If Applied Econometrics Were Easy, LLMs Could Do It!

Summary

Can AI do applied econometrics and causal inference? Can LLMs pick up on the nuances and social norms that dictate so many of the decisions made in applied work and reflect them in response to a prompt? LLMs bring to the table incredible capabilities and efficiencies and opportunities to create value. But there are risks when these tools are used like Dunning-Kruger-as-a-Service (DKaaS), where the critical thinking and actual learning begins and ends with prompt engineering and a response. We have to be very careful to recognize as Philip Tetlock describes in his book "Superforecasters" that there is a difference between mimicking and reflecting meaning vs. originating meaning.  To recognize that it’s not just what you know that matters, but how you know what you know. The second-handed tendency to believe that we can or should be outsourcing, nay, sacrificing our thinking to AI in exchange for misleading if not false promises about value, is philosophically and epistemically disturbing.

AI vs. Causal Thinking

This is a good article, from causal lens: Enterprise Decision Making Needs More Than Chatbots. 

"while LLMs are good at learning and extracting information from a corpus, they’re blind to something that humans do really well – which is to measure the impact of one’s decisions." 

In a recent talk Cassie Kozrykov puts it well: "AI does not automate thinking!"

   

Channelling Judea Pearl, understanding what makes a difference (causality)requires more than data, it also requires something not in the data to begin with. So much of the hype around AI is based on a tools and technology mindset. As Captain Jack Sparrow says about ships in Pirates of the Caribbean, a ship is more than sails and rudders, those are things a ship needs. What a ship really is, is freedom. Causal inference is more than methods and theorems, those are things causal inference needs, but what it really is, is a way of thinking. And in business, what is required is an alignment of thinking. For instance, in his article The Importance of Being Causal, Ivor Bojinov describes the Causal Data Analysis Review Committee at LinkedIn. It is a common best practice in learning organizations that leverage experimentation and causal inference. 

If you  attended very many of those reviews you begin to appreciate the amount of careful thinking required to understand the business problem, frame the hypothesis, and translate it to an analytical solution....then interpret the results and make a recommendation about what action to take next. Similarly a typical machine learning workflow requires up front thinking and problem framing. But unlike training an ML model, as Scott Lundberg describes (see my LI Post: Beyond SHAP Values and Crystal Balls), understanding what makes a difference is not just a matter of letting an algo figure out the best predictors  and calling it a day, there is an entire garden of forking paths to navigate and each turn requires more thinking and a vast difference in opinions among 'experts' about which direction to go.

As I discussed in a past post about forking paths in analysis: 

"even if all I am after is a single estimate of a given regression coefficient, multiple testing and researcher degrees of freedom may actually become quite a relevant concern...and this reveals the fragility in a lot of empirical work that prudence would require us to view with a critical eye"

Sure you could probably pair a LLM with statistical software and a data base connection and ask it to run a regression, but getting back to Jack Sparrow's ship analogy, a regression is more than just fitting a line to data and testing for heteroskedasticity and multicollinearity (lets hope if LLMs train on econometrics textbooks they don't weight the value of information by the amount of material dedicated to multicollinearity!!!) and the laundry list of textbook assumptions. AI could probably even describe in words a mechanical interpretation of the results. All of that is really cool, and something like that could save a lot of time and augment our workflows (which is valuable) but we also have to be careful about that tools mindset creeping back on us. All those things that AI may be able to do are only the things regression needs, but to get where we need to go, to understand why, we need way more than what AI can currently provide. We need thinking. So even for a basic regression, depending on our goals, the thinking required is currently and may always be beyond the capabilities of AI.

When we think about these forking paths encountered in applied work, each path can end with a different measure of impact that comes with a number of caveats and tradeoffs to think about. There are seldom standard problems with standard solutions. The course of action taken requires conscious decisions and the meeting of minds among different expert judgements (if not explicitly then implicitly) that considers all the tradeoffs involved in moving from what may be theoretically correct and what is practically feasible. 

In his book, "A Guide to Econometrics" Peter Kennedy states that "Applied econometricians are continually faced with awkward compromises" and offers a great story about what it's like to do applied work: 

"Econometric theory is like an exquisitely balanced French recipe, spelling out precisely with how many turns to mix the sauce, how many carats of spice to add, and for how many milliseconds to bake the mixture at exactly 474 degrees of temperature. But when the statistical cook turns to raw materials, he finds that hearts of cactus fruit are unavailable, so he substitutes chunks of cantaloupe; where the recipe calls for vermicelli he used shredded wheat; and he substitutes green garment die for curry, ping-pong balls for turtles eggs, and for Chalifougnac vintage 1883, a can of turpentine."

What choice would AI driven causal inference make when it has to make the awkward compromise between Chalifougnac vintage 1883 and turpentine and how would it explain the choice it made and the thinking that went into it? How would that choice stack up against the opinions of four other applied econometricians who would have chosen differently? 

As Richard McElreath discusses in his great book Statistical Rethinking:

"Statisticians do not in general exactly agree on how to analyze anything but the simplest of problems. The fact that statistical inference uses mathematics does not imply that there is only one reasonable or useful way to conduct an analysis. Engineers use math as well, but there are many ways to build a bridge." 

This is why in applied economics so much of what we may consider as 'best practices' are as much the result of social norms and practices as they are textbook theory. These norms are often established and evolve informally over time and sometimes adapted to the particulars of circumstances and place unique to a business or decision making environment, or research discipline (this explains the language barriers for instance between economists and epidemiologists and why different language can be used to describe the same thing and the same language can mean different things to different practitioners). A kind of result of human action but not human design, many best practices may seldom be formally codified or published in a way accessible to train a chatbot to read and understand. Would an algorithm be able to understand and relay back this nuance? I gave this a try by asking chatGPT about linear probability models (LPMs), and while I was impressed with some of the detail, I'm not fully convinced at this point based on the answers I got. While it did a great job articulating the pros and cons of LPMs vs logistic regression or other models, I think it would leave the casual reader with the impression that they should be wary of relying on LPMs to estimate treatment effects in most situations. So they miss out on the practical benefits (the 'pros' that come from using LPMs) while avoiding the 'cons' that as Angrist and Pischke might say, are mostly harmless. I would be concerned about more challenging econometric problems with more nuance and more appeal to social norms and practices and thinking that an LLM may not be privy to.

ChatGPT as a Research Assistant

Outside of actually doing applied econometrics and causal inference, I have additional concerns with LLMs and AI when it comes to using them as a tool for research and learning. At first it might seem really great if instead of reading five journal articles you could just have a tool like chatGPT do the hard work for you and summarize them in a fraction of the time! And I agree this kind of summary knowledge is useful, but probably not in the way many users might think. 

I have been thinking a lot about how much you get out of putting your hands on a paper or book and going through it and wrestling with the ideas, the paths leading from from hypotheses to the conclusions, and how the cited references let you retrace the steps of the authors to understand why, either slowly nudging your priors in new directions or reinforcing your existing perspective, and synthesizing these ideas with your own. Then summarizing and applying and communicating this synthesis with others. 

ChatGPT might give the impression that is what it is doing in a fraction of the time you could do it (literally seconds vs. hours or days). However, even if it gave the same summary you could write verbatim the difference couldn't be as far apart as night and day in terms of the value created. There is a big difference between the learning that takes place when you go through this process of integrative complex thinking vs. just reading a summary delivered on a silver platter from chatGPT. I’m skeptical what I’m describing can be outsourced to AI without losing something important. I also think there are real risks and costs involved when these tools are used like Dunning-Kruger-as-a-Service (DKaaS), where the critical thinking and actual learning begins and ends with prompt engineering and a response. 

When it comes to the practical application of this knowledge and thinking and solving new problems it’s not just what you know that matters, but how you know what you know. If all you have is a summary, will you know how to navigate the tradeoffs between what is theoretically correct and what is practically feasible to make the best decision in terms of what forking path to take in an analysis? Knowing about the importance of social norms and practices in doing applied work, and if the discussion above about LPMs is any indication, I'm not sure. And with just the summary, will you be able to quickly assimilate new developments in the field....or will you have to go back to chatGPT. How much knowledge and important nuance is lost with every update? What is missed? Thinking!

As Cassie says in her talk, thinking is about:

"knowing what is worth saying...knowing what is worth doing, we are thinking when we are coming up with ideas, when we are solving problems, when we are being creative"

AI is not capable of doing these things, and believing and even attempting or pretending that we can get these things on a second-handed basis from an AI tool will ultimately erode the real human skills and capabilities essential to real productivity and growth over the long run. If we fail to accept this we will hear a giant sucking sound that is the ROI we thought we were going to get from AI in the short run by attempting to automate what can't be automated. That is the false promise of a tools and technology mindset.

It worries me that this same tools and technology based data science alchemy mindset has moved many managers who were once were sold the snake oil that data scientists could simply spin data into gold with deep learning, will now buy into the snake oil that LLMs will be able to spin data into gold and do it even cheaper and send the thinkers packing! 

Similarly Cassie says: "that may be the biggest problem, that management has not learned how to manage thinking...vs. what you can measure easily....thinking is something you can't force, you can only get in the way of it."

She elaborates a bit more about this in her LinkedIn post: "A misguided view of productivity could mean lost jobs for workers without whom organizations won't be able to thrive in the long run - what a painful mistake for everyone."

Thunking vs. Thinking

I did say that this kind of summary info can be useful. And I agree that the kinds of things that AI and LLMs will be useful for are what Cassie refers to in her talk as 'thunking.'  The things that consume our time and resources but don't require thinking. Having done your homework, the kind of summary information you get from an LLM can help reinforce your thinking and learnings and save time in terms of manually googling or looking up a lot of things you once knew but have forgotten. If there is an area you haven't thought about in a while it can be a great way to help get back up to speed. And when trying to learn new things, it can be leveraged to speed up some aspects of your discovery process or make it more efficient, or even help challenge or vet your thinking (virtually bouncing ideas back and forth). But to be useful, this still requires some background knowledge and should never be a substitute for putting your hands on a paper and doing the required careful and critical thinking.

One area of applied econometrics I have not mentioned is the often less glamorous work it takes to implement a solution. In addition to all the thinking involved in translating the solution and navigating the forking paths, there is a lot of time spent accessing and transforming the data and implementing the estimation that involves coding (note even in the midst of all that thunking work there is still thinking involved - sometimes we learn the most about our business and our problem while attempting to wrangle the data - so this is also a place where we need to be careful about what we automate). Lots of data science folks are also using these tools to speed up some of their programming tasks. I'm a habitual user of stack-exchange and git hub and constantly recycle my own code or others' code. But I burn a lot of time somedays in search of what I need. That's the kind of thunking that it makes since to enlist new AI tools for!

Conclusion: Thinking is Our Responsibility

I've observed two extremes when it comes to opinions about tools like ChatGPT. One is that LLMs have the knowledge and wisdom of Yoda and will solve all of our problems. The other extreme is that because LLMs don't have the knowledge and wisdom of Yoda they are largely irrelevant. Obviously there is middle ground and I am trying to find it in this post. And I think Cassie has found it:

"AI does not automate thinking. It doesn't! There is a lot of strange rumblings about this that sound very odd to me who has been in this space for 2 decades"

I have sensed those same rumblings and it should make us all feel a bit uneasy. She goes on to say:

"when you are not the one making the decision and it looks like the machine is doing it, there is someone who is actually making that decision for you...and I think that we have been complacent and we have allowed our technology to be faceless....how will we hold them accountable....for wisdom...thinking is our responsibility"

Thinking is a moral responsibility. Outsourcing our thinking and fooling ourselves into thinking we can get knowledge and wisdom and judgment second-handed from a summary written by an AI tool, and to believe that is the same thing and provides the same value as what we could produce as thinking humans is a dangerous illusion when ultimately, thinking is the means by which the human race and civil society ultimately thrives and survives. In 2020 former President Barak Obama emphasized the importance of thinking in a democracy: 

"if we do not have the capacity to distinguish what's true from what's false, then by definition the marketplace of ideas doesn't work. And by definition our democracy doesn't work. We are entering into an epistemological crisis." 

The wrong kind of tools and technology mindset, and obsequiousness toward the technology, and a second-handed tendency to believe that we can or should be outsourcing, nay, sacrificing our thinking to AI in exchange for misleading if not false promises about value, is philosophically and epistemically disturbing.

LLMs bring to the table incredible capabilities and efficiencies and opportunities to create value. But we have to be very careful to recognize as Philip Tetlock describes in his book Superforecasters, that there is a difference between mimicking and reflecting meaning vs. originating meaning.  To recognize that it’s not just what you know that matters, but how you know what you know. To repurpose the closing statements from the book Mostly Harmless Econometrics: If applied econometrics were easy, LLMs could do it.

Additional Resources:

Thunking vs Thinking: Whose Job Does AI Automate? Which tasks are on AI’s chopping block? Cassie Kozrykov. https://kozyrkov.medium.com/thunking-vs-thinking-whose-job-does-ai-automate-959e3585877b

Statistics is a Way of Thinking Not a Just a Box of Tools. https://econometricsense.blogspot.com/2020/04/statistics-is-way-of-thinking-not-just.html 

Will There Be a Credibility Revolution in Data Science and AI? https://econometricsense.blogspot.com/2018/03/will-there-be-credibility-revolution-in.html 

Note on updates: An original version of this post was written on July 29 in conjunction with the post On LLMs and LPMs: Does the LL in LLM Stand for Linear Literalism? Shortly after posting I ran across Cassie's talk and updated to incorporate many of the points she made, with the best of intentions. Any  misrepresentation/misappropriation of her views is unintentional. 

Causal Decision Making with non-Causal Models

In a previous post I noted: 

" ...correlations or 'flags' from big data might not 'identify' causal effects, but they are useful for prediction and might point us in directions where we can more rigorously investigate causal relationships"

Recently on LinkedIn I discussed situations where we have to be careful about taking action on specific features in a correlational model, for instance changing product attributes or designing an intervention based on interpretations of SHAP values from non-causal predictive models. I quoted Scott Lundberg:

"regularized machine learning models like XGBoost will tend to build the most parsimonious models that predict the best with the fewest features necessary (which is often something we strive for). This property often leads them to select features that are surrogates for multiple causal drivers which is "very useful for generating robust predictions...but not good for understanding which features we should manipulate to increase retention."

So sometimes, we may go into a project with the intention of only needing predictions. We might just want to target offers or nudges to customers or product users but not think about this in causal terms at first. But, as I have discussed before the conversation often inevitably turns to causality, even if stakeholders and business users don't use causal language to describe their problems. 

"Once armed with predictions, businesses will start to ask questions about 'why'... they will want to know what decisions or factors are moving the needle on revenue or customer satisfaction and engagement or improved efficiencies...There is a significant difference between understanding what drivers correlate with or 'predict' the outcome of interest and what is actually driving the outcome."

This would seem to call for causal models. However, in their recent paper Carlos Fernández-Loría and Foster Provost make an exciting claim:

“what might traditionally be considered “good” estimates of causal effects are not necessary to make good causal decisions…implications above are quite important in practice, because acquiring data to estimate causal effects accurately is often complicated and expensive. Empirically, we see that results can be considerably better when modeling intervention decisions rather than causal effects.”

Now in this case they are not talking about causal models related to identifying key drivers of an outcome, so it is not contradicting anything mentioned above or in previous posts. Particularly they are talking about building models for causal decision making (CDM) that are simply focused on making decisions about who to 'treat' or target.  In this particular scenario businesses are leveraging predictive models to target offers, provide incentives, or make recommendations. As discussed in the paper, there are two broad ways of approaching this problem. Let's say the problem is related to churn.

1) We could predict risk of churn and target members most likely to churn. We could do this with a purely correlational machine learning model. The output or estimand from this model is a predicted probability p() or risk score. They also refer to these kinds of models as 'outcome' models

2) We could build a causal model, that predicts causal impact of an outreach. This would allow us to target customers that we can most likely 'save' as a result of our intervention. They refer to this estimand as a causal effect estimate CEE. Building machine learning models that are causal can be more challenging and resource intensive.

It is true at the end of the day we want to maximize our impact. But the causal decision is ultimately who do we target in order to maximize impact. They point out this causal decision does not necessarily hinge on how accurate our point estimate is related to causal impact as long as errors in prediction still lead to the same decisions about who to target.

What they find is that in order to make good causal decisions about who to 'treat' we don't have to have super accurate estimates of the causal impact of treatment (or models focused on CEE). In fact they talk through scenarios and conditions where outcome models like #1 above that are non-causal, can perform just as well or sometimes better than more accurate causal models focusing on CEE. 

In other words, correlational outcome models (like #1) can essentially serve as proxies for the more complicated causal models (like #2), even if the data used to estimate these 'proxy' models is confounded.

 Scenarios where this is most likely include:

1) Outcomes used as proxies and (causal)effects are correlated

2) Outcomes used as proxies are easier to estimate than causal effects

3) Predictions are used to rank individuals

They also give some reasons why this may be true. Biased non-causal models built on confounded data may not be able to identify true causal effects, but still be useful for identifying the optimal decision. 

"This could occur when confounding is stronger for individuals with large effects - for example if confounding bias is stronger for 'likely' buyers, but the effect of adds is also stronger for them...the key insight here is that optimizing to make the correct decision generally involves understanding whether a causal effect is above or below a given threshold, which is different from optimizing to reduce the magnitude of bias in a causal effect estimate."

"Models trained with confounded data may lead to decisions that are as good (or better) than the decisions made with models trained with costly experimental data, in particular when larger causal effects are more likely to be overestimated or when variance reduction benefits of more and cheaper data outweigh the detrimental effect of confounding....issues that make it impossible to estimate causal effects accurately do not necessarily keep us from using the data to make accurate intervention decisions."

Their arguments hinge on the idea that what we are really solving for in these decisions is based on ranking:

"Assuming...the selection mechanism producing the confounding is a function of the causal effect - so that the larger the causal effect the stronger the selection-then (intuitively) the ranking of the preferred treatment alternatives should be preserved in the confounded setting, allowing for optimal treatment assignment policies from data."

A lot of this really comes down to proper problem framing and appealing to the popular paraphrasing of George E. P. Box - all models are wrong, but some are useful. It turns out in this particular use case non-causal models can be as useful or more useful than causal ones.

And we do need to be careful about the nuance of the problem framing. As the authors point out, this solves one particular business problem and use case, but does not answer some of the most important causal questions businesses may be interested in:

"This does not imply that firms should stop investing in randomized experiments or that causal effect estimation is not relevant for decision making. The argument here is that causal effect estimation is not necessary for doing effective treatment assignment."

They go on to argue that randomized tests and other causal methods are still core to understanding the effectiveness of interventions and strategies for improving effectiveness. Their use case begins and ends with what is just one step in the entire lifecycle of product development, deployment, and optimization. In their discussion of further work they suggest that:

"Decision makers could focus on running randomized experiments in parts of the feature space where confounding is particularly hurtful for decision making, resulting in higher returns on their experimentation budget."

This essentially parallels my previous discussion related to SHAP values. For a great reference for making practical business decisions about when this is worth the effort see the HBR article in the references discussing when to act on a correlation.

So some big takeaways are:

1) When building a model for purposes of causal decision making (CDM) even a biased model (non-causal) can perform as well or better than a causal model focused on CEE.

2) In many cases, even a predictive model that provides predicted probabilities or risk (as proxies for causal impact or CEE) can perform as well or better than causal models when the goal is CDM.

3) If the goal is to take action based on important features (i.e. SHAP values as discussed before) however, we still need to apply a causal framework and understanding the actual effectiveness of interventions may still require randomized tests or other methods of causal inference.

HT: This paper was previously discussed at Andrew Gelman's blog here: https://statmodeling.stat.columbia.edu/2021/11/01/how-different-are-causal-estimation-and-decision-making/ 

References: 

Causal Decision Making and Causal Effect Estimation Are Not the Same... and Why It Matters. Carlos Fernández-Loría and Foster Provost. 2021. https://arxiv.org/abs/2104.04103

When to Act on a Correlation, and When Not To. David Ritter. Harvard Business Review. March 19, 2014. 

Be Careful When Interpreting Predictive Models in Search of Causal Insights. Scott Lundberg. https://towardsdatascience.com/be-careful-when-interpreting-predictive-models-in-search-of-causal-insights-e68626e664b6  

Additional Reading:

Laura B Balzer, Maya L Petersen, Invited Commentary: Machine Learning in Causal Inference—How Do I Love Thee? Let Me Count the Ways, American Journal of Epidemiology, Volume 190, Issue 8, August 2021, Pages 1483–1487, https://doi.org/10.1093/aje/kwab048

Petersen, M. L., & van der Laan, M. J. (2014). Causal models and learning from data: integrating causal modeling and statistical estimation. Epidemiology (Cambridge, Mass.), 25(3), 418–426. https://doi.org/10.1097/EDE.0000000000000078

Explaining the Behavior of Black-Box Prediction Algorithms with Causal Learning. Numair Sani, Daniel Malinsky, Ilya Shpitser arXiv:2006.02482v3  

Related Posts:

Will there be a credibility revolution in data science and AI? 

Statistics is a Way of Thinking, Not a Toolbox

Big Data: Don't Throw the Baby Out with the Bathwater

Big Data: Causality and Local Expertise Are Key in Agronomic Applications 

The Use of Knowledge in a Big Data Society

Science Communication for Business and Non-Technical Audiences: Stigmas, Strategies, and Tactics

If you are a reader of this blog you are familiar with the number of posts I have shared about machine learning and causal inference and the benefits of education in economics. I have also discussed how there are important gaps sometimes between theory and application. 

In this post I am going to talk about another important gap related to communication. How do we communicate the value of our work to a non-technical audience? 

We can learn a lot from formal coursework, especially in good applied programs with great professors. But if not careful we can pick up on mental models and habits of thinking that turn out to weigh us down too, particularly for those that end up working in very applied business or policy settings. How we deal with these issues becomes important to career professionals and critical to those involved in science communication in general whether we are trying to influence business decision makers, policy makers, or consumers and voters.

In this post I want to discuss communicating with intent, paradigm gaps, social harassment costs, and mental accounting.

As stated in The Analytics Lifecycle Toolkit: "no longer is it sufficient to give the technical answer, we must be able to communicate for both influence and change."

Communicating to Business and Non-Technical Audiences - or - The Laffer Curve for Science Communication

For those who plan to translate their science backgrounds to business audiences (like many data scientists coming from scientific backgrounds) what are some strategies for becoming better science communicators?  In their book Championing Science: Communicating your Ideas to Decision Makers Roger and Amy Aines offer lots of advice. You can listen to a discussion of some of this at the BioReport podcast here. 

Two important themes they discuss is the idea of paradigm gaps and intent. Scientists can be extremely efficient communicators through the lens of the paradigms they work in. 

As discussed in the podcast, a paradigm is all the knowledge a scientist or economist may have in their head specific to their field of study and research. Unfortunately there is a huge gap between this paradigm and its vocabulary and what non-technical stakeholders can relate to. They have to meet stakeholders where they are, vs. the audience they may find at conferences or research seminars. From experience, different stakeholders and audiences across different industries have different gaps. If you work for a consultancy with external pharma clients they might have a different expectation about statistical rigor than say a product manager in a retail setting. Even within the same business or organization, the tactics used in solving for the gap for one set of stakeholders might not work at all for a new set of stakeholders if you change departments. In other words, know your audience. What do they want or need or expect? What are their biases? What is their level of analytic or scientific literacy? How risk averse are they? Answers to these questions is a great place to start in terms of filling the paradigm gaps and to address the second point made in the podcast - speaking with intent.

As discussed in the podcast: "many scientists don't approach conversations or presentations with a real strategic intent in terms of what they are communicating...they don't think in terms of having a message....they need to elevate and think about the point they are trying to make when speaking to decision makers." 

As Bryan Caplan states in his book The Myth of the Rational Voter, when it comes to speaking to non-economists and the general public, they should apply the Laffer curve of learning, "they will retain less if you try to teach them more."

He goes on to discus that its not just what we say, but how we position it, especially when dealing with resistance related to misinformation and disinformation and systemic biases:

"irrationality is not a barrier to persuasion, but an invitation to alternative rhetorical techniques...if beliefs are in part consumed for their direct psychological benefits then to compete in the marketplace of ideas, you need to bundle them with the right emotional content."

In the Science Facts and Fallacies podcast (May 19, 2021) Kevin Folta and Cameron English discuss:

"We spend so much time trying to convince people with scientific principles....it's so important for us to remember what we learn from psychology and sociology (and economics) matters. These are turning out to be the most important sciences in terms of forming a conduit through which good science communication can flow."

Torsten Slok offers great advice in his discussion with Barry Ritholtz about working in the private sector as a PhD economist in the Masters in Business Podcast back in 2018: 

"there is a different sense of urgency and an emphasis on brevity....we offer a service of having a view on what the economy will do what the markets will do - lots of competition for attention...if you write long winded explanations that say that there is a 50/50 chance that something will happen many customers will not find that very helpful."

So there are a lot of great data science and science communicators out there with great advice. A big problem is this advice is often not part of the training that many of those with scientific or technical backgrounds receive, and an even bigger problem is that it is often looked down upon and even punished! I'll explain more below.

The Negative Stigma of Science Communication in the Data Science and Scientific Community

One of the most egregious things I see on social media is someone trying their best to help mentor those new to the analytical space (and improve their own communication skills) by sharing some post that attempts to describe some complicated statistical concept in 'layman's' terms - to only be rewarded by harassing and trolling comments. Usually this is about how they didn't capture every particular nuance of the theory, failed to include a statement about certain critical assumptions, or over simplified the complex thing they were trying to explain in simple terms to begin with. This kind of negative social harassment seems to be par for the course when attempting to communicate statistics and data science on social media like LinkedIn and Twitter.

Similarly in science communication, academics can be shunned by their peers when attempting to do popular writing or communication for the general public. 

In 'The Stoic Challenge' author William Irvine discusses Danial Kahneman's challenges with writing a popular book: 

"Kahneman was warned that writing a popular book would cause harm to his professional reputation...professors aren't supposed to write books that normal people can understand."

He describes, when Kahneman's book Thinking Fast and Slow made the New York Times best selling list Kahneman "sheepishly explained to his colleagues that the book's appearance there was a mistake."

In an EconTalk interview with economist Steven Levitt, Russ Roberts asks Levitt about writing his popular book Freakonomics:

"What was the reaction from your colleagues in the profession...You know, I have a similar route. I'm not as successful as you are, but I've popularized a lot of economics...it was considered somewhat untoward to waste your time speaking to a popular audience."

Levitt responded by saying the reaction was not so bad, but the fact that Russ had to broach the topic is evidence of the toxic culture that academics face when doing science communication. The negative stigma associated with good science communication is not limited to economics or the social and behavioral sciences. 

In his Talking Biotech podcast episode Debunking the Disinformation Dozen, scientist and science communicator Kevin Folta discusses his strident efforts facing off these toxic elements:

"I have always said that communication is such an important part of what we do as scientists but I have colleagues who say you are wasting your time doing this...Folta why are you wasting your time doing a podcast or writing scientific stuff for the public."

Some of this is just bad behavior, some of it is gatekeeping done in the name of upholding the scientific integrity of their field, some of it is the attempt of others to prove their competence to themselves or others, and maybe some of it is the result of people genuinely trying to provide peer review to their colleagues that they think have gone astray. But most of it is unhelpful when it comes to influencing decision makers or improving general scientific literacy. It doesn't matter how great the discovery, how impactful the findings, we have all seen from the pandemic that effective science communication is critical for overcoming the effects of misinformation and disinformation. A culture that is toxic toward effective science communication becomes an impediment to science itself and leaves a void waiting be filled by science deniers, activists, policy makers, decision makers, and special interests.

This can be challenging when you add the Dunning-Kruger effect to the equation. Those that know the least may be the most vocal while scientists and those with expertise sit on the sidelines. As Bryan Caplan states in his book The Myth of the Rational Voter:

"There are two kinds of errors to avoid. Hubris is one, self abasement is the other. The first leads experts to over reach themselves; the second leads experts to stand idly by while error reigns."

How Does Culture and Mental Accounting Impact Science Communication?

So as I've written above, in the scientific community there is sort of a toxic culture that inhibits good science communication. In the Two Psychologists Four Beers podcast  (WARNING: the intro of this podcast episode may contain vulgarity) behavioral scientist Nick Hobson makes an interesting comparison between MBAs and scientists. 

"as scientists we need to be humble with regards to our data...one thing we are learning from our current woes of replication (the replication crisis) is we know a lot less than we think. This has conditioned us to be more humble....vs. business school people that are trained to be more assertive and confident."

I'd like to propose an analogy relating to mental accounting. It seems like when a scientist gets their degree it comes with a mental account called scientific credibility. Speaking and writing to a general audience risks taking a charge against that account, and they are trained to be extremely frugal about managing it. Communication becomes an exercise in risk management.  If they say or communicate something that is of the slightest error, missing the slightest nuance, a colleague may call them out.  Gotcha! Psychologically, this would call for a huge charge against their 'account' and reputation. Its not quite a career ending mistake like making a fraudulent claim or faking data, but it's bad enough to be avoided at great cost. MBAs don't have a mental account called scientific credibility. They aren't long on academic credibility so they don't require putting on the communication hedges the way scientists often do. They come off as better communicators and more confident while scientists risk becoming stereotyped as unable to be effective communicators. 

To protect their balance at all costs and avoid social harassment from their peers, economists and scientists may tend to speak with caveats, hedges, and qualifications. This may also mean a delayed response. Before even thinking about communicating results in many cases requires in depth rigorous analysis, sensitivity checks etc. It requires doing science which is by nature slow while the public wants answers fast. Faster answers might mean less time for analysis which calls for more caveats. This can all be detrimental to effective communication to non-technical audiences. Answers become either too slow or too vague to support decision making (recall Torsten Slok's comments above). It gives the impression of a lack of confidence and relevance and a stereotype that technical people (economists, scientists, data scientists etc.) fail to offer definitive or practical conclusions. As Bryan Caplan notes discussing the role of economists in The Myth of the Rational Voter:

"when the media spotlight gives other experts a few seconds to speak their mind, they usually strive to forcefully communicate one or two simplified conclusions....but economists are reluctant to use this strategy. Though the forum demands it they think it unseemly to express a definitive judgement. This is a recipe for being utterly ignored."

Students graduating from economics and science based graduate programs may inherit these mental accounts and learn these 'hedging strategies' from their professors, from the program, and the seminar culture that comes with it.

Again, Nick Hobson offers great insight about how to deal with this kind of mental accounting in his own work:

"what I've wrestled with as I've grown the business is maintaining scientific integrity and the rigor but knowing you have to sacrifice some of it....you have to find and strike a balance between being data driven and humble while also being confident and strategic and cautious about the shortcuts you take."

In Thinking Fast and Slow, Kahneman argues that sometimes new leaders can produce better results because fresh thinkers can view problems without the same mental accounts holding back incumbents. The solution isn't to abandon scientific training and the value it brings to the table in terms of rigor and statistical and causal reasoning. The solution is to learn how to view problems in a way that avoids the kind of mental accounting I have been discussing. This also calls for a cultural change in the educational system. As Kevin Folta stated in the previous Talking Biotech Podcast:

"Until we have a change in how the universities and how the scientific establishment sees these efforts as positive and helpful and counts toward tenure and promotion I don't think you are going to see people jump in on this." 

Given graduate and PhD training may come with such baggage, one alternative may be to develop programs with more balance, like Professional Science Master's degrees or at least create courses or certificates that focus on translational knowledge and communication skills. Or seek out graduate study under folks like Dr. Folta who are great scientists and researchers that can also help you overcome the barriers to communicate science effectively. If that is the case we are going to need more Dr. Folta's.

References:

The Myth of the Rational Voter: Why Democracies Choose Bad Policies. Bryan Caplan. Princeton University Press. 2007.

The stoic challenge : a philosopher's guide to becoming tougher, calmer, and more resilient. William Braxton Irvine. Norton & Co. NY. 2019

The Analytics Lifecycle Toolkit: A Practical Guide for an Effective Analytics Capability. Gregory S. Nelson. 2018.

Experimentation and Causal Inference: Strategy and Innovation

Knowledge is the most important resource in a firm and the essence of organizational capability, innovation, value creation, strategy, and competitive advantage. Causal knowledge is no exception.In previous posts I have discussed the value proposition of experimentation and causal inference from both mainline and behavioral economic perspectives. This series of posts has been greatly influenced by Jim Manzi's book 'Uncontrolled: The Surprising Payoff of Trial-and-Error for Business, Politics, and Society.' Midway through the book Manzi highlights three important things that experimentation and causal inference in business settings can do:

1) Precision around the tactical implementation of strategy
2) Feedback on the performance of a strategy and refinements driven by evidence
3) Achievement of organizational and strategic alignment

Manzi explains that within any corporation there are always silos and subcultures advocating competing strategies with perverse incentives and agendas in pursuit of power and control. How do we know who is right and which programs or ideas are successful, considering the many factors that could be influencing any outcome of interest?  Manzi describes any environment where the number of causes of variation are enormous as an environment that has 'high causal density.' We can claim to address this with a data driven culture, but what does that mean? How do we know what is, and isn't supported by data? Modern companies in a digital age with AI and big data are drowning in data. This makes it easy to adorn rhetoric in advanced analytical frameworks. Because data seldom speaks, anyone can speak for the data through wily data story telling.  Decision makers fail to make the distinction between just having data, and having evidence to support good decisions.

As Jim Manzi and Stefan Thomke discuss in Harvard Business Review:

"business experiments can allow companies to look beyond correlation and investigate causality....Without it, executives have only a fragmentary understanding of their businesses, and the decisions they make can easily backfire."

Without experimentation and causal inference, there is know way to connect the things we do with the value created. In complex environments with high causal density, we don't know enough about the nature and causes of human behavior, decisions, and causal paths from actions to outcomes to list them all and measure and account for them even if we could agree how to measure them. This is the nature of decision making under uncertainty. But, as R.A. Fisher taught us with his agricultural experiments, randomized tests allow us to account for all of these hidden factors (Manzi calls them hidden conditionals). Only then does our data stand a chance to speak truth. Experimentation and causal inference don't provide perfect information but they are the only means by which we can begin to say that we have data and evidence to inform the tactical implementation of our strategy as opposed to pretending that we do based on correlations alone. As economist F.A. Hayek once said:

"I prefer true but imperfect knowledge, even if it leaves much undetermined and unpredictable, to a pretense of exact knowledge that is likely to be false"

In Dual Transformation: How to Reposition Today's Business While Creating the Future authors discuss the importance of experimentation and causal inference as a way to navigate uncertainty in causally dense environments in what they refer to as transformation B:

“Whenever you innovate, you can never be sure about the assumptions on which your business rests. So, like a good scientist, you start with a hypothesis, then design and experiment. Make sure the experiment has clear objectives (why are you running it and what do you hope to learn). Even if you have no idea what the right answer is, make a prediction. Finally, execute in such a way that you can measure the prediction, such as running a so-called A/B test in which you vary a single factor."

Experiments aren't just tinkering and trying new things. While these are helpful to innovation, just tinkering and observing still leaves you speculating about what really works and is subject to all the same behavioral biases and pitfalls of big data previously discussed.

List and Gneezy address this in The Why Axis:

"Many businesses experiment and often...businesses always tinker...and try new things...the problem is that businesses rarely conduct experiments that allow a comparison between a treatment and control group...Business experiments are research investigations that give companies the opportunity to get fast and accurate data regarding important decisions."

Three things distinguish experimentation and causal inference from just tinkering:

1) Separation of signal from noise (statistical inference)

2) Connecting cause and effect  (causal inference)

3) Clear signals on business value that follows from 1 & 2 above

Having causal knowledge helps identify more informed and calculated risks vs. risks taken on the basis of gut instinct, political motivation, or overly optimistic and behaviorally biased data-driven correlational pattern finding analytics. 

Experimentation and causal inference add incremental knowledge and value to business. No single experiment is going to be a 'killer app' that by itself will generate millions in profits. But in aggregate the knowledge created by experimentation and causal inference probably offers the greatest strategic value across an enterprise compared to any other analytic method.

As discussed earlier, experimentation and causal inference creates value by helping manage the knowledge problem within firms, it's worth repeating again from List and Gneezy:

"We think that businesses that don't experiment and fail to show, through hard data, that their ideas can actually work before the company takes action - are wasting their money....every day they set suboptimal prices, place adds that do not work, or use ineffective incentive schemes for their work force, they effectively leave millions of dollars on the table."

As Luke Froeb writes in Managerial Economics, A Problem Solving Approach (3rd Edition):

"With the benefit of hindsight, it is easy to identify successful strategies (and the reasons for their success) or failed strategies (and the reason for their failures). It's much more difficult to identify successful or failed strategies before they succeed or fail."

Again from Dual Transformation:

"Explorers recognize they can't know the right answer, so they want to invest as little as possible in learning which of their hypotheses are right and which ones are wrong"

Experimentation and causal inference offer the opportunity to test strategies early on a smaller scale to get causal feedback about potential success or failure before fully committing large amounts of irrecoverable resources. They allow us to fail smarter and learn faster. Experimentation and causal inference play a central role in product development, strategy, and innovation across a range of industries and companies like Harrah's casinos, Capital One, Petco, Publix, State Farm, Kohl's, Wal-Mart, and Humana who have been leading in this area for decades in addition to new ventures like Amazon and Uber. 

"At Uber Labs, we apply behavioral science insights and methodologies to help product teams improve the Uber customer experience. One of the most exciting areas we’ve been working on is causal inference, a category of statistical methods that is commonly used in behavioral science research to understand the causes behind the results we see from experiments or observations...Teams across Uber apply causal inference methods that enable us to bring richer insights to operations analysis, product development, and other areas critical to improving the user experience on our platform." - From: Using Causal Inference to Improve the Uber User Experience (link)

Economist Joshua Angrist explains about his students that have went on to work for companies like Amazon: "when I ask them what are they up to they say...we're running experiments."

Achieving the greatest value from experimentation and causal inference requires leadership commitment.  It also demands a culture that is genuinely open to learning through a blend of trial and error, data driven decision making informed by theory and experiments, and the infrastructure necessary for implementing enough tests and iterations to generate the knowledge necessary for rapid learning and innovation. It requires business leaders, strategists, and product managers to think about what they are trying to achieve and asking causal questions to get there (vs. data scientists sitting in an ivory tower dreaming up models or experiments of their own). The result is a corporate culture that allows an organization to formulate, implement, and modify strategy faster and more tactfully than others.

See also:
Experimentation and Causal Inference: The Knowledge Problem
Experimentation and Causal Inference: A Behavioral Economics Perspective
Statistics is a Way of Thinking, Not a Box of Tools

Experimentation and Causal inference: A Behavioral Economic Perspective

In my previous post I discussed the value proposition of experimentation and causal inference from a mainline economic perspective. In this post I want to view this from a behavioral economic perspective. From this point of view experimentation and causal inference can prove to be invaluable with respect to challenges related to overconfidence and decision making under uncertainty.

Heuristic Data Driven Decision Making and Data Story Telling

In a fast paced environment, decisions are often made quickly and often based on gut decisions. Progressive companies have tried as much as possible to leverage big data and analytics to be data driven organizations. Ideally, leveraging data would help to override biases and often gut instincts and ulterior motives that may stand behind a scientific hypothesis or business question. One of the many things we have learned from behavioral economics is that humans tend to over interpret data into unreliable patterns that lead to incorrect conclusions. Francis Bacon recognized this over 400 years ago:

"the human understanding is of its own nature prone to suppose the existence of more order and regularity in the world than it finds" 

Anyone can tell a story with data. And with lots of data a good data story teller can tell a story to support any decision they want, good or bad. Decision makers can be easily duped by big data, ML, AI, and various BI tools into thinking that their data is speaking to them. As Jim Manzi and Stefan Thomke state in Harvard Business Review in the absence of experimentation and causal inference

"executives end up misinterpreting statistical noise as causation—and making bad decisions"

Data seldom speaks, and when it does it is often lying. This is the impetus behind the introduction of what became the scientific method. The true art and science of data science is teasing out the truth, or what version of truth can be found in the story being told. I think this is where experimentation and causal inference are most powerful and create the greatest value in the data science space. John List and Uri Gneezy discuss this in their book 'The Why Axis.' 

"Big data is important, but it also suffers from big problems. The underlying approach relies heavily on correlations, not causality. As David Brooks has noted, 'A zillion things can correlate with each other depending on how you structure of the data and what you compare....because our work focuses on field experiments to infer causal relationships, and because we think hard about these causal relationships of interest before generating the data we go well beyond what big data could ever deliver."

Decision Making Under Uncertainty, Risk Aversion, and The Dunning-Kruger Effect

Kahneman (in Thinking Fast and Slow) makes an interesting observation in relation to managerial decision making. Very often managers reward peddlers of even dangerously misleading information (data charlatans) while disregarding or even punishing merchants of truth. Confidence in a decision is often based more on the coherence of a story than the quality of information that supports it. Those that take risks based on bad information, when it works out, are often rewarded. To quote Kahneman:

"a few lucky gambles can crown a reckless leader with a Halo of prescience and boldness"

The essence of good decision science it understanding and seriously recognizing risk and uncertainty. As Kahneman discusses in Thinking Fast and Slow, those that often take the biggest risks are not necessarily any less risk averse, they simply are often less aware of the risks they are actually taking.  This leads to overconfidence and lack of appreciation for uncertainty, and a culture where a solution based on pretended knowledge is often preferred and even rewarded. Its easy to see how the Dunning-Kruger effect would dominate. This feeds a viscous cycle that leads to collective blindness toward risk and uncertainty. It leads to taking risks that should be avoided in many cases, and prevents others from considering smarter calculated risks.  Thinking through an experimental design (engaging Kahneman's system 2) provides a structured way of thinking about business problems and all the ways our biases and the data can fool us..  In this way experimentation and causal inference can ensure a better informed risk appetite to support decision making.

Just as rapid cycles of experiments in a business setting can aid in the struggle with the knowledge problem, experimentation and causal inference can aid us in our struggles with biased decision making and biased data.  Data alone doesn't make good decisions because good decisions require something outside the data. Good decision science leverages experimentation and causal inference that brings theory and subject matter expertise together with data so we can make better informed business decisions in the face of our own biases and the biases in data.

A business culture that supports risk taking coupled with experimentation and causal inference will come to value a preferred solution over pretended knowledge. That's valuable. 

Read Part 3: Strategy and Tactics

See also:

The Value of Business Experiments: The Knowledge Problem

Statistics is a Way of Thinking, Not a Box of Tools

Experimentation and Causal Inference Meet the Knowledge Problem

Why should firms leverage experimentation and causal inference? With recent advancements in computing power and machine learning, why can't they simply base all of their decisions on predictions or historical patterns discovered in the data using AI?  Perhaps statisticians and econometricians and others have a simple answer. The kinds of learnings that will help us understand the connections between decisions and the value we create require understanding causality. This requires something that may not be in the data to begin with. Experimentation and causal inference may be the best (if not the only) way of answering these questions. In this series of posts I want to focus on a number of fundamental reasons that experimentation and causal inference are necessary in business settings from the perspective of both mainline and behavioral economics:

Part 1: The Knowledge Problem
Part 2:  Behavioral Biases
Part 3:  Strategy and Tactics

In this post I want to discuss the value of experimentation and causal inference from a basic economic perspective. The fundamental problem of economics, society, and business is the knowledge problem. In his famous 1945 American Economic Review article The Use of Knowledge in Society, Hayek argues:

"the economic problem of society is not merely a problem of how to allocate 'given resources'....it is a problem of the utilization of knowledge which is not given to anyone in its totality."

A really good parable explaining the knowledge problem is the essay I, Pencil by Leonard E. Read. The fact that no one person possesses the necessary information to make something that seems so simple as a basic number 2 pencil captures the essence of the knowledge problem.

If you remember your principles of economics, you know that the knowledge problem is solved by prices which reflect tradeoffs based on the disaggregated incomplete and imperfect knowledge and preferences of millions (billions) of individuals. Prices serve both the function of providing information and the incentives to act on that information. It is through this information creation and coordinating process that prices help solve the knowledge problem. Prices solve the problem of calculation that Hayek alluded to in his essay, and they are what coordinate all of the activities discussed in I, Pencil. 

In Living Economics: Yesterday, Today, and Tommorow by Peter J. Boettke, discusses the knowledge problem in the context of firms and the work of economist Murray Rothbard:

"firms cannot vertically integrate without facing a calculation problem....vertical integration eliminates the external market for producer goods."

 Coase, also recognized that as firms integrate to eliminate transactions costs they also eliminate the markets which generate the prices that solve the knowledge problem! This tradeoff has to be managed well or firms go out of business. In a way firms could be viewed as little islands with socially planned economies in a sea of market competition. As Luke Froeb masterfully illustrates in his text Managerial Economics: A Problem Solving Approach (3rd Ed), decisions within firms in effect create regulations, taxes, and subsidies that destroy wealth creating transactions. Managers should make decisions that consummate the most wealth creating transactions (or do their best not to destroy, discourage, or prohibit wealth creating transactions).

So how do we solve the knowledge problems in firms without the information creating and coordinating role of prices? Whenever mistakes are made, Luke Froeb provides this problem solving algorithm that asks:

1) Who is making the bad decision?
2) Do they have enough information to make a good decision?
3) Do they have the incentive to make a good decision?

In essence, in absence of prices, we must try to answer the same questions that market processes often resolve. And we could leverage experimentation and causal inference to address each of the questions above:

How do we know a decision was good or bad to begin with? 

How do we get the information to make a good decision? 

What incentives or nudges work best to motivate good decision making? 

What does failure to solve the knowledge problem in firms look like in practical terms? Failure to consummate wealth creating transactions implies money left on the table - but experimentation and causal inference can help us figure out how to reclaim some of these losses. List and Gneezy address this in The Why Axis:

"We think that businesses that don't experiment and fail to show, through hard data, that their ideas can actually work before the company takes action - are wasting their money....every day they set suboptimal prices, place adds that do not work, or use ineffective incentive schemes for their work force, they effectively leave millions of dollars on the table."

Going back to I, Pencil and Hayek's essay, the knowledge problem is solved through the spontaneous coordination of multitudes of individual plans via markets. Through a trial and error process where feedback is given through prices, the plans that do the best job coordinating peoples choices are adopted. Within firms there are often only a few plans compared to the market and these are in the form of various strategies and tactics. But as discussed in Jim Manzi's book Uncontrolled, firms can mimic this trial and error feedback process through iterative experimentation.

While experimentation and causal inference cannot perfectly emulate the same kind of evolutionary feedback mechanisms prices deliver in market competition, an iterative test and learn culture within a business may provide the best strategy for dealing with the knowledge problem. And that is one of many ways that experimentation and causal inference can create value.

Read Part 2: Behavioral Biases

See also:

Statistics is a Way of Thinking, Not a Box of Tools

Statistics is a Way of Thinking, Not Just a Box of Tools

If you have taken very many statistics courses you may have gotten the impression that it's mostly a mixed bag of computations and rules for conducting hypothesis tests or making predictions or creating forecasts. While this isn't necessarily wrong, it could leave you with the opinion that statistics is mostly just a box of tools for solving problems. Absolutely statistics provides us with important tools for understanding the world, but to think of statistics as 'just tools' can have some pitfalls (besides the most common pitfall of having a hammer and viewing every problem as a nail)

For one, there is a huge gap between the theoretical 'tools' and real world application. This gap is filled with critical thinking, judgment calls, and various social norms, practices, and expectations that differ from field to field, business to business, and stakeholder to stakeholder. The art and science of statistics is often about filling this gap. That's a stretch more than 'just tools.'

The proliferation of open source programming languages (like R and Python) and point and click automated machine learning solutions (like DataRobot and H2Oai) might give the impression that after you have done your homework in framing the business problem, data and feature engineering, then all that is left is hyper-parameter tuning and plugging and playing with a number of algorithms until the 'best' one is found. It might reduce to a mechanical (sometimes time consuming if not using automated tools) exercise. The fact that a lot of this work can in fact be automated probably contributes to the 'toolbox' mentality when thinking about the much broader field of statistics as a whole. In The Book of Why, Judea Pearl provides an example explaining why statistical inference (particularly causal inference) problems can't be reduced to easily automated mechanical exercises:

"path analysis doesn't lend itself to canned programs......path analysis requires scientific thinking as does every exercise in causal inference. Statistics, as frequently practiced, discourages it and encourages "canned" procedures instead. Scientists will always prefer routine calculations on data to methods that challenge their scientific knowledge."

Indeed, a routine practice that takes a plug and play approach with 'tools' can be problematic in many cases of statistical inference. A good example is simply plugging GLM models into a difference-in-differences context. Or combining matching with difference-in-differences. While we can get these approaches to 'play well together' under the correct circumstances its not as simple as calling the packages and running the code. Viewing methods of statistical inference and experimental design as just a box of tools to be applied to data could leave one open to the plug and play fallacy. There are times you might get by with using a flathead screwdriver to tighten up a phillips head screw, but we need to understand that inferential methods are not so easily substituted even if it looks like a snug enough fit on the surface.

Understanding the business problem and data story telling are in fact two other areas of data science that would be difficult to automate . But don't let that fool you into thinking that the remainder of data science including statistical inference is simply a mechanical exercise that allows one to apply the 'best' algorithm to 'big data'. You might get by with that for a minority set of use cases that require a purely predictive or pattern finding solution but the remainder of the world's problems are not so tractable. Statistics is about more than data or the patterns we find in it. It's a way of thinking about the data.

"Causal Analysis is emphatically not just about data; in causal analysis we must incorporate some understanding of the process that produces the data and then we get something that was not in the data to begin with." - Judea Pearl, The Book of Why

Statistics is A Way of Thinking

In their well known advanced text book "Principles and Procedures of Statistics, A Biometrical Approach", Steel and Torrie push back on the attitude that statistics is just about computational tools:

"computations are required in statistics, but that is arithmetic, not mathematics nor statistics...statistics implies for many students a new way of thinking; thinking in terms of uncertainties of probabilities.....this fact is sometimes overlooked and users are tempted to forget that they have to think, that statistics cannot think for them. Statistics can however help research workers design experiments and objectively evaluate the resulting numerical data."

At the end of the day we are talking about leveraging data driven decision making to override biases and often gut instincts and ulterior motives that may stand behind a scientific hypothesis or business question.  Objectively evaluating numerical data as Steel and Torrie put it above. But what do we actually mean by data driven decision making? Mastering (if possible) statistics, inference, and experimental design is part of a lifelong process of understanding and interpreting data to solve applied problems in business and the sciences. It's not just about conducting your own analysis and being your own worst critic, but also about interpreting, criticizing, translating and applying the work of others. Biologist and geneticist Kevin Folta put this well once in a Talking Biotech podcast:

"I've trained for 30 years to be able to understand statistics and experimental design and interpretation...I'll decide based on the quality of the data and the experimental design....that's what we do."

In 'Uncontrolled' Jim Manzi states:

"observing a naturally occurring event always leaves open the possibility of confounded causes...though in reality no experimenter can be absolutely certain that all causes have been held constant the conscious and rigorous attempt to do so is the crucial distinction between an experiment and an observation."

Statistical inference and experimental design provide us with a structured way to think about real world problems and the data we have to solve them while avoiding as much as possible the gut based data story telling that intentional or not, can sometimes be confounded and misleading. As Francis Bacon once stated:

"what is in observation loose and vague is in information deceptive and treacherous"

Statistics provides a rigorous way of thinking that moves us from mere observation to useful information.

*UPDATE: Kevin Gray wrote a very good article that really gets at the spirit of a lot of what I wanted to convey in this post.

https://www.linkedin.com/pulse/statistical-thinking-nutshell-kevin-gray/

See also:

To Explain or Predict

Applied Econometrics

When Wicked Problems Meet Biased Data

In "Dissecting racial bias in an algorithm used to manage the health of populations" (Science, Vol 366 25 Oct. 2019) the authors discuss inherent racial bias in widely adopted algorithms in healthcare. In a nutshell these algorithms use predicted cost as a proxy for health status. Unfortunately, in healthcare, costs can proxy for other things as well:

"Black patients generate lesser medical expenses, conditional on health, even when we account for specific comorbidities. As a result, accurate prediction of costs necessarily means being racially biased on health."

So what happened? How can it be mitigated? What can be done going forward?

 In data science, there are some popular frameworks for solving problems. One widely known approach is the CRISP-DM framework. Alternatively, in The Analytics Lifecycle Toolkit a similar process is proposed:

(1) - Problem Framing
(2) - Data Sense Making
(3) - Analytics Product Development
(4) - Results Activation

The wrong turn in Albuquerque here may have been at the corner of problem framing and data understanding or data sense making.

The authors state:

"Identifying patients who will derive the greatest benefit from these programs is a challenging causal inference problem that requires estimation of individual treatment effects. To solve this problem health systems make a key assumption: Those with the greatest care needs will benefit the most from the program. Under this assumption, the targeting problem becomes a pure prediction public policy problem."

The distinctions between 'predicting' and 'explaining' have been made in the literature by multiple authors in the last two decades. The problem with this substitution has important implications. To quote Galit Shmueli:

"My thesis is that statistical modeling, from the early stages of study design and data collection to data usage and reporting, takes a different path and leads to different results, depending on whether the goal is predictive or explanatory."

Almost a decade before, Leo Brieman encouraged us to think outside the box when solving problems by considering multiple approaches:

"Approaching problems by looking for a data model imposes an a priori straight jacket that restricts the ability of statisticians to deal with a wide range of statistical problems. The best available solution to a data problem might be a data model; then again it might be an algorithmic model. The data and the problem guide the solution. To solve a wider range of data problems, a larger set of tools is needed."

A number of data analysts today may not be cognizant of the differences in predictive vs explanatory modeling and statistical inference. It may not be clear to them how that impacts their work. This could be related to background, training, or the kinds of problems they have worked on given their experience.  It is also important that we don't compartmentalize so much that we miss opportunities to approach our problem from a number of different angles (Leo Breiman's 'straight jacket') This is perhaps what happened in the Science article, once the problem was framed as a predictive modeling problem other modes of thinking may have shut down even if developers were aware of all of these distinctions.

The take away is that we think differently when doing statistical inference/explaining vs. predicting or doing machine learning. Making the substitution of one for the other impacts the way we approach the problem (things we care about, things we consider vs. discount etc.) and this impacts the data preparation, modeling, and interpretation.

For instance, in the Science article, after framing the problem as a predictive modeling problem, a pivotal focus became the 'labels' or target for prediction.

"The dilemma of which label to choose relates to a growing literature on 'problem formulation' in data science: the task of turning an often amorphous concept we wish to predict into a concrete variable that can be predicted in a given dataset."

As noted in the paper 'labels are often measured with errors that reflect structural inequalities.'

Addressing the issue with label choice can come with a number of challenges briefly alluded to in the article:

1) deep understanding of the domain - i.e subject matter expertise
2) identification and extraction of relevant data - i.e. data engineering and data governance
3) capacity to iterate and experiment - i.e. understanding causality, testing and measurement strategy

Data science problems in healthcare are wicked problems defined by interacting complexities with social, economic, and biological dimensions that transcend simply fitting a model to data. Expertise in a number of disciplines is required.

Bias in Risk Adjustment

In the Science article, the specific example was in relation to predictive models targeting patients for disease management programs. However, there are a number of other predictive modeling applications where these same issues can be prevalent in the healthcare space.

In Fair Regression for Health Care Spending, Sherri Rose and Anna Zink discuss these challenges in relation to popular regression based risk adjustment applications. Aligning with the analytics lifecycle discussed above, they point out there are several places where issues of bias can be addressed including pre-processing, model fitting, and post processing stages of analysis. In this article they focus largely on the modeling stage leveraging a number of constrained and penalized regression algorithms designed to optimize fairness. This work looks really promising, but the authors point out a number of challenges related to scalability and optimizing fairness across a number of metrics or groups.

Toward Causal AI and ML

Previously I referenced Galit Shmueli's work that discussed how differently we approach and think about predictive vs explanatory modeling. In the Book of Why, Judea Pearl discusses causal inferential thinking:

"Causal Analysis is emphatically not just about data; in causal analysis we must incorporate some understanding of the process that produces the data and then we get something that was not in the data to begin with." 

There is currently a lot of work fusing machine learning and causal inference that could create more robust learning algorithms. For example, Susan Athey's work with causal forests, Leon Bottou's work related to causal invariance, and Elias Barenboim's work on the data fusion problem.  This work, including the kind of work mentioned before related to fair regression will help inform the next generation of predictive modeling, machine learning, and causal inference models in the healthcare space that hopefully will represent a marked improvement over what is possible today.

However, we can't wait half a decade or more while the theory is developed and adopted by practitioners. In the Science article, the authors found alternative metrics for targeting disease management programs besides total costs that calibrate much more fairly across groups. Bridging the gap in other areas will require a combination of awareness of these issues and creativity throughout the analytics product lifecycle. As the authors conclude:

"careful choice can allow us to enjoy the benefits of algorithmic predictions while minimizing the risks."

References and Additional Reading:

This paper was recently discussed on the Casual Inference podcast.

Annual Review of Public Health 2020 41:1

Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell 1, 206–215 (2019) doi:10.1038/s42256-019-0048-x

Breiman, Leo. Statistical Modeling: The Two Cultures (with comments and a rejoinder by the author). Statist. Sci. 16 (2001), no. 3, 199--231. doi:10.1214/ss/1009213726. https://projecteuclid.org/euclid.ss/1009213726

Shmueli, G., "To Explain or To Predict?", Statistical Science, vol. 25, issue 3, pp. 289-310, 2010.

Fair Regression for Health Care Spending. Anna Zink, Sherri Rose. arXiv:1901.10566v2 [stat.AP]

Wicked Problems and The Role of Expertise and AI in Data Science

In 2018, an article in Science characterized the challenge of pesticide resistance as a wicked problem:

“If we are to address this recalcitrant issue of pesticide resistance, we must treat it as a “wicked problem,” in the sense that there are social, economic, and biological uncertainties and complexities interacting in ways that decrease incentives for actions aimed at mitigation.”

In graduate school, I worked on this same problem, attempting to model the social and economic systems with game theory and behavioral economics and capturing biological complexities leveraging population genetics. 

Wicked vs. Kind Environments

In data science, we also have 'wicked' learning environments in which we try to train our models. In the EconTalk podcast with Russ Roberts, Mastery, Specialization, and Range, David Epstein discusses wicked and kind learning environments:

"The way that chess works makes it what's called a kind learning environment. So, these are terms used by psychologist Robin Hogarth. And what a kind learning environment is, is one where patterns recur; ideally a situation is constrained--so, a chessboard with very rigid rules and a literal board is very constrained; and, importantly, every time you do something you get feedback that is totally obvious...you see the consequences. The consequences are completely immediate and accurate. And you adjust accordingly. And in these kinds of kind learning environments, if you are cognitively engaged you get better just by doing the activity."

"On the opposite end of the spectrum are wicked learning environments. And this is a spectrum, from kind to wicked. Wicked learning environments: often some information is hidden. Even when it isn't, feedback may be delayed. It may be infrequent. It may be nonexistent. And it maybe be partly accurate, or inaccurate in many of the cases. So, the most wicked learning environments will reinforce the wrong types of behavior."

As discussed in the podcast, many problems fall within some spectrum ranging between very kind environments like Chess to more complex environments like self driving cars or medical diagnosis. What do experts have to offer where AI/ML falls short? The type of environment determines to a great extent the scope of disruption we might be able to expect from AI applications.

The Role of Human Expertise

In Thinking Fast and Slow, Kahneman discusses two conditions for acquiring skill:

1) an environment that is sufficiently regular to be predictable

2) an opportunity to learn these regularities through prolonged practice

This sounds a lot like the 'kind' environments discussed above. Based on research by Robin Hogarth, Kahneman also makes these distinctions describing 'wicked' environments as those environments in which those with expertise are likely to learn the wrong lessons from experience. The problem is that with wicked environments, experts often default to heuristics which can lead to wrong conclusions. Even if aware of these biases, social norms often nudge experts into the wrong direction. Kahneman gives an example involving physicians:

"Generally it is considered a weakness and a sign of vulnerability for clinicians to appear unsure. Confidence is valued over uncertainty and there is a prevailing censure against disclosing uncertainty to patients...acting on pretended knowledge is often the preferred solution."

This likely explains many of the mistakes and low value care that are problematic with healthcare delivery as well as dissatisfaction with both the quality and costs of healthcare. How many of us want our physicians to pretend to know what they are talking about? On the other hand, how many people are willing to accept an answer from their physician that rhymes with "let me look this up and get back to you later." 

One advantage AI may have over experts in kind environments is as Kahneman puts it, the opportunity to learn through prolonged practice. Machine learning can handle many more training examples than a human so to speak.

Even in kind environments, an expert may swing and miss when dealing with cases where the correct decision is like a pitch straight over the plate. One reason Kahneman discusses in Thinking Fast and Slow is the idea of 'ego' depletion. This is related to the idea that mental energy can become exhausted after significant exertion. As self-control breaks down, its easy to default to heuristics and biases that can lead to decisions that look like careless mistakes. This would certainly apply to physicians given the number of stories we hear about burnout in the profession. 

The solution seems to be what polymath economist Tyler Cowen suggested several years ago in the econtalk podcast discussion he had about his book Average is Over with Russ Roberts:

"I would stress much more that humans can always complement robots. I'm not saying every human will be good at this. That's a big part of the problem. But a large number of humans will work very effectively with robots and become far more productive, and this will be one of the driving forces behind that inequality."

Imagine the clinical situation where a physician's 'ego' is substantially depleted from a difficult case. They could then lean on AI to prevent mistakes treating more routine decisions that follow. Or perhaps leveraging AI tools, a clinician could conserve additional mental energy throughout the day so that they are less likely to default to heuristics when they encounter more complex issues. The way this synergy materializes is uncertain, but it will certainly continue to involve substantial expertise on the part of many professionals going forward. Together human expertise and AI might have the greatest chance tackling the most wicked problems.

References:

Wicked evolution: Can we address the sociobiological dilemma of pesticide resistance? | Science  https://science.sciencemag.org/content/360/6390/728.full

Thinking Fast and Slow. Daniel Kahneman. 2011

EconTalk:David Epstein on Mastery, Specialization, and Range
https://www.econtalk.org/david-epstein-on-mastery-specialization-and-range/

EconTalk: Tyler Cowen on Inequality, the Future, and Average is Over
https://www.econtalk.org/tyler-cowen-on-inequality-the-future-and-average-is-over/

The Multiplicity of Data Science

There was a really good article on LinkedIn some time ago regarding how Airbnb classifieds its data science roles: https://www.linkedin.com/pulse/one-data-science-job-doesnt-fit-all-elena-grewal/

"The Analytics track is ideal for those who are skilled at asking a great question, exploring cuts of the data in a revealing way, automating analysis through dashboards and visualizations, and driving changes in the business as a result of recommendations. The Algorithms track would be the home for those with expertise in machine learning, passionate about creating business value by infusing data in our product and processes. And the Inference track would be perfect for our statisticians, economists, and social scientists using statistics to improve our decision making and measure the impact of our work."

I think this helps tremendously to clarify thinking in this space.

Performance of Machine Learning Models on Time Series Data

In the past few years there has been an increased interest among economists in machine learning. For more discussion see here, here, here, here, here, here, here,  and here.  See also Mindy Mallory's recent post here.

While some folks like Susan Athey are beginning to develop the theory to understand how machine learning can contribute to causal inference, it has carved out a niche in the area of prediction. But what about times series analysis and forecasting?

That is a question taken up by authors this past March in an interesting paper (Statistical and Machine Learning forecasting methods: Concerns and ways forward). They took a good look at the performance of popular machine learning algorithms relative to traditional statistical time series approaches. The authors found that traditional approaches including exponential smoothing and econometric time series approaches out performed algorithmic approaches from machine learning across a number of model specifications, algorithms, and time series data sources.

Below are some interesting excerpts and takeaways from the paper:

When I think of time series methods, I think of things like cointegration, stationarity, autocorrelation, seasonality, auto-regressive conditional heteroskedasticity etc. (I recommend Mindy Mallory's posts on time series here)

Hearing so much about the ability of some machine learning approaches (like deep learning) to mimick feature engineering, I wondered how well algorithmic approaches would handle these issues in time series applications. The authors looked at some of the previous literature in relation to this:

"In contrast to sophisticated time series forecasting methods, where achieving stationarity in both the mean and variance is considered essential, the literature of ML is divided with some studies claiming that ML methods are capable of effectively modelling any type of data pattern and can therefore be applied to the original data [62]. Other studies however, have concluded the opposite, claiming that without appropriate preprocessing, ML methods may become unstable and yield suboptimal results [28]."

One thing about this paper, as I read it, is that it does not take an adversarial or luddite tone toward machine learning methods in favor of more traditional approaches. While they found challenges related to predictive accuracy, they seemed to proactively look deeper to understand why ML algorithms performed the way they did and how to make ML approaches better at time series.

One of the challenges with ML, even with crossvalidation was overfitting and confusion of signals, patterns, and noise in the data:

"An additional concern could be the extent of randomness in the series and the ability of ML models to distinguish the patterns from the noise of the data, avoiding over-fitting....A possible reason for the improved accuracy of the ARIMA models is that their parameterization is done through the minimization of the AIC criterion, which avoids over-fitting by considering both goodness of fit and model complexity."

They also recommend instances where ML methods may offer advantages:

"even though M3 might be representative of the reality when it comes to business applications, the findings may be different if nonlinear components are present, or if the data is being dominated by other factors. In such cases, the highly flexible ML methods could offer significant advantage over statistical ones"

It was interesting that basic exponential smoothing approaches outperformed much more complicated ML methods:

"the only thing exponential smoothing methods do is smoothen the most recent errors exponentially and then extrapolate the latest pattern in order to forecast. Given their ability to learn, ML methods should do better than simple benchmarks, like exponential smoothing."

However the authors note it is often the case that smoothing methods can offer advantages over more complex econometric time series as well (i.e. ARIMA, VAR, GARCH etc.)

Toward the end of the paper the authors go on to discuss in detail the differences in the domains where we have seen a lot of success in machine learning (speech and image recognition, games, self driving cars etc. ) vs. time series and forecasting applications.

In table 10 of the paper, they drill into some of these specific differences and discuss structural instabilities related to time series data, how the 'rules' change and how forecasts themselves can influence future values, and how this kind of noise might be hard for ML algorithms to capture.

This paper is definitely worth going through again and one to keep in mind if you are about to embark on an applied forecasting project.

Reference: 

Makridakis S, Spiliotis E, Assimakopoulos V (2018) Statistical and Machine Learning forecasting methods: Concerns and ways forward. PLoS ONE 13(3): e0194889. https://doi.org/10.1371/journal.pone.0194889

See also Paul Cuckoo's LinkedIn post on this paper: https://www.linkedin.com/pulse/traditional-statistical-methods-often-out-perform-machine-paul-cuckoo/