Debiasing Feedback

Biased feedback can have detrimental effects on employees' psychological safety, engagement, and performance.

Feedback is crucial to growth; the evidence on that is clear and needn’t be rehashed here. It’s not easy to give feedback well and it’s often not received the way it’s intended. Giving good feedback is a skill that requires continual refinement, and nature of the skill is that its development will never be complete and perfection will perpetually be out of reach; usually, we’ll fall far from it. The flip side is that receiving feedback well is also a skill, and biases can also affect how we receive even well-intended feedback. These bi-directional biases can have really detrimental effects on relationships, motivation, trust, and performance, so it’s really important that as feedback givers and receivers we work to debias our judgments and feedback. The most effective way to do that is to be aware of your biases, which you’re doing simply by reading this post.

I’ve been reading a lot of the scientific literature on bias in feedback, and I’ve been adjusting my own feedback framework to incorporate some of the things I’ve learned. Jump to the end of the post to read about it. I hope you find it, and the rest of this post, useful.

Feedback inflation and aversion

The most commonly-discussed biases - implicit and explicit - manifest regularly in feedback, and you may have even participated in training or discussions at your workplace about recognizing and eliminating these biases. Typically, implicit/explicit bias training is focused on the feedback giver so their biases don’t manifest as negative feedback. Biased feedback is usually poorer in quality and quantity, and the efficacy of the feedback and developmental opportunities that should come from feedback is usually diminished. But the feedback doesn’t have to be negative to be of poorer quality or quantity or result in fewer developmental opportunities. Feedback inflation is a phenomenon where feedback is more favorable when given to someone from an underrepresented or negatively stereotyped group because the giver is concerned about the appearance of bias or prejudice, or harbors an underlying assumption or stereotype about the receiver’s ability to perform. Operating on the assumption that feedback is meant to help people grow, feedback inflation can be as powerful a stall or stop to career development as feedback laden with implicit or explicit bias. Feedback givers who are concerned about appearing prejudiced can also become averse to delivering feedback at all, which can be even worse. The purpose of feedback is to enact a change in the recipient, and feedback should be used as a constructive, instructive tool. But if feedback is delivered poorly (whether it’s positive or negative) or not at all, it has destructive effects. This can also be true if the feedback is good but is interpreted with bias.

Negative interpretations of feedback

You’ve heard the cliché, “Feedback is a gift.” I think a more apt metaphor is a gift exchange, because feedback should be a dialogue. The feedback giver isn’t the only one harboring biases: studies have found that racially and ethnically underrepresented groups view positive feedback more negatively than do dominant groups, and they tend to attribute praise to low expectations. This is sad for so many reasons, and a slippery slope to really poor outcomes. Even if the feedback is constructive and delivered well, we may not be aware it’s being interpreted negatively. This self-harming bias can negatively impact the recipient’s self-efficacy (judgments that people hold about their own capabilities to learn or perform tasks), and ultimately result in poor performance due to a chain reaction of lower belief in themselves, lower motivation to complete a task, and tasks being completed poorly or not at all. There are myriad reasons why underrepresented or marginalized groups would harbor these biases; suffice to say that they can manifest when receiving any kind of feedback from a member of a dominant group. In the spirit of the two-way dialogue, just as giving feedback is a skill, so is receiving feedback, and receiving feedback isn’t typically a focus of feedback or bias training. I’ll be recording a training video on receiving feedback this month; keep an eye out for it in the Lecture Series section of my website.

While feedback givers and receivers should be aware of how biases manifest based on their position in a feedback conversation, there are some common cognitive biases that can affect both participants. I’ve highlighted some of the most common that can occur specifically in a workplace setting which I keep in mind when giving or receiving feedback.

Common cognitive biases that can affect feedback

Stereotype

Both feedback givers and receivers should be aware of stereotyping. Humans have a tendency to look for traits that correspond with previously formed stereotypes (this is known as a problem from the representativeness heuristic) which can lead to poorly timed or poorly delivered feedback, including inflation and aversion. On the receiving side, stereotype threat occurs when an individual feels that they could be judged based on some stereotype about an aspect of their identity. The truly sinister characteristic of stereotype threat is that a belief in a stereotype isn’t needed to negatively impact performance, but merely knowing that others could judge us because of an aspect of our identity is enough to negatively impact performance.

Anchoring

Anchoring is when an assessment is made by starting from an initial value (the “anchor”) and adjusting to yield a final decision. The anchor can be derived from historical precedent, the framing of a problem, from random information, or even from another cognitive bias. Humans tend to over-rely on anchors and seldom question their validity or appropriateness in a particular situation. When the pandemic hit in 2020, my colleagues and I went from working in the office one day to working from home the next, and naturally there was an adjustment period which took longer for some than for others. I remember having a conversation with my teammate Jessica about task completion. Prior to the pandemic, Jessica completed a high number of tasks per week, but after we started working from home, the number dropped, and a couple months later it was fairly stable around the level it had dropped to. I recall having a one-to-one with Jessica to talk about it, and my anchor was the pre-pandemic number. Indeed, we’ve heard the phrase “returning to pre-pandemic numbers” in so many contexts; the pre-pandemic number is the anchor. It wasn’t appropriate of me to compare the new numbers with pre-pandemic numbers because I was totally inconsiderate of so many variables that resulted in the new number. If the situation was reversed, we went from working from home to working in the office, and Jessica completed a much higher number of tasks in the office than at home, then I’m certain I would have praised her effusively for doing an outstanding job in the office compared to at home. Eventually, this could have tapered off, though…

Regression to the mean

The subtitle of this section is the most math you’ll have to worry about in this post, and it’s another bias from the representativeness heuristic. Individuals typically assume that future outcomes will be directly predictable from past outcomes, so we tend to (naïvely) develop predictions based on the assumption of a perfect correlation with past data. Let’s rely on Jessica’s high performance for another example. If she performed extraordinarily well during a certain period, I might expect similar performance the next period, and the period after that, and so on. But this doesn’t take into consideration the fact that outstanding performance tends to fall back to a baseline over time because circumstances that unlock unusually high performance don’t last. Let’s take the anecdote in the previous example further and we’ll flip regression on its head, this time finding the mean. Jessica’s been back in the office for six months. For the first one or two months, her task completion may have been extraordinary, but after a few months, it’s become average; all the more so because now everyone else is back in the office and performing at that level. A new baseline has been established. Let’s say that Jessica’s employer just hired 100 new people in one week and Jessica has tasks to complete for every one of them. She’ll have at least 100 more tasks completed in this period, putting her above the baseline. But this circumstance that enabled this extraordinary achievement is a one-off; she’ll be back at baseline - regressed to the mean - next month. Not considering regression to the mean (or the powerful combination of regression and anchoring) is a good way to make sure that feedback is demotivating. If I had hounded Jessica to get her numbers back up to the levels that she was doing when the hiring event happened, and there’s no way she could reasonably do that, the effects are destructive to morale, motivation, and engagement. And as eye-rollingly obvious as this sounds, I am rarely surprised to see this bias appear in written feedback.

Conjunctive and disjunctive events

Conjunctive and disjunctive events are those that occur with or independent of other events. We tend to overestimate the probability of events that occur in conjunction with one another (conjunctive events) and underestimate the probability of events that occur independently (disjunctive events). In other words, conjunctive bias occurs when people overestimate the likelihood of success when multiple events must all occur for success to happen, and disjunctive bias occurs when people underestimate the likelihood of at least one of many possible events occurring (especially when failure of any single event can lead to a problem). A complex system will fail if any one single component fails, and the likelihood of a complex system failure is underestimated because only one of many events needs to occur. For people in technical fields, this might be easier to conceptualize. Let’s look at a scenario in a sales organization.

The sales team is launching a new product and assumes that their goal of achieving $1 million in sales within three months of launch is highly likely because the marketing campaign will generate enough leads; the sales team will effectively convert leads into customers; and there will be no product issues. Despite each individual event having a high probability of success, the overall likelihood of achieving the goal is overestimated because all events must happen without fail. If each event has a 90% chance of success, the combined probability is only about 73% (ok, maybe that last section was almost all the math you’ll need to worry about), but the sales team might assume near certainty. This is conjunctive event bias.

Conversely, the sales manager assumes there’s a very low chance the sales target will be missed because the top-performing sales reps will achieve their quotas; regional sales teams will hit their growth targets; and there won’t be any key customer churn. Each failure scenario seems individually unlikely (e.g., a 10% chance of a top sales rep underperforming, 15% chance of regional targets being missed, and 5% chance of losing a key customer), but the combined likelihood of at least one of these events occurring is higher than expected. The manager could underestimate the risk that one failure could derail the entire sales target. This is an example of disjunctive bias.

Conjunctive and disjunctive bias, like regression to the mean, is often not considered when giving feedback. I guess managers don’t want to add probability and statistics to their performance review workload, but it would behoove them to at least be aware of them even if they’re not running the numbers so they can deliver less biased and fairer feedback.

Availability

Enough already - I’ll stop with this cognitive bias… for now. We too easily assume that our available recollections are truly representative of the larger pool of events that exist outside our range of experience. While the availability heuristic (the degree to which an event are readily “available” in memory) can lead to accurate judgments of an event, it can also be misused because of other cognitive biases that come from the availability heuristic, like misjudging the frequency of an event by the availability of its instances (ease of recall, e.g., vivid instances of an employee’s negative behavior will be most easily recalled from memory or will appear more numerous or weighted more heavily than commonplace incidents) or retrieving something more easily from memory because of a familiar commonality or pattern (retrievability bias, e.g., “upscale” retailers tend to be in the same area (think a high-end mall or Rodeo Drive) because they want to be in an area that consumers associate with upscale products/stores). This is probably one of the more pernicious cognitive biases that affect feedback because it’s so easy to recall an outstanding event (be it positive or negative), even if it’s a complete outlier, and that event can make an appearance in a feedback conversation or on a written review and be highly demoralizing if it was a one-off or the result of uncontrollable, external events.

A framework for debiasing feedback

These examples highlight the common and difficult-to-reconcile fracture between connecting performance to a set of standards while maintaining the individualization and personalization of the feedback process, and recognizing where our implicit, explicit, and cognitive biases could influence what feedback we give and how we give it. Basic judgmental biases are unlikely to correct themselves over time, so it’s important that we engage in intentional exercise to debias our judgment.

My personal feedback framework is based on a simple premise:

Deliver specific and actionable feedback, focused on behaviors, intended to motivate the recipient to achieve mastery.

You can learn more about some of the following phases of my framework in my Lecture Series video on giving effective feedback.

1. Start with planning the feedback conversation. Ask yourself hard, specific questions. What are you trying to achieve? What biases could be clouding my judgment? What assumptions am I making about the situation, about myself, or about the recipient? Is my feedback focused on behaviors (right) or personality (wrong)? Am I helping the recipient achieve mastery of a specific thing? Do I even know what it is they were trying to accomplish?

2. When conducting the conversation, be an active listener and seek to understand rather than speak to judge. The focus is on what is right, not who is right. Explain your understanding of events, and ask for their understanding of events. Find common ground, and move forward together.

   1. One of the most potent ways to debias feedback is to increase self-awareness of personal biases, which you’re doing by reading this post. Even offering warnings about the possibility of bias works well, something as simple as, “I’m approaching this from my perspective based on my level of understanding, and it could be biased. I need your help so I can understand a fuller picture of the situation.”

   2. Provide feedback that is rooted in concrete examples and factual observations rather than generalizations or assumptions. Cite specific instances and behaviors to reduce the influence of personal bias and make feedback more actionable and fair.

   3. If giving a performance review, and even occasionally in a regular feedback conversation, employ a standardized methodology that applies to everyone, like a rubric or a checklist, that focuses on specific, observable behaviors and outcomes. This will help minimize subjective judgments and ensures that feedback is based on consistent criteria for everyone. See this article for a great walk-through for creating excellent rubrics.

3. Reflecting on the conversation. Again, ask yourself hard, specific questions. Did I accomplish what I wanted? Are we moving forward with a plan? Did I express myself in a way that made me proud? Was my intention clearly understood by the recipient?

4. Following up in writing.

This framework focuses on the first two parts of my premise: Deliver specific and actionable feedback, focused on behaviors. The final part, motivate the recipient to achieve mastery, is a topic all its own, which will be the topic of my next post.

I hope you found this useful. Thanks so much for reading!