Cognitive Biases in Software Engineering

http://www.jonathanklein.net/2013/06/cognitive-biases-in-software-engineering.html

by charlieirish • 12 years ago

69 21 12 years ago

dustingetz• 12 years ago

The Bandwagon Effect

I don't know if this is a negative bias. Popular tech is well understood with known risks. Successful outcomes in software are hard enough without adding unknowns to the equation.

chriswarbo2• 12 years ago

The effect isn't referring to popular technologies remaining popular; the effect is that, when we make decisions in a group, we unconsciously tend to agree with other people more than we otherwise would.

For example, lets say you're asked to submit a survey of the suitability of a bunch of languages for a new project; say Python, Java, PHP, C# and Ruby. You may give Python 8/10, Java 7/10, PHP 6/10, C# 7/10 and Ruby 8/10.

Now, imagine that instead of that survey, you were invited to a meeting where the chief architect informs everyone that the they haven't decided whether to use Java or C# yet, and since the CEO suggested PHP they decided to open it up for comments. Would you still give the same scores?

The classic experiment for this used a social music website, where people could vote for what they liked. Unknown to the users, they were internally separated into groups, so that people in group A only saw scores based on other voters in group A, and so on. In each group, there were a couple of really high-scoring songs, with the rest scoring very badly; but the each group had chosen different songs as their favourites! Those which were regarded highly in group A were seen as terrible by group B, and vice versa. As a control, one of the groups was never shown any scores at all, so their votes were less biased. That group showed much less extreme results, with some songs scoring better than others, but not massively so.

mathattack• 12 years ago

I found that overconfidence is the worst. In part everyone thinks "I know my work, leave me alone" but interdependencies mean that even if everyone is 90% confident, large projects slip. And most people's 90% confidence is really only 50% accurate.

The first antidote is educating people about interdependencies and overconfidence. The second is introducing management.

alphagenerator• 12 years ago

I think projecting overconfidence is a social/political survival strategy. The 50% accurate thing is just the hidden reality everyone acknowledges. It's sort of like how people are polite when confronted with ugly babies.

calinet6• 12 years ago

Love that first one (attribution error, tendency to over-estimate individual influence and under-estimate systematic effect). It's one of the biggest pitfalls I've seen in software (or any industry, really, but let's talk about software).

When something goes wrong, it's so easy to pull the "accountable party" into the CTO's office and give him an earful, but it is so wrong, so ineffective, and so anti-progress it's not even funny. Not funny at all.

The correct solution is the blameless post-mortem that Etsy does, as mentioned. Kill all fear of punishment, bring out the honesty and systemic analysis, and recognize the complex factors involved to come up with a real solution.

This type of thing is exactly what's required for a successful engineering culture, and for quality to be able to flourish. Give in to your reptilian instincts to get mad at a figurehead, however, and you'll kill your company culture and your quality all at once.

afarrell3• 12 years ago

In my opinion the way around this is to deliberately stop and do an estimation exercise. First think about how long the refactor will take, and be extremely generous (e.g. double your first estimate)

That isn't estimation. That is pulling a number out of your ass. Try something for me. For each of these data points, write down a low estimate and a high estimate that give you a 90% probability of being within that range.

Surface Temperature of the Sun: Low [___] -- High [__] Latitude of Shanghai: Low [___] -- High [__] Area of the Asian continent: Low [___] -- High [__] The year of Alexander the Great's birth: Low [___] -- High [__] Total value of U.S. currency in circulation in 2004: Low [___] -- High [__] Total volume of the Great Lakes: Low [___] -- High [__] Worldwide box office receipts for the movie Titanic: Low [___] -- High [__] Total length of the coastline of the Pacific Ocean: Low [___] -- High [__] Number of Book titles published in the US since 1776: Low [___] -- High [__] Heaviest blue whale ever recorded: Low [___] -- High [__]

Now, you might think this is a silly exercise because you don't have experience with the scale of these things, but how often do you estimate the time it takes you to build a product you aren't yet familiar with something using a tool you're not yet familiar with or people you aren't familiar with?

The answers are here: http://my.safaribooksonline.com/book/software-engineering-an...

(This quiz is taken from Software Estimation by Steve McConnel (Microsoft Press, 2006) and is copyright 2006 Steve McConnel. All rights Reserved. Permission to copy this quiz is granted provided that this copyright notice is included.)

skwirl• 12 years ago

What exactly is the point you are trying to make?

afarrell3• 12 years ago

That in this particular section they are using the word "estimation" wrong in a way that is both common in the software industry and leads to failed projects.

pessimizer2• 12 years ago

Sometimes, for business reasons, it's crucial that you pull a number out of your ass. My policy is to make that number as high as will be acceptable by management, and to refuse the project if that number is less than twice the length of my worst case scenario nightmare. Then, deliver early. Basically the same as the OP.

Are you advocating something different here?

chriswarbo2• 12 years ago

That's known as the Scotty Principle, or Scotty Factor ;)

http://c2.com/cgi/wiki?ScottyFactor

afarrell3• 12 years ago

Well, I'm advocating finding a way to estimate things that doesn't involve pulling things out of your ass. Doing that has always felt a bit dishonest to me. Yet, I don't actually know any other method because I haven't finished the book.

paulhodge• 12 years ago

The negativity bias really irks me. I went through a phase in my career where I tried to communicate everything in a positive way without talking about blame or judgement. And I found that people don't respond to that as well - they are more likely to ignore you or forget what you said. But instead, if you talk about how X library sucks and Y developer did such and such wrong, people perk up their ears and listen to you, and they are more likely to start to regard you as the local expert. Seems like the effect is worse when talking to the non-technical folks.

There's definitely way too much crankiness in the software world, and I don't think it's because the people are naturally negative people. It's because we're pragmatic people and we unconsciously gravitate towards the styles of communication that get the most results.

jamesbritt• 12 years ago

Interesting. I'd gotten the impression that negative comments were a form of dick-waving. I.e., if you can spot the the numerous flaws in some tool or concept then you are clearly the superior-minded hacker.

But I've never thought about it in terms of idea adoption or whether positive comments might have a lesser effect on people.

I wonder if negative comments play off people's fear of being wrong or fear of being seen on the "losing" side.

Maybe if you hear that there are good points for both Dreamweaver and [emas|vi|sublime] and it's a matter of personal preference and what works for you, then it's just some bit of data to file away. OTOH if you hear that Dreamweaver is simply evil and profoundly flawed then your adrenaline spikes; you don't want to be one of those people who use that tool.

Kind of disappointing if true.

mreiland• 12 years ago

Because if "something sucks", you clearly are making the right decision by going with the other thing.

As opposed to worrying about whether or not your decision was the right one because all of the options had really good upsides.

ogdenyogly• 12 years ago

I have this problem. I am perceived as relentlessly negative for pointing out obvious points of failure, ridiculous management, etc. Most people just parrot what management wants instead of what they need-- good advice.

In reality I am trying to ensure the success of the project-- FAILURE IS THE DEFAULT OPTION.

mbesto• 12 years ago

Umm, how can forget the greatest cognitive bias ever in software development: http://en.wikipedia.org/wiki/Planning_fallacy

The planning fallacy is a tendency for people and organizations to underestimate how long they will need to complete a task, even when they have experience of similar tasks over-running.

superuser2• 12 years ago

My personal favorite is organizations' vast overestimations of the size/complexity and performance needs for their software projects, probably related to delusions of their own grandeur.

Rails is more than capable of serving an intranet application for your 150 employees. But nooope, we're an enterprise, so we need J2EE/Oracle.

bluesnowmonkey• 12 years ago

Poor design decisions and shortcuts are sexy because they give you a small amount of value right now (not having to do the work to architect things properly), and you dramatically discount the value you would get in the future by doing it right the first time.

What do you call this bias? People like to think that if a different person had been in charge, or if they'd had more time, then things could have been done "properly". But really there is no "proper" way to do anything in software engineering. Every decision is a judgement call. Everywhere you look it's shades of gray. Especially without the benefit of hindsight.

jdmichal• 12 years ago

Honestly, I call it "lack of proper documentation of decisions." It's pretty systemic throughout software engineering, myself included. It's why we seem doomed to "refactor" current solutions, only to end up with the same mess at the end. We see the 80% use-case, think we can do better, then by the time we hit those 20% edge cases, the code is as bad as when we started. This would be avoided if all the decisions and use-cases leading to the current mess were well documented, which would allow people to actually engineer a better solution with all the use-cases, or just leave it alone because they can't.

deciplex• 12 years ago

At larger established companies, do the refactor. At a startup that is trying to release an MVP, maybe do the hack (your code will probably be rewritten anyway). That said, always think of who else will be working on your code...

Sadly, at the larger companies, and especially ones where software is a cost center, this is basically impossible. They'd rather replace their last bullshit app with a new bullshit app, since it was the last director responsible for that POS anyway, making all the same mistakes as before, and rely on collective delusion (i.e. threats) to get everybody to go along with the idea that the old mistakes are being corrected, and that mistakes made now, if any, are at least new ones.

Maybe it's the same at startups. I wouldn't know.

Cognitive Biases in Software Engineering

Related Stories