening, he turned his vehicle and rammed into the highway guardrail. While investigating the incident, the RPT concluded the reason as ‘driver was over-speeding.’

In both these incidents, the actual causes were ignored – glare in the first incident and a vehicle trying to cut across at high speed in the second incident. We realized that many a times, the RPT interpret the cause of the incident and prepare the report accordingly rather than objectively recording the information, resulting in skewed data and misrepresentation of the incidents.

We redesigned the Crash Investigation Form to nudge the RPT to capture the details in sequence and to hold off analysis of the cause of the incident till the very end of the report. Collecting details like ‘position of the vehicle(s)’ [Fig 1] before and after incident and ‘vehicle damage status’ [Fig 2] is crucial in identifying the cause of the incident and person responsible for the incident.

 

Fig 1

 

Fig 2

Also, capturing incident details is a monotonous process due to which the person collecting data might fail to focus on specific details resulting in skewed data. We designed the Crash Investigation Form with visuals [Fig 3] to make the data collection process more engaging, self explanatory and easy.    

Fig 3

A new incident recording format for the police has been approved and introduced recently by the Transport Research Wing, Ministry of Road Transport & Highways, Government of India. This report is intended to minimize subjectivity while recording the incident details and to arrive at the actual cause of the incident by capturing the technical details like road surface and traffic control systems in place at the incident site. Will this new format aid the police personnel in accurate collection of crash data and minimize subjectivity?

To begin with, it could be difficult for few police personnel to understand and remember terms like ‘staggered junction,’ ‘four arm junction’ ‘paved/unpaved surface,’ etc., (even though workshops are planned to train the police) while filling the details at the incident site. The process to arrive at the ‘cause of the incident’ in the new format could have been more analytical by capturing details like ‘position of the vehicle(s) before and after incident’ and ‘vehicle damage status’ – this is very crucial as solutions/preventive measures depend on the cause of the incidents. The text-heavy report could also have been made easier with illustrations to aid data-capture and reduce monotony. All these aspects might eventually result in incorrect data collection negating the purpose of designing a new format.

For an unbiased and accurate data collection, it is imperative that the Incident Recording Form should be comprehensive, yet easy to understand; visually more engaging; and follow a sequence in data collection with an objective approach in determining the cause of the incident.

P.S.  Only few sections from the Crash Investigation Report designed by us have been published here for reference.

The American Health Care Act’s (dis)Incentives 

http://www.commondreams.org/views/2017/03/09/american-health-care-act-wealth-grab-not-health-plan

The Republicans have recently released their proposed alternative to the Affordable Care Act (ACA), entitled: the American Health Care Act (AHCA).

One of the prime concerns they hoped to address was the Individual Mandate, as defined by the ACA. Their replacement, however, appears set to miss the behavioral objective which the enrollment incentive was designed to achieve.

Behavioral principles such as Prospect Theory, Certainty Effect and Present Bias define the short comings of their proposed legislation.

 

The Ongoing Challenge

After 7 years of fighting against the ACA, which the Democratically led 111th Congress enacted and President Obama signed into law in 2010, the Republican Party, under the 114th Congress, has presented their first detailed plan to replace the ACA. They are calling it the American Health Care Act.

The ACA is a complicated bit of policy, which in large measure reflects the nature of the problem(s) that it tries to address. Equally, the objections that people have with the ACA are complex. There are any number of aspects to the conversation regarding health care in America, and to both of these policies in particular, that might benefit from a behavioral science perspective. At the risk of taking a reductionists approach to a systemic issue, I’d like to focus on just one of those concerns.

The ACA enacted an individual mandate which said, in principle, if you don’t enroll yourself in an approved health insurance plan, the federal government will fine you with a tax penalty. For each full month that you forego health insurance coverage, the tax penalty would equal, roughly, $58.00 per adult. The maximum penalty for going without coverage for a full year would be $695.00.

For the ACA to work, from a purely economic perspective, healthy/ lucky people, who do not really need insurance, would need to pay into the insurance system so that unhealthy/ unlucky people can benefit. This is how insurance of all stripes works, generally. One of the challenges that the ACA was trying to address with the individual mandate was: ‘how to encourage young, healthy people to buy into the insurance system’ so that the system might function as a whole.

 

Prospect Theory & Bounded Rationality Aligned

From a behavioral economics perspective, the disincentives of the ACA seemed to align with the desired behavioral outcomes. The economic and emotional logic being: “As a healthy person I can either pay the government a small tax penalty, which will help subsidize the system, and forego health insurance. Or, I can purchase an approved insurance plan, which will help subsidize the system, and receive some benefits of being insured.”

The bounded rationality of the trade-off is pretty simple: “I know that I’m going to pay something, no matter what. I can either pay a penalty and receive nothing, or I can pay for a plan that at least covers me for the unexpected.” In both cases there is a certain near-term loss (either the insurance premium or the tax penalty). But, if I choose to go uninsured, there is a higher degree of uncertainty in longer-term outcomes. Whereas, if I choose the insured route, I have the certainty that I will be covered (relative to the chosen plan).

One of the caveats to this disincentive, however, was uncertainty regarding the relative effectiveness of the tax penalty. At $695 per year, it was significantly lower than the cost of any given insurance premium, then or now. People therefore criticized it, speculatively, as lacking the economic muscle to be an effective disincentive of foregoing insurance. While others made the argument that the tax penalty worked more from an emotional utility perspective, creating a new social norm through rules based signaling.

 

Alternative Assumptions, Alternative Penalties

The American Health Care Act, on the other hand, attempts to eradicate the individual mandate, as this was a major ideological sticking point with Republicans. (Not to mention that the Republicans promised to repeal the whole of the ACA in its entirety). What they propose in place of the individual mandate is to increase the insurance premiums on those Americans whose insurance coverage has lapsed for more than 60 days.

For example, assume you must forego health insurance for more than 60 days; this may be due to a chronic economic situation in which insurance is normally beyond your budget, or a shorter-term concern related to unemployment or other temporary factor. You won’t be fined by the government right away under the AHCA. In fact, nothing will happen until you attempt to insure yourself again. At which time you will face the prospect of a 30% increase in premium cost on whatever insurance plan you propose to purchase. That 30% increase would be in effect for the next 12 months of insurance coverage.

 

Consequences of Structure and Timing

From a behavioral design perspective, the proposed structure of the Republicans replacement for the individual mandate appears set to satisfy nobody’s interests. In fact, it may have potentially devastating impact on both individual health outcomes (as fewer people will be able to afford coverage) and on the bottom line of the insurers (as an insurance death spiral may ensue).

For young, healthy people who have never had an insurance plan before, the AHCA disincentive appears set to achieve the opposite of it’s stated goal. A 30% increase of an insurance policy that these young people never had equals zero in their accounting. They have no idea what the difference in cost (the true cost of the disincentive) would be because they have no baseline with which to make a comparison. Additionally, the penalty is off in the distant, uncertain future in which they may eventually need insurance. Given the impact of our present bias, this more or less insures that the additional cost is out-of-sight and out-of-mind, therefore rendering the disincentive wholly ineffective.

The other segment of the population that is most likely to forego insurance coverage are folks that are in the unenviable economic situation as to be unable to afford a policy in the immediate term due to circumstances. They would be balancing a near-term certainty (“I can’t afford it right now.”) with a distant uncertainty (“Will I absolutely need insurance in the next X# of days?”). The certainty effect in this case encourages a wait and see approach.

 

Intent is the Key Word in Communication

What the penalty might fully accomplish, however, is a guaranteed death spiral as the incentives seem structured to encourage people of all kinds to accept a lack of coverage in the short-term and to raise the bar-to-entry over the long-term.

How is such an incentive design supposed to be interpreted? Is this an unintended consequence resulting from chasing some other objective? Or is this the intentional application of a dark pattern? Your answer to that question probably depends on your political inclinations…