Relying more on data to make decisions can increase susceptibility to confirmation bias. Learn what confirmation bias is and how to avoid its potentially high costs.
What is confirmation bias and why is it important?
Our brains use shortcuts to help us process information. These shortcuts, while useful, also create cognitive biases that can lead us to jump to incorrect conclusions and make suboptimal decisions.
One such bias can cloud our objectivity in acquiring and processing information, causing us to search for evidence, or favor it, in ways that tend to confirm our prior beliefs. This is confirmation bias.
When using shortcuts associated with confirmation bias, we essentially purchase the ability to make faster decisions in exchange for some possibility of making worse decisions. Inherent in this transaction is the risk that what we gain (speed, lower information costs) will turn out to be less valuable than what we gave up (the incremental benefits making a more considered decision). In some cases, we may never know how much we gave up because opportunity costs are often hidden.
Doesn’t data and modern tech make confirmation bias irrelevant?
A lot of what we do at Horizon Data Science comes down to helping teams develop and use tools to make better decisions with data. Ironically, more data can make confirmation bias worse by providing us with more ways to find the answers we favor (data mining). Beware the dashboard that has an answer for everything!
So why not let machines make the decisions? Unfortunately, hi-tech methods are no vaccine against confirmation bias. For example, supervised machine learning is only as good as its training data. If confirmation bias is present in the human-generated labels of that data, then algorithms will learn and replicate the bias.
How To rely on data while protecting Your organization from confirmation bias
Just because more data can worsen the problem doesn’t mean it isn’t valuable.
When using data for decision making, we are often doing one of two things: looking for new ideas or testing an existing idea/theory. In both cases, we should aim to use mental shortcuts optimally, accounting for the potential costs of confirmation bias. The key to doing so is to become more aware of our goals with data, acknowledge our prior beliefs and exactly how they might impact our decisions, and then develop a strategy to address this issue head-on.
Below are a few steps to help decision makers accomplish this…
Step 1: Write down your objectives
Why are you looking at data – to come up with new ideas, or to test an idea you already have? Both are perfectly valid but addressing confirmation bias requires us to be aware of our aims.
Confirmation bias adversely affects ideation by creating blinders to new/outside-the-box ideas. It impedes idea testing by skewing the questions we ask and preventing us from marshalling the full potential of our data to challenge prior beliefs.
Step 2: Acknowledge your beliefs and the decision that they tend to lead you towards
Ask yourself what your prior belief is. If you had to place a large bet on the outcome, what would it be? How confident in that bet would you be? What is that confidence based on?
“Gut instinct” is not a wrong answer here – again, such instincts often help us humans respond to the world more efficiently. Our goal here is not to remove instinct from the equation. Instead, it’s to acknowledge our prior beliefs so that we can understand their implications and appropriately challenge them in cases where the costs of being led astray are high.
Now ask yourself whether your prior belief will tend to affect the action or decision that you need to take. If the answer is “no” then, as Google’s Chief Decision Scientist Cassie Kozyrkov points out, your prior belief is irrelevant.
If the answer is “yes”, then ask yourself what inference/action/decision your prior belief tilts you towards. If that turns out to be wrong, what would the costs be? If low (e.g., if the decision can be easily reversed), then go with your gut. But if the costs are unknown (e.g., you may not know you’ve made the wrong decision) and potentially high, then you need a strategy to help you overcome confirmation bias. Continue to the next step.
Step 3: Create a strategy
A strategy is a plan – the set of actions you intend to take for each contingency. The important thing regarding confirmation bias is to ask yourself the right questions while making a plan, and to do so before you look at the data.
If using data to generate new ideas, ask yourself these questions:
Why do my data exist? How does that compare with what my current goals are? For example, why was this dashboard built? A given data set almost never covers all aspects of something – especially if that something is currently aspirational (e.g., a new product).
Which aspects of my goal are these data suitable to inform? If your prior beliefs are that metrics X and Y are important, plan to look at X and Y, but give A and B a shot too.
Where are my blind spots (i.e., what important aspects of my decision are my data ill-suited to inform)? Be specific about the limitations of your data and approach.
What are my criteria for “finding something” – and under what circumstances would I find nothing?
And if testing an idea/theory that you already have, ask yourself:
What are the chances that the patterns I’m seeing in my data reflect causal connections between variables vs. simple correlations? How important to my decision is observing causality in my data?
What implications flow from your prior beliefs? Identify the ones that you can test with the data you have – the ones that your data have a decent chance of rejecting if your prior beliefs aren’t true. You can’t claim to have found data to support one interpretation vs. another if you could never have found evidence supporting the alternative explanation with the data and approach you are using.
What kinds of patterns in my data would be inconsistent with the theory I’m testing?
What alternative theories could realistically be reflected in my data?
What alternative explanations can you think of for the patterns you’re observing in the data? Do those have testable implications? Plan to test them if possible: “If X were true instead of Y, then we should see Z in the data – the fact that we don’t see Z casts serious doubt on X as an explanation over Y.”
What evidence will I require to make one decision vs. another? For example, is that new customer loyalty program justified if your test experiment shows that the loyalty program reduces customer attrition by 5%? 2%? Be as specific as possible.
Step 4: Invest in getting your most important decisions right
Use a balanced approach. Cognitive biases can be useful, and making exactly the “right” decision every time probably isn’t the best use of limited resources. But do be aware of your prior beliefs and the potential for confirmation bias to skew your decision making. Before looking at data, we should be able to answer questions like, “Does it make sense to make this mental leap? What’s the likelihood that I’m wrong, and what are the costs of being wrong? How do these compare with the costs of investing in a more thoroughly-considered decision?”
Sometimes making the wrong decision could be very costly, and this is when it makes sense to invest in a more thorough, data-informed decision-making process. That investment involves being honest about how capable our data are of generating new ideas or challenging our existing ones, and then asking questions of it that fully utilize that capability. Targeted decision-making tools (NOT sprawling dashboards) can help systematize this by forcing us to look at metrics identified as important independent from our prior beliefs, thus nudging us to ask the right questions of our data.