Hacker News Reader

Comment by darksaints | Hacker News Reader

I’ve always been bothered by the five whys, simply because it’s such a dramatic oversimplification of cause and effect. The fatal flaw is the assumption that causality is linear, when it is more often combinatorial in nature.

To illustrate consider a drunk pedestrian jaywalking and being hit by a speeding driver who was texting while driving. Who caused the accident? They both did, together. Or maybe one of them did and the other didn’t. Or maybe the stars aligned just right and both would have been fine if they hadn’t been doing it at the exact same time?

I can’t help but feel like anybody who has found the “root cause” for any problem, their answer will reflect their biases far more than will reflect true causality. If you can come up with a single answer for a complex causality problem, something hints to me that you would probably point to the same “cause”for other problems too.

View on HN

> the fatal flaw

> dramatic oversimplification

> single answer

If someone pitched it to you that way, I agree that's a dramatic oversimplification -- not of causality, but of what's meant by the "Five Why's."

"Five Why's" is a tool, a reminder to separate the symptoms from <waving hands> whatever complex system dynamics that caused them. Nothing about that analysis has to be linear nor yield a single answer.

The key lesson is to never solve a problem on the same level it was created -- i.e., the symptoms. Take the time to think through the causality. If you can go five levels up/away from the symptoms and still find an actionable mitigation for a whole class of problems, great! More likely, you go up one level and at least you won't experience the exact same symptom twice.

> The key lesson is to never solve a problem on the same level it was created -- i.e., the symptoms. Take the time to think through the causality.

An example of how to illustrate the wrong behavior is -- for software devs -- something like a null pointer dereference. I've seen code reviews where the suggested fix is simply to guard the dereference to check if the pointer is non-null. Sure, sometimes that actually is the right fix. But a lot of time - maybe even most of the time - you need to to take a step back and ask yourself why it was null, whether it should have been and what else must change if it shouldn't have.

The big brain move is to ask yourself why you haven't enabled strict null checking yet.

Then again, NULL was a $1B mistake anyway, so why not...

reWriTE It iN RUST, bABy!

Agreed. I think some people must take the "5" portion very literally, but that's never how I've done it.

I just recently did a 5 whys session on some production data loss. We ended up with a branching tree that had 22 items and was up to 6 items deep. In the past I've ended up with bigger and deeper ones. To me, the "5" is just a reminder to really dig.

For me it's also a practice that's most valuable when used frequently. As you see similar things crop up in different sessions, that's a great sign to dig deeper. E.g., one of the leaf nodes in our recent retro was "no official process". That's deep enough for us; we'll create an agreed process for this particular thing and put it in the wiki. But if I see that a couple more times, I'll definitely drive the cause analysis deeper.

>I think some people must take the "5" portion very literally, but that's never how I've done it.

My personal experience has been anyone taking the "5" part literally is a cargo cult 5 why'er and they will not produce anything of value from the exercise.

There's a common fallacy in your example in that you're asking "who" caused the accident and not "what." This is why blameless retros are important. Whenever you're dealing with who, the responsibility is never binary. There's always some responsibility someone can take for a situation they're a part of. That's good advice for life in general.

The intention of the five whys is to find breakdowns at each level...not just the last one, and mitigate each.

This is why 5 why's is (or should) be paired with something like fishbone diagrams, or why tree's what ever to reinforce that it's not sufficient to regress along a single line - you must try to build out your causal tree out to 5 layers where possible.

Ultimately 5 why's is an arbitrary guideline that balances trying to keep you from being complacent in your analysis, with the need to actually like... stop and make a decision.

Who caused the accident isn't a problem statement. A driver hit a pedestrian is. Sure, from there maybe you have some weird bias against pedestrians, but in most cases five whys would say the root cause is driver inattention. Poor framing of the initial problem is as much an issue with the method as finding bad answers to "why".

Right, and "root cause" only has meaning in the context of some goal being achieved. If you're a detective investigating the scene of the accident to determine culpability, your goal is very different than if you're a civil engineer investigating the accident for ways to improve traffic management. It'd be really strange for either of them to conclude their report with a recommendation to increase World Health Organization's budget for alcohol abuse treatment.

Therefore, Your Honor, my client is innocent. Why has this tragic accident occurred? I shall tell you to look no further than the root cause...

The Big Bang.

> If you can come up with a single answer for a complex causality problem, something hints to me that you would probably point to the same “cause”for other problems too.

That's basically the point of the 5 whys approach. You work up the tree/into the graph of contributing factors based on the specific problem and you end up solving lots of problems. If you don't go back far enough, you're likely just putting a band-aid on the problem and some other variation of it will waste more time again. You get "5 whys" back and now you can really start figuring out what went wrong and how to fix/prevent/mitigate/avoid/etc. it for next time.

If you asked my dad to root cause literally any problem in my life, from my failed marriage all the way to a random nasty smell emanating from my kitchen, he would trace it back to the fundamental root cause that I’m no longer a believing mormon.

I don’t have any problem with the idea that you should dig deeper for any form of causal reasoning. My problem is with the inherent linear flow and singular root cause reasoning of the 5-whys methodology. It is just too easy to lead you back to the thing that you are personally biased to think anyway. If you want to solve actual problems, you more than likely have to solve multiple problems that contribute to a complex system than to solve any singular problem.

I don't mean to be snarky, but if you're looking for a linear flow or singular root cause, you're doing it wrong. It's literally just getting you in the door and past the most immediate symptoms.

If you can work your way back from a smelly kitchen to a change in religion, you ask, "Does anyone who didn't change religions have a smelly kitchen?" Obviously the answer is yes, so flag it as possibly correlated and move on to investigate factors that have a better chance at causality and aren't so easily dismissed. Or maybe you need to go even further back, and there's some link between why your kitchen smells and why you changed religions. It's just an approach, like therapy, not an algorithm.

> I don't mean to be snarky, but if you're looking for a linear flow or singular root cause, you're doing it wrong. It's literally just getting you in the door and past the most immediate symptoms.

If you read this thread there are literally multiple interpretations of what the 5-whys actually means, each of which results in a roundabout No True Scotsman argument. Including yours.

I’d prefer to see the 5-why’s as what it actually is: an aphorism about not treating symptoms but rather problems. The aphorism is useful…attempts to define it as some kind of formal tool are a waste of everybody’s time. There are plenty of tools (such as Ishikawa diagrams) out there that represent a better implementation of the aphorism, and 5-whys can be replaced entirely by them.

Is it actually that? I'm pretty sure that the term comes out of the Toyota Production System, in which kaizen, or continuous improvement, is the heart of the process. Given that the Ishikawa diagram also comes out of post-war Japan's industrial process quality improvement community, I would be very surprised if Totota didn't make use of Ishikawa diagrams in practice.

In all the five-whys investigations I’ve been a part of, it’s always been accepted that each node can have more than a single cause, and you try to explore as many of them as the time allows, and provide betterments for as many of them as possible.

I've always looked at these exercises (like the NTSB reports) as trying to find as many possible ways that this could have been prevented - not trying to find the single most proximate way to prevent it.

And in doing so, you likely realize that you came close to disaster a number of times and never realized it.

5 why's refers (afaik) to the depth of the why tree. If the first why has 3 answers, then you ask why 3 more times to each answer (n = 4, depth = 2)

If each of those 3 whys have 2 answers, you would ask why 6 more times (n=10, depth = 3)

If each of those whys have 3 answers, we would need to ask why a further 18 times! (N=28, depth = 4)

That is leaving out the last depth, but suffice to say there are a lot more than just 5 questions. It seems rare for any system where the cause and effect tree is linear. Hence, 5 whys is meant to be very holistic, not just deep.

Sure it’s combinatorial in any specific instance, but you can still identify how much each individual root cause contributed to the final outcome and then alter outcomes in aggregate that way.

For your example, if texting and driving is associated with 80% of pedestrian collisions but a drunk pedestrian is only associated with 5% of such collisions, you can probabilistically influence most potential outcomes by focusing your effort on the texting-while-driving root cause.

All models are false, some are useful. Five whys strikes a useful balance between the overly simplistic single cause model and the incomprehensible detail of full reality.

Very interesting, hadn't thought about this.

It's funny how humans try to simplify things to black and white all too often. Why do we do this? I'm sitting here trying to think why I do it, and I can't figure it out. Something maybe about needing someone to blame.

For example, when I think of the 2008 financial crisis, I always feel this need to find one group to blame, but really there wasn't one group.

Five why's states that if I have to ask "Why?" more than 5 times to get at why something is the way it is, it needs to be simplified, in the sense that lack of simplicity/excessive cognitive load is in and of itself, a form of defect.

This is not a formulation of the approach that I've ever encountered.

Because it isn't. 5 whys is for exploring the neighborhood of the problem without going deep down a rabbit hole. It's like timeboxing an activity or putting a bounds on a search algorithm. The objective is to find something near enough to act on and partially [0] address the systemic causes, but far enough to not just be the proximate cause. "Battery was dead" is a proximate cause, many people stop at proximate causes. 5 whys forces you beyond that.

It should also be combined with other techniques that elaborate, more properly, on the model of the system in which the problem occurred. There you will find the deeper systemic causes and can start addressing and mitigating the issues. But that will take a lot longer to explore and resolve than the 5 whys answer of "Get a proper tuneup every 5-10k miles".

[0] Ideally fully, but 5 whys is an expedient, non-thorough technique so that's a reach.

https://www.toolshero.com/problem-solving/5-whys-analysis/

No, you have it exactly backwards. Five whys is explicitly depth first search. It was originally employed by Toyota Motors, and the key to actually getting at what the actual causes of issues were. This allowed Japan to avoid entire classes of defect due to their acceptance and implementations of the teachings of W. Edwards Demmings. It never made the leap to the United States, because doing things right is such a foreign concept over here, where management would gasp at the concept of a single worker shutting down the entire factory upon detecting a problem, whereas in Japan, the andon was key to getting issues that would compromise the Quality of the end product addressed.

It's:

I have a problem, Why? Because of this other thing. Why? What causes other thing? The other thing is dependent on yet another thing and that thing. Are those things a problem? If so, why?

Iterate until done.

The "Five" aspect came into play in that after a certain number of "Why's", it was often the case that the fact it took so many Why's to get to the root cause, that the complexity itself constituted it's own type of defect.

I might need to revisit the book I first read of it in, believe it was titled "Elegance" or some such.

EDIT: Found it.

In Pursuit of Elegance: Why the Best Ideas Have Something Missing by Matthew E. May

EDIT2: Nope, false alarm. Drat it, I will find that reference. It's in this library somewhere... In the meantime, forget it; maybe I'm hallucinating it, because clearly I haven't refreshed it in a while, and the old wetware has taken a hell of a beating over the last couple years.

> Five whys is explicitly depth first search.

I didn't say it wasn't depth first so I have no idea what you're responding to with this. Can you point out where in my comment I said it wasn't depth first?

I just said it's bounded, because it is. Even the link you've posted states that "why?" is only asked 5 times. But even if you take the 5 as suggestive rather than a strict limit, giving a limit (even a soft one) implies a boundary to the search. If it was meant to be unbounded it would just be called "Whys". 5 pushes you beyond the proximate cause and towards the actual root cause (of course, "root cause" is itself a poor phrase as it implies a singular cause that can be addressed to prevent the problem from reoccurring), but then the root cause could be 10 or 20 or 30 whys deep. So unless you go beyond the point suggested by the name and publications about it then you won't find it.

------------

EDIT: I see you edited your comment after I hit reply and didn't mark it as such, good form.

Anyways, about it being depth first I'd actually disagree. Even if you believe that it should be taken as an unbounded search (which I also disagree with, obviously, because there are better tools than 5 whys for such a major task) it should not be an unbounded depth first search. That would be a form of malpractice.

The more accurate description would be something like iterative deepening. You go down a few layers with something approximating a depth first search, and then you back track and try the other branches. If you go down one rabbit hole arbitrarily deep you may or may not find a causal factor but you will have missed many others along the branches you chose to ignore for the sake of your unbounded depth first search (again, a form of malpractice).

Reply on news.ycombinator.com