Confirmation Bias

Confirmation bias: a definition

Confirmation bias (التحيز التأكيدي) is a tendency to look for evidence that agrees with an existing belief, and to interpret and recall evidence in a way that leans towards that belief, instead of treating agreeing and contrary evidence alike. Research on it comes from cognitive and social psychology. Among the references cited in that research are an article by Nickerson, published in 1998, and an article by Wason, published in 1960.

The concept describes a pattern of thinking and passes judgement on no one. It does not accuse an interviewer, a manager or an investigator of bad faith, it does not state that every prior judgement is wrong, and it does not describe what happens in Saudi establishments.

Where does the concept come from?

We checked the bibliographic data of five works in Crossref. That data establishes that the works exist, not what they contain. The titles by Snyder and Swann and by Dipboye are given without their hyphens:

• Nickerson. The article “Confirmation Bias: A Ubiquitous Phenomenon in Many Guises” in Review of General Psychology, volume 2, issue 2, 1998, pages 175 to 220.
• Wason. The article “On the Failure to Eliminate Hypotheses in a Conceptual Task” in Quarterly Journal of Experimental Psychology, volume 12, issue 3, 1960, pages 129 to 140.
• Lord, Ross and Lepper. The article “Biased Assimilation and Attitude Polarization: The Effects of Prior Theories on Subsequently Considered Evidence” in Journal of Personality and Social Psychology, volume 37, issue 11, 1979, pages 2098 to 2109.
• Snyder and Swann. The article “Hypothesis testing processes in social interaction” in Journal of Personality and Social Psychology, volume 36, issue 11, 1978, pages 1202 to 1212.
• Dipboye. The article “Self Fulfilling Prophecies in the Selection Recruitment Interview” in The Academy of Management Review, volume 7, issue 4, 1982.

We have not read the text of any of these works, so we quote no wording and no finding from them, and we draw from their titles only what the titles indicate. Nickerson’s title describes the phenomenon as widespread and as appearing in many forms. Wason’s title concerns a failure to eliminate hypotheses. The title of Lord and his colleagues concerns biased assimilation of later evidence under the influence of prior theories.

The title of Snyder and Swann concerns testing hypotheses in social interaction, and Dipboye’s title concerns self fulfilling prophecies in the selection interview. Our description of the concept and its examples is a summary, in our own wording, of what secondary summaries say about it. The Arabic name of the term is our choice of rendering.

The core idea of confirmation bias

Secondary summaries describe the concept as bias in searching for evidence, in interpreting it and in recalling it, all at once. They report that Nickerson regarded it as a phenomenon that takes many forms rather than as a single flaw. The summaries group those forms into 3 kinds:

• Selective search: A person asks questions whose answer would be “yes” if the belief were correct.
• Biased interpretation: A single piece of evidence is taken as strong support by those who back the belief and as a weak objection by those who oppose it.
• Selective recall: More of the agreeing facts are brought back to mind than of the contrary ones.

The summaries recount Wason’s well known experiment. A participant is told that the sequence 2, 4, 6 follows a rule and is asked to discover the rule by proposing new sequences, each of which is answered as fitting the rule or not. According to the summaries, many participants assume “increasing even numbers” and try sequences of the same kind, such as 8, 10, 12. They receive a yes every time, while the actual rule is broader: “increasing numbers”.

We did not read the original text, so we give neither the number of participants nor their proportions. The lesson the summaries draw is that a test which cannot contradict a hypothesis does not reveal that the hypothesis is wrong.

How confirmation bias differs from nearby concepts

• First impression error: An early judgement about a person settles, and what comes after is interpreted through it. The mechanism set out for First Impression Error has two steps, anchoring and then filtering. Confirmation bias is broader in scope: the belief need not be a first impression or concern a person, and it covers beliefs about a plan or about the cause of an incident. Treating that filtering step as close to confirmation bias is our reading.
• Overconfidence bias: A systematic gap between how accurate a judgement is and how much confidence is placed in it, with confidence running higher. The confidence side of judgement has a separate entry in our glossary, Overconfidence Bias. Confirmation bias concerns the way evidence is gathered for an existing judgement, whereas the other bias concerns how much confidence is placed in one, and the two can occur together in the same person.
• Implicit bias: Our glossary defines Implicit Bias as a tendency of which its holder is unaware, and which affects judgements of other people according to the group they belong to rather than what they have delivered. In implicit bias the judgement stems from a person’s group, while in confirmation bias it stems from a belief already held by the person judging, on any subject.
• Rater bias: A systematic tendency in performance appraisal that makes a rating reflect something in the rater, or in the way the appraisal is carried out, instead of the employee’s performance alone. The recency effect, similarity and the halo effect are among the patterns listed in Rater Bias. Confirmation bias is not one of those patterns. Rater bias is defined within performance appraisal, whereas confirmation bias is not confined to it.

A hypothetical example: an interview of 10 questions

Suppose a hypothetical hiring manager had formed a belief from a CV that a candidate is strong in sales. The manager ran an interview of 10 questions. Of these, 7 were worded to draw out what supports the belief, such as “Tell me about the biggest deal you closed”, and 3 were worded to ask about what contradicts it, such as “Tell me about a deal you lost and what caused it”. The share of supporting questions is 7 ÷ 10 = 70%.

During the interview the manager wrote 20 notes, 14 agreeing with the belief and 6 contrary to it, so the agreeing share is 14 ÷ 20 = 70%. A week later the manager was asked to list what they remembered of the interview and named 11 agreeing notes and 2 contrary ones, 13 in total. The agreeing share of what was recalled is 11 ÷ 13, about 84.6%, a gap of about 14.6 percentage points between the two shares.

Of the contrary notes, 4 of 6 dropped out, which is 4 ÷ 6, about 66.7%. Of the agreeing notes, 3 of 14 dropped out, which is 3 ÷ 14, about 21.4%. Checking the arithmetic: 11 + 3 = 14 and 2 + 4 = 6.

The numbers are hypothetical, set out for illustration, and represent no real interview or study. They do not show that selective recall occurs to this degree in people. In a real situation the gap might be smaller, larger or absent, and the forgetting might come from the time that had passed or from the vagueness of some notes rather than from the belief. Nor does the equality of the two 70% shares show that the wording of the questions caused what was noted. What the example does raise is one question: were the contrary notes recorded in the same detail as the agreeing ones?

Where confirmation bias can appear at work

• Candidate selection: An interviewer forms a belief from the CV and then words the questions to fit it.
• Performance appraisal: A manager holds on to the facts that support their view of an employee and overlooks what contradicts it when writing the appraisal.
• Looking into an incident: Whoever examines it starts from a hypothesis about who caused it, gathers the accounts that fit the hypothesis and treats the contrary ones as exceptions.

These settings are our estimate of where the same pattern of thinking is conceivable. They are not the result of a survey of establishments and do not describe any particular establishment. Dipboye’s title points to the selection interview, but a title tells us neither the size nor the direction of any effect, and we have no measurement of the effect in workplaces.

Steps suggested to reduce the bias

• Common questions written before the candidate is seen: According to Structured Interview, a common question set prevents the interview from being tailored around one candidate. The aim is to lower the chance that questions are worded to support a belief formed in advance.
• A box for contrary evidence on the record form: A box for the evidence on which the judgement rests is part of the written form described in Interview Scorecard. The suggestion is to add a parallel box for whatever contradicts the judgement, and that addition is our view, not an item in the scorecard entry.
• Asking the rater directly about contrary evidence: What did you see that does not fit your judgement? The answer is written down before the decision is approved.
• Recording at the time, not a week later: Selective recall, as the summaries describe it, works on what is retrieved, not on what was recorded at the moment it happened.

All four steps are our own judgement, based on the logic of each step. We have not seen a study we can rely on that measures the effect of any of them on this particular concept.

How the concept is misused

Holding firmly to a belief is not bias in itself. Some beliefs hold steady because the evidence does support them, and the concept describes how evidence is handled, not whether a belief is true. The name is also broad enough to tempt people to explain every error with it, yet not every error of judgement comes from confirmation bias. Other causes include missing information and lack of time. The misuses we see as worth avoiding are these:

• Calling every disagreement confirmation bias: Two people weighing the same evidence differently does not mean that one of them is biased.
• Accusing a rater on the strength of one case: The pattern shows across several cases, not in a single one.
• Assuming that knowing the concept is enough to remove it: We have not seen a study that establishes this. We think a written procedure is more useful than a general warning, and this is our view rather than a finding.
• Using it to dismiss a sound belief: Calling a belief confirmatory does not make it false and does not replace an examination of the evidence.

The role of HR

HR can use the concept to frame questions for reviewing interview, appraisal and fact finding procedures. Were the questions written before the person was seen? Is contrary evidence asked for? Is what was observed recorded at the time? Decisions stay with the managers who own them, and HR contributes to them through its knowledge of procedures, not by classifying how people think.

What the evidence on confirmation bias permits us to conclude

We do not claim that confirmation bias weakens the quality of hiring, appraisal or fact finding in establishments, nor that any particular step removes it. We have not read the articles recorded in Crossref, and in the sources we reviewed we found no Saudi study that we could rely on, so we attribute no effect, figure or percentage to any source.

We think it is better to treat the concept as a vocabulary for describing a pattern in how evidence is handled, and to judge any procedure in an establishment by what shows up in it: whether contrary facts are recorded as fully as agreeing ones, and whether the people appraised know on what basis the judgement was made. What is decided after that depends on the establishment and the nature of the work.

This is an explanation of the concept, not legal advice.