REPLY Definition of root cause (SD6779)

[SDMAIL Jack Harich]

Posted by  Jack Harich <jack at thwink.org>

Colin Beveridge wrote:
> Jack
> My answer to your second question (below) might be rather naïve but I 
> think that finding a single root cause for a complex problem might not 
> always be possible.   

Sorry, my mistake. The last sentence in the quoted section was "For 
simplicity we say 'root cause' when actually it may be singular or 
plural." I will modify the manuscript to mention this sooner.

Done. The first paragraph in the section on step 2B now ends with "For 
simplicity we say 'root cause' when actually it is usually plural."

> A combination of asynchronous influences might contribute to a problem 
> and the asynchronicity might disguise the combination of influences to 
> the point of invisibility, defeating analysis to identify a single 
> root cause. In other words, your analysis might identify strong causal 
> candidates but no single cause actually manifests itself for long 
> enough, or indeed ever presents itself again.
>   
A really good point, that the system is changing and hence so are the 
root causes you are trying to resolve. The System Improvement Process 
accommodates this by dividing difficult social problems into three 
subproblems: overcoming change resistance, achieving proper coupling, 
and avoiding excessive model drift. If the last one, model drift, is 
solved then your solution is self-managing and self-evolving.

Thanks for pointing this out.

Bill Braun wrote:
> Jack Harich asks about the definition of root cause. Broadly stated, 
> it would be the structure that produces the reference mode up to the 
> present and, if policies were left as is, would result in the "feared" 
> mode in the future.

Bill,

Ahhh, what an expansive way to put it! How succinct.

"it would be the structure that produces the reference mode..." - Let's 
see if I understand this. It usually takes the entire model to do this. 
So the entire model is the root cause?

This seems to be the viewpoint that everything that contributes to a 
problem's symptoms is a cause, and that trying to find a so called root 
cause is counter productive. I wonder if there is a term for this school 
of thought? The Thou Shalt Have No Root Cause school? :-)


Regarding "the structure that produces the reference mode" -  It's easy 
to create a model that can create a problem's symptoms, without 
including the root causes and without including time series data.

A simple example would be a model of an organism that, once infected, 
gets sick. The model can beautifully handle the infection event, the 
illness that follows, and maybe even getting well. The first root cause 
of such a model would be the combination of an infection event and 
susceptibility to infection. But if you want to go deeper, you could ask 
WHY was the organism so susceptible? Was its immune system weak? Was 
that in turn caused by bad diet and exposure to pollutants? Etc. So 
there is a deeper root cause. Whether it's in the model or not does not 
change the reference mode behavior. It will of course change other 
scenarios.

As another example, consider World3. It roughly duplicates the 
consequences of living unsustainably. But does it explain why there is 
so much change resistance to solving the problem? Does it help us to 
understand why in 1999 the US Senate voted an amazing 95 to 0 against 
the Kyoto Protocol, even though the US had a democratic president and Al 
Gore himself was vice president? No.

As far as I can tell, World3 only identified the problem, by modeling 
the immediate causes in a convincing manner. These are the well known 
IPAT forces. The model was a tremendous contribution. The first step is 
usually the hardest. Next we need to dig quite a bit deeper and find out 
the underlying causes, the ones that explain the intermediate causes 
captured in World3. Only then will we be able to move on to resolving 
the underlying causes.

> Posted by  "Jack Homer" <jhomer at comcast.net>
>
> I don't see how Jack Harich's emphasis on root causes is helpful.  
> Trying to find the "true" original or root causes in complex social 
> systems (unlike engineered systems) is a fruitless endeavor, because 
> when it comes to human behavior among multiple stakeholders there is 
> no end to the forces at play. The best we can do is to peel the onion 
> of causation to get a few more layers down.  The way we do this is 
> with models that help us understand why current policy approaches are 
> not working or may lead to problems in the future, and why certain 
> other approaches may hold more promise.
Jack,

Sorry to cause such a ruckus. This is a great, thoughtful message. Such 
disagreement shows the field of system dynamics is at last in the early 
Model Revolution phase of the Kuhn Cycle. The cycle has five phases: 
Normal Science, Model Drift, Model Crisis, Model Revolution, and 
Paradigm Change, as described at 
http://www.thwink.org/sustain/glossary/KuhnCycle.htm.

In the Model Revolution phase, new ideas of how to most productively 
tackle a field's central problems compete against each other. As 
consensus on which of these will work develops, a new problem solving 
model emerges. For SD this model contains things like how to model most 
effectively, what principles to use where, notational standards, 
completed model standards as discussed recently, etc. A mature model 
allows the field to solve problems the previous one could not.

My humble offerings are an attempt to help build a new model for SD, one 
that is good enough to solve difficult social problems. I'm sure my work 
is chock full of errors and weakness. But because I'm an outsider and 
new to the field, such efforts have the potential to offer fresh and 
possibly more productive lines of attack.
   
Thanks. It is true that "when it comes to human behavior among multiple 
stakeholders there is no end to the forces at play." But it does not 
follow that this prevents us from finding the underlying causes of a 
problem that, when resolved, will solve the problem.

You say "Trying to find the 'true' original or root causes in complex 
social systems (unlike engineered systems) is a fruitless endeavor..." 
This says that finding root causes is easier in engineered systems. I 
agree. But that does not prevent us from finding them in non-engineered 
systems. It only makes it harder.

Is a corporation an engineered system? I think not. They are big pieces 
of putty shaped by the intuitive whims of many managers, over time in an 
evolutionary manner. But we routinely apply SD to corporate problems 
with great success. We do this by finding the hidden underlying causes 
and resolving them.


> Such models need to be able to reproduce the problem and plausibly 
> explain why it is occurring.  I believe Mr. Harich is wrong to 
> minimize the importance of this step in the process, a step which is 
> more often than not the key to identifying more effective policy.  I 
> have previously described this process of scientific investigation and 
> discovery in modeling; see "Why We Iterate", SD Review 1996.
>   
I think we are in agreement here. SIP step 2A (see below) reproduces the 
problem, by modeling the dominant feedback loops that cause the 
symptoms. Step 2B then explains why they are dominant by finding the 
root causes. Thus steps 2A and 2B are the same as "reproduce the problem 
and plausibly explain why it is occurring." I wrote that 2A is 
relatively easy and 2B is hard. Together, 2A and 2B are hard, so there 
seems to be no difference in our viewpoints. I've only decomposed mine 
differently from the way modelers usually approach their task.

The key difference may be that I'm deliberately using stronger, more 
precise language than "plausibly explain." I prefer "find the root 
cause" because that emphasizes we must dig deep to solve big hairy 
problems. This also allows a finer grained process, as the steps below 
show.

I do think finding the immediate causes is easy, compared to finding the 
root causes. That's what makes difficult problems difficult.

For reference, step 2 of the System Improvement Process (SIP) analyzes 
the problem/system until the key cause and effect relationships are 
understood. Step 2 has these five substeps:

A. Find the feedback loops that are currently dominant.
B. Find the root cause of why they are dominant.
C. Find the low leverage points and symptomatic solutions.
D. Find the feedback loops that should be dominant.
E. Find the high leverage points to make them go dominant.

I had not read your "Why We Iterate" article before. A good read. I 
learned a lot. Marvelous cases. The Lessons section was meaty and I will 
refer to them in a new project I'm starting this week. Thanks!



> Mr. Harich is under the impression that the Urban Dynamics model was 
> somehow more insightful and penetrating than the World3/World Dynamics 
> model.  I have argued previously that both of these models produced 
> striking conclusions because of their artful weaving together of key 
> feedback structures and empirical observations and evidence; see 
> "Structure, Data, and Compelling Conclusions: Notes from the Field", 
> SD Review 1997.  Both models used the some process of inquiry, both 
> peeled the onion equally deep, and both delivered compelling 
> conclusions that altered the policy debates on urban decline and on 
> global development.  
Regarding "both peeled the onion equally deep" - I think there was a 
major difference. There are several proofs:

(1) One is that the conclusions of the urban decay model allowed urban 
managers to adopt policies that were immediately effective, although the 
problem has not been completely solved. This has not happened on the 
environmental sustainability problem. 36 years after publication of LTG, 
the world is nowhere near solving the problem. No policies have been 
introduced that were immediately effective. Many have been slightly 
effective. This indicates that problem solvers have been addressing 
intermediate causes, rather than root causes. This causes them to push 
on low leverage points (LLPs) instead of HLPs.

(2) Another proof is from the analytical viewpoint of the System 
Improvement Process. Examination of the urban decay model shows it 
performed all five steps of SIP step 2. Looking at World3, I only see 
step 2A performed. Others may see this quite differently, but when I 
examine the model, I ask where are the root causes? What in the model 
explains WHY the dominant social agents are so addicted to perpetual 
growth? What in the model explains WHY the preferred, intuitive solution 
to the problem has largely been to lower impact per unit of consumption 
and continue to raise the limits? That this only postpones the day of 
reckoning and makes collapse bigger when it finally comes shows this is 
an erroneous solution. But WHY has it been followed? There is no hint of 
this in the model. And so on. Thus when problem solvers attempt to use 
World3 to solve the problem, they have no deep root causes to resolve. 
They have nothing to steer by, and are as lost as a ship at see without 
a compass.... (Whoops, getting carried away.)

(3) If we search for the root causes of the sustainability problem using 
SIP, it's possible to come to radically different conclusions than 
World3 did. (But of course the authors caution their goal was not a deep 
analysis, but to alert the world to the presence and magnitude of the 
problem.) I've been working on this problem full time since 2001. If you 
read the Dueling Loops of the Political Powerplace paper at 
http://www.thwink.org/sustain/articles/005/DuelingLoops_Paper.htm, you 
will see an example of what happens when you keep asking WHY longer than 
the rest of the herd. You keep throwing away false root causes until you 
come to ones that fully satisfy the process. On page 9 of the full 
length version the paper says:

"We now have enough pieces of the puzzle to draw an important 
conclusion: The dueling loops, their cyclic nature, the inherent 
advantage of the race to the bottom, the presence of the New Dominant 
Life Form, and its successful exploitation of the race to the bottom are 
the structural root cause of most of the stiff, prolonged resistance to 
adopting a solution to the environmental sustainability problem. 
Civilization is presently stuck in the dominant race to the bottom part 
of the cycle. Our challenge is to cause this cycle to end as soon as 
possible, and then to prevent it from ever starting again. If we can do 
that civilization will not only enter the Age of Transition to 
Sustainability. It will also enter an entirely new mode: a permanent 
race to the top among politicians, along with all that has to offer, but 
has never been achieved."

This is not a simplistic root cause. Several dozen readers have told me 
it makes complete sense and explains so much. No one has found a flaw in 
it. But even if there is a flaw (I'm sure there are at least some) the 
analysis should illustrate that it is possible to go much deeper than 
the LTG analysis went, by finding the true root causes, and then the 
LLPs and the HLPs.



> They did not attempt to peel the onion to its center, which is 
> impossible, but only deep enough to reveal underlying 
> forces--supported by the evidence at hand--that could help 
> policymakers and the general public act with greater foresight.
That's why we need a tight definition of root cause - so modelers will 
know when to stop and avoid going too far or not far enough.

My definition of root cause is "those portions of the system that most 
deeply explain why the system’s emergent behavior produces the problem 
symptoms" and "A root cause has three identifying characteristics: (1) 
It is clearly a major cause of the symptoms (2) It has no deeper cause. 
(3) It can be resolved."

Step 2B is done when you have achieved the goal of finding the root 
cause, as defined. There's no cookbook procedure to guarantee you will 
get there. But the clarity of the goal makes it much more likely you 
will arrive. (Perhaps we need to add more characteristics, so we can 
have a more reliable checklist.)

In a short paper there's little room to discuss the intricacies of 
finding the root causes. This takes a lot of iteration through steps 2 
and 3, and eventually step 4. It's these iterations that allow you to 
test whether you have found root causes that can be resolved within the 
constraints of the problem definition. If they can't, then it's back to 
step 2. I'm a believer in the power of intelligent iteration, just like 
the fellow who wrote "Why We Iterate." :-)

I will modify the manuscript to mention the importance of iteration.... 
There, done. I've added this paragraph: "It takes a lot of iteration to 
find the root causes. The steps (especially 2E and experimentation in 3) 
that follow 2B allow you to test whether you have found a satisfactory 
set of root causes and to iteratively refine them until they are correct 
enough to solve the problem."


Jack, I really hope this helps. All I'm trying to do is help our field 
zoom its way out of infancy, and on into its mature years. After all, 
there are a lot of fascinating, challenging problems just sitting out 
there, waiting for analysis and solution. Some have been waiting for a 
long, long time. A few, if they are not solved soon, have consequences 
that are too horrific to grasp....

Muchos gracias,

Jack
Posted by  Jack Harich <jack at thwink.org>
posting date  Tue, 04 Mar 2008 21:58:13 -0500