Cause-alities

The Last Post . . .

2013-04-01T14:13:14-06:00

I've really enjoyed writing the Cause-alities blog but the time has come to move on. I hope to see you over at the DecisioTech Blog!

Thanks!

Big Data, Analytics, And Storytelling

2011-03-01T14:25:53-07:00

“I think storytelling is becoming one of the new frontiers,” said Luke Lonergan, co-founder of Greenplum, now part of EMC Corp. But beyond that, “it really matters a lot to bring the brain to the problem in a way that you can untangle the complexities.” "Social Media, Genomics Driving Data Tsunami" Wall Street Journal 18 Feb 2011 http://on.wsj.com/g9Lt5A

I've found many times that it is very difficult for audiences to use and consume analysis -- no matter how insightful it might be. I suspect this is why effective analysts always find and present "the story" that the data tells. Audiences, especially many decision makers, simply glaze over when presented with the details of a complex analysis. But presented as a story they can interact, explore, test, and consume the analysis. Used correctly, storytelling can be the common language for both consumers of analytics and the those that truly revel in the abstractions of the analysis process.

There is a measure of irony in Lonergan's comment about storytelling being a "new frontier" since it has to be one of the most ancient and powerful modes of human thinking and communication. I'm guessing he means storytelling as a means to facilitate the application of big data (I don't know Lonergan, although I'd like to, so all I can do is guess) and that would be a new, but not unprecedented, application for storytelling.

I think that storytelling is more than a communication mechanism -- something that we think about after the analysis is complete. Storytelling can provide an analytic framework. As I read the interesting WSJ blog post and got to Lonergan's quote at the end I was prompted to describe some of my thinking about the relationship of storytelling and analytics and explore some ideas about how it might be relevant to the promise of "Big Data".

This is not a meta-physical argument. As I'm sure you remember from your literature class (You did take one, right? You do remember it don't you?) a story has plot, characters, and a narrative point of view. Let's see how these concepts are relevant in data driven analysis.

Aristotle argues that a plot has "a beginning, middle, and an end, and the events of the plot must causally relate to one another as being either necessary, or probable." In other words, the "analytic story" has to describe events over time that are driven by cause and effect relationships. To say we know the story the data tells means we know the cause-and-effect framework that has caused the events to transpire. Or, if this is a story of the future, then we assert that we know the cause-and-effect relationships that will cause events to happen at some future time. Knowing these relationships means that we have synthesized a set of rules that can describe events over time -- we have created a model. Generally by exploring (analyzing) the data we first develop a "mental model" and sometimes we go on to describe an explicit, shareable, simulation model. If you are a reader of this blog then you know that this is a pretty exact description of a system simulation model and its creation.

To describe cause-and-effect relationships in stories we generally need characters with motivations or purpose that act based on the events. To tell a story of consumer choice our analysis needs some characters -- consumers, manufacturers, retailers for example. And a series of events -- purchases, inventory orders, etc. And the set of cause-and-effect physics that ties this all together. In the system simulation modeling world we usually call characters "actors" or "agents". The modeling paradigm that places characters foremost is called "agent based modeling" for example.

And that leads to narrative point of view, or "narrative mode", meaning the methods used to tell the story. The analogy in the analytics world is the modeling paradigm and tool set we choose to describe the characters and plot. Existing system simulation tools are oriented towards "data poor" or, at best, moderately "data rich" environments. Certainly nothing like the developing "data tsunami" world we are headed into. Tools powerful enough to model the stories buried in the "tsunami" of data that is beginning to be available truly are on the cutting edge -- and I suspect that is exactly what Lonergan meant.

In a future post I'll try to describe what those tools might look like.

Outcome Maps

2011-02-03T10:23:41-07:00

There are a lot of reasons to create system simulation models. Many efforts start by simply wanting to understand what is causing some situation to develop; or, just the desire to understand how things work. In these cases a simulation model becomes a rich and transparent cause-and-effect hypothesis. Now, let me observe that having a solid understanding of how your business (or whatever you are exploring) works, its driving structure, and the baseline values of its parameters is a basic and broadly useful result in and of itself. One that is surprisingly rare.

However, in this "what have you done for me lately" world, inevitably, the "so what?" question comes up. As in, "So you have a simulation model . . . so what?" Because, as soon as a basis for system understanding has been established, we want to improve, control, change, the system. We want to make insightful resource allocation decisions. So, as I've discussed many times in this blog, it's usually not enough to build a system model simply to know how things work -- we need to think about how to harness it to do useful work.

This is trickier than it might first appear. Commonly, the initial approach runs along the classical scientific reductionist line: "Now that we have a model that predicts the future we simply act in accordance with that insight." This is so common it has a name: The predict-and-act decision framework.

In a very real sense, however, system models don't predict the future. They describe the cause-and-effect physics that connect our actions with assumptions about the future that lie outside our control. They describe the rules that allow us to "shape" but not dictate the future.

As the saying goes, this is not a bug, it's a feature. Because, shocking as it might appear, good decision making does not require that we predict the future. Good decision making requires that we understand the implications of our actions. System modeling is a practical way to differentiate the implications of our decisions from uncertain factors that are out of the sphere of our influence. And by doing so we gain deep insight into both.

Working with my clients I've created a visualization that helps them put their system model to use in a decision making environment. I call it an "Outcome Map". I've drawn heavily on work from "Real Options" and "Robust Decision Making" and married it to system simulation. Take a look at this Prezi to learn more.

Some Prezi Hints: After you fire up the Prezi, use the "more" menu to switch to full screen mode. Advance the presentation using the "next" arrow at the bottom. After seeing the presentation explore the canvas using the pan (left click and drag) and zoom (scroll wheel).

It's Fact-Based AND Quantitative

2010-07-26T15:16:49-06:00

As one would expect I spend a lot of time describing how system modeling works as a problem solving approach. My usual description -- in fact the one that I wrote again this morning -- goes something like this:

"System Modeling works by explicitly mapping the causal drivers that link today's resource allocations (your management decisions) to future outcomes. The technique provides a fact-based, quantitative, and transparent basis for management policy development."

Since we're describing simulation models that run on a computer it's easy to assume that all of the "facts" in the simulation are quantitative as they appear to be in a spreadsheet. But in a systems model that's not really true. The "non-quantitative facts" identify people and things in the system and, crucially, the logical relationships between those things. They describe the "physics" of the system. Things like "We have to provide a quote to the prospect before they can buy it". Or that "I have to build a widget before I can put it in inventory". Or that "I have to ship it from inventory to get it to the customer". Sometimes the physics are about human behavior: "If supply is constrained I need to accelerate ordering" is an all time favorite of mine.

Maybe this seems simple and obvious. But the sum of all of these relationships is often complex and (this is important) can feed back onto itself in a feedback loop. Also, these facts may be well known to the players in the system but they are seldom written down anywhere and are therefore "implicit" knowledge.

Finally, I don't know of any other modeling or problem-solving approach that offers to capture these causal relationships, marry them to the quantitative data (eg: how long does that take? How many of those things are there?) so that potential management policies can be evaluated in light of "all of the facts".

Why Model?

2009-08-18T12:17:29-06:00

I recently ran onto a short paper (speakers notes, really) by Joshua M. Epstein titled "Why Model?" I spend a lot of my life answering that question and I am excited by Epstein's concise, reasoned explanation. He boils it right down to the basics:

We're all modelers, but most of our models are implicit, not explicit.
Sometimes we model to predict.
Sometimes we model to explain.
And there are at least 15 other good reasons to build explicit models . . .

In some sense Epstein's position on modeling is a presentation of the scientific worldview and its moral advantages. So, mixed in with some really concrete reasons (eg: #2 -- Guide data collection) are some seemingly more esoteric objectives (eg: #6 -- Promote a scientific habit of mind).

Alas, business, financial, and other organizational leaders are mostly not swayed by a "scientific approach". I find that business and organizational clients generally need an additional level of motivation to justify an investment in explicit modeling. Usually, for the business person it is not enough to believe that an investment in explicit modeling will accomplish any of Epstein's 16 reasons. The business person wants to know What Then? Often phrased as a somewhat derisive "So What?"

For most business leaders explicit modeling has to be linked to some decision making or problem solving process. And, unfortunately, this often boils down to a focus on prediction at the expense of the other 16 reasons that are also part of excellent decision making and problem solving.

If you like this paper by Epstein try his book Generative Social Science: Studies in Agent-Based Computational Modeling. It's one of my favorites.

Decisio Described Concisely

2009-01-22T15:06:31-07:00

I was having coffee with a colleague recently and he challenged me to be much more concise and concrete about what makes Decisio's approach to modeling, simulation, and analysis novel and valuable. In response I've boiled Decisio's proposition down to four dimensions. In this post I am going to try to summarize these as concisely as possible. In future posts I'll elaborate each one and present some concrete examples.

The first dimension is purpose. Modeling and simulation is used for a wide array of purposes. Decisio uses modeling, simulation, and visualization to support decision making and problem solving.

The second dimension is application. Models don't make decisions, people do. People do it by developing a belief about how the relevant part of the world works and then acting on this belief. I call this process sense-making. It can be very hard for people to accomplish it successfully when faced with difficult circumstances. Decisio uses modeling and simulation to accelerate and improve people's ability to make sense when faced with unique, unfamiliar, complex, and ambiguous circumstances.

The third dimension is modeling technology and approach. Dynamic systems models are uniquely capable of describing the time based interactions of relevant actors and processes, of which there may be many. The ability to describe all of the relevant structure and resulting behavior over time makes systems models the tool of choice to support the sensemaking process. Agent based modeling and system dynamics modeling techniques form the core technical approaches.

The fourth dimension is managing and accumulating knowledge. Explicit systems models are artifacts that become the means to share understanding between individuals and groups, support collaborative development and expansion of knowledge, and support the reuse of insight.

Well, I think that is about as direct as I can make it! Let me know what you think!

Making Sense and "Barabba's Law"

2008-11-28T11:13:44-07:00

Through the years I've had the good fortune to work on several complex projects for Vince Barabba (more about Vince here and here). Vince is a true leader -- visionary, creative, and effective. One of the management approaches he often applied was the use of modeling to help him and his team understand a complex situation and make good decisions. In fact, I consider Vince the ultimate "model consumer." That is, he did not write complex systems models -- he used them and guided others in their use and interpretation. This paper provides one detailed example of how he worked.

Vince had a guiding principle in the application of complex models that became known as "Barabba's Law". Here it is:

Never Say "The Model Says"
-- Vince Barabba

I've think I've sat in a hundred meetings where we were using a systems simulation model to understand some complex, uncertain, situation and at some point someone would say -- but the model says . . . . If you where working on a project for Vince (or even if you had EVER worked on a project for Vince) then you knew it was time to pause the action and reflect about what was happening. Because as soon as those words are uttered then somebody is about to depend on the model as a literal prediction of the future instead of a tool to "make sense" of the situation to support their decision making.

I started writing about sense-making in my last post but here it is again: Making Sense is the development of situational awareness including an understanding of the future trajectory of the system.

At the time, though, I didn't spend much energy thinking about the underlying philosophy of Barabba's Law. What I observed is that forcing a different choice of language inherently guided stakeholders towards a different and more effective application of the modeling. The nature of the team discussion changed from predictive thinking towards evaluating the correctness and completeness of the underlying causal hypothesis that the model represented.

Barabba's Law closes the, often disastrous, thinking shortcut that allows leaders to abdicate responsibility for understanding the relevant system and its behavior. ( "Gee, we thought we were doing the right thing because the model said we were. . ." )

I think that one of the reasons Vince was so effective in using sophisticated models is that he instinctively understood the difference between prediction and sense making (although I never heard him use exactly those words). Through his extensive experience he understood how leaders actually make decisions and he knew how to integrate sophisticated modeling into that process. And he distilled some of that into his law.

Making Sense with Systems Science

2008-11-07T13:16:39-07:00

When I named Decisio (almost 10 years ago now!) I was casting about for a tag line that extended the "decision motif" to capture the essence of what we do. I settled on "Making Sense of the Future." My idea was (and is) that if clients are going to be able to make good decisions in complicated situations then they first had to understand that situation -- they had to "make sense" of what was happening. Then, they could use that understanding to make good decisions. The invocation of the "future" in this was intended, firstly, to suggest that comprehending the role of time is important to understand problems. Secondly, that we make decisions today in order to reap rewards in the future.

This idea of using systems modeling to "make sense" and support decision making was not and still isn't very common. There seem to be two prevailing ideas about the role of models and modeling. One common view is that they are sophisticated black box tools that consume data and produce predictions of the future. My observation is that while good models have predictive qualities the future is slippery. All models are wrong (but some are useful). Decision making based on a "forecast" mentality will not turn out well. An alternative perspective is that, since forecasting is difficult or impossible, modeling should be used for individual and organizational learning. Well, that's fine but sooner or later somebody has to make decisions!

I've recently become aware of the science and some of the research around the formal idea of "sensemaking." Gary Klein, well known in the field, describes sensemaking as "a motivated, continuous effort to understand connections (which can be among people, places, and events) in order to anticipate their trajectories and act effectively". Well, that's exactly what I help clients accomplish using systems models. In my projects the modeling activity guides an effective sensemaking process that results in high quality decisions.

Recently, I think I've been guilty of describing my work from the perspective of systems science and modeling to the detriment of the "making sense of the future" perspective. In fact, successful projects always integrate modeling with the sensemaking perspective.

I think that the intersection of systems modeling and sensemaking is not as well explored as it needs to be so I'll be blogging more about it. To read more about sensemaking in general try this wikipedia article and publications by Gary Klein and K. E. Weick.

What's System Science and Why Should We Care?

2008-09-25T13:26:41-06:00

It is very common for people to use the idea of a "system" pretty freely when discussing their ideas, projects, and problems . Alas, they often have a pretty fuzzy idea about what a system is and how that perspective can be put to work. "Systems Science" offers a concise definition of a system that is easy to contrast with traditional analytics. In this post I'd like to start at the beginning and try to create a clear mental image of what systems science is and how it provides useful insights.

Let's start by recapping the paradigm that we're all familiar with -- traditional analytics. Then I'll introduce the Systemic perspective.

Traditional analytics, from before the time of the Greek philosophers, is based on a reductionist principle. This is the idea that to understand a thing we decompose it into its constituent parts and then further decompose those sub-parts and so forth until we reach something that cannot be further decomposed -- the "atoms" of the system. This produces the familiar hierarchical tree structure that appears so often in our world and is the result of this process. The idea is that once we have the atoms figured out we can assemble them to understand how each level of the hierarchy works. This perspective is so deeply ingrained into our thinking, institutions, and science that we aren't always aware of its influence. It leads directly to the saying that a "thing is the sum of its parts."

Traditional analysis is an undeniably useful way to organize, catalog, and index. But it does not describe how things work. Just try to understand how a firm works by examining an organizational chart, for example. Or how the body works by doing an autopsy and cataloguing the organs.

Systemic analysis employs a different approach. Instead of thinking about each component as a thing, we conceive it as a process. Accordingly, each component of the system has an input, an output, and a description of how the input is transformed into the output -- its function. Components are interconnected via their inputs and outputs to form the system. And, in the systems view, hierarchy exists such that each component can be broken down into a sub-system.

The specific pattern of interconnections is called the system structure. In most interesting real-life cases what the system does, its behavior, is more sensitive to the system structure than to the function of the individual components. As a result, in a systems analysis the focus is not so much on the details of the components but on the relationships between them. Because system behavior is dependent on the connections between components as well as the component themselves it's sometimes said that "a system is more than the sum of its parts."

By applying system science we can describe useful things about how a firm, market, or organism works and how it will behave over time. This is something traditional analysis can't do. This is why systems science is interesting. If we want analysis that provides insight into how the future will emerge from the present we need to apply a systemic approach.

Fish Tails and Teenagers

2008-08-22T12:39:07-06:00

I found this New York Times story about a couple of high school students foray into genetic fingerprinting fascinating on so many levels. Here it is in a nutshell:

. . . In a tale of teenagers, sushi and science, Kate Stoeckle and Louisa Strauss, who graduated this year from the Trinity School in Manhattan, took on a freelance science project in which they checked 60 samples of seafood using a simplified genetic fingerprinting technique to see whether the fish New Yorkers buy is what they think they are getting.

They found that one-fourth of the fish samples with identifiable DNA were mislabeled . . .

As the father of a teenaged woman I know how clever and motivated these young folk can be. There is nothing they cannot do if they set their minds to it. I certainly related to one girl's father who noted this about their field technique: “It involved shopping and eating, in which they were already fluent.”

At a different level, as a consumer of a fair bit of sushi, I'm totally appalled. If you can't trust your sushi-master, who CAN you trust!?!

Finally, the usefulness of the DNA Barcoding Technique, despite its apparent limitations, is pretty impressive. I think that supermarkets should go way beyond just labeling fresh food with the origin. I want a BAR CODE that I can read with a pocket scanner to determine EXACTLY what I'm getting. Those green beans, for instance, what variety are they really?

I'm going to setup a DNA Barcoding system in my garage . . . .