Similarity Bias

Similarity Bias: a definition

Similarity Bias (تحيز التشابه), called the Similar to Me Effect in the interview literature, is a rater’s tendency to score more highly someone who resembles them in background, interests, career path or style, so that resemblance enters a judgement of competence without ever being written down as a reason for it. In Arabic the concept also appears as «انحياز التشابه» and «التحيز للمشابه», and the English term Affinity Bias names something close to it. The Arabic title is our choice of rendering.

The concept describes a pattern in rating and passes no judgement on anyone. It does not accuse an interviewer or a manager of bad faith, it does not hold that every resemblance between a rater and a candidate is a flaw in the decision, and it is not a description of what happens in Saudi organisations.

Which sources did we check?

We verified in Crossref the bibliographic data of the works we cite. Crossref confirms that a work exists and gives its details, but says nothing about its content. We dropped hyphens from the English titles and names.

• Byrne. The article “Interpersonal attraction and attitude similarity” in The Journal of Abnormal and Social Psychology, volume 62, issue 3, 1961, pages 713 to 715, identifier 10.1037/h0044721.
• Rand and Wexley. The article “Demonstration of the Effect, ‘Similar to Me,’ in Simulated Employment Interviews” in Psychological Reports, volume 36, issue 2, 1975, pages 535 to 544, identifier 10.2466/pr0.1975.36.2.535.
• Wexley, Alexander, Greenawalt and Couch. The article “Attitudinal Congruence and Similarity as Related to Interpersonal Evaluations in Manager Subordinate Dyads” in the Academy of Management Journal, volume 23, issue 2, 1980, pages 320 to 330, identifier 10.2307/255434.
• Tsui and O’Reilly. The article “Beyond Simple Demographic Effects: The Importance of Relational Demography in Superior Subordinate Dyads” in the Academy of Management Journal, volume 32, issue 2, 1989, pages 402 to 423, identifier 10.2307/256368.
• McPherson, Smith Lovin and Cook. The article “Birds of a Feather: Homophily in Social Networks” in the Annual Review of Sociology, volume 27, issue 1, 2001, pages 415 to 444, identifier 10.1146/annurev.soc.27.1.415.
• Sears and Rowe. The article “A personality based similar to me effect in the employment interview: Conscientiousness, affect versus competence mediated interpretations, and the role of job relevance” in the Canadian Journal of Behavioural Science, volume 35, issue 1, 2003, pages 13 to 24, identifier 10.1037/h0087182.
• Montoya, Horton and Kirchner. The article “Is actual similarity necessary for attraction? A meta analysis of actual and perceived similarity” in the Journal of Social and Personal Relationships, volume 25, issue 6, 2008, pages 889 to 922, identifier 10.1177/0265407508096700.
• Rivera. The article “Hiring as Cultural Matching” in the American Sociological Review, volume 77, issue 6, 2012, pages 999 to 1022, identifier 10.1177/0003122412463213.

We read the abstracts of Rand and Wexley, McPherson and colleagues, Sears and Rowe, Montoya and colleagues, and Rivera as OpenAlex publishes them, and we read the full text of none of these articles. Of the abstract of Wexley and colleagues we saw only the first sentence, and of the Tsui and O’Reilly abstract the first sentence and the start of the second. We found no abstract for the Byrne article.

What is the core idea?

The oldest of the eight sources is Byrne (1961), whose title links the attraction between two people with the similarity of their attitudes. We attribute no finding to it, because we did not read its abstract.

In sociology, the abstract of McPherson and colleagues (2001) opens with the phrase “Similarity breeds connection” and calls this principle homophily. The abstract reports that the principle structures ties of every type, including work, advice and the exchange of information, and that personal networks turn out homogeneous in social, behavioural and psychological characteristics.

The abstract of Montoya and colleagues (2008) states that they analysed 460 effect sizes from 313 laboratory and field studies, and found attraction correlated with actual similarity at 0.47 and with perceived similarity at 0.39. It adds that the effect of actual similarity weakened outside studies with no interaction and was not significant in existing relationships, while perceived similarity remained a predictor of attraction in all of them. The study separates actual resemblance from perceived resemblance, and its subject is attraction between people, not hiring decisions.

Where does similarity bias appear in selection and appraisal?

• The videotaped interview. The abstract of Rand and Wexley (1975) describes videotaped employment interviews in which two candidates gave background information that either matched or differed from the interviewer’s, judged by 160 participants. It reports that similarity of background raised ratings of the candidate’s suitability for the job and of the candidate’s other personal characteristics. These were videotaped interviews, not real ones.
• Personality traits. The abstract of Sears and Rowe (2003) states that similarity between rater and candidate in conscientiousness had a significant effect on ratings of suitability for the job. The authors suggest that this effect does not reflect rater error in every case, and that in some cases similarity may support the validity of the interview. The article’s title makes the job relevance of the trait part of the question, so the question, in our view, is whether the shared trait bears on the job, not whether a resemblance exists.
• Hiring in professional firms. The abstract of Rivera (2012) reports that she built her study on 120 interviews with employers and on observation of a hiring committee, and that employers looked for candidates who were competent and also culturally similar to them in leisure pursuits, experiences and the way they presented themselves. In the employers’ eyes, concern for shared culture repeatedly outweighed productivity. The firms studied were elite professional service firms.
• A manager rating an employee. The sentence we saw from the abstract of Wexley and colleagues (1980) says the field study examined how actual similarity and perceived congruence relate to job satisfaction and to performance ratings in 194 manager and subordinate pairs. We do not know its results.

The article by Tsui and O’Reilly (1989) speaks in its title of “relational demography” between superior and subordinate, which, as we understand it, means the difference between the two in demographic characteristics rather than the characteristics of either one alone. We did not read enough of its abstract to attribute a finding to it.

How does similarity bias differ from related concepts?

• Rater bias. Similarity is one item among the patterns listed in the entry on Rater Bias, which describes it as favouring whoever resembles the rater in career path or working style, alongside the recency effect, central tendency error, leniency or severity, and the halo effect. That entry sets the concept in performance appraisal, whereas in the abstracts we read the resemblance effect also reaches the employment interview.
• Implicit bias. Our glossary defines Implicit Bias as a tendency of which the person who holds it is not aware, and which affects their judgement of other people according to the group they belong to rather than what they have delivered. Its interview example is favouring the candidate who resembles the interviewer in background or interests. The difference, in our reading, lies in the reference point: in similarity it is the rater as a person, and in implicit bias it is a group to which the person being rated belongs.
• Confirmation bias. A tendency to look for evidence that agrees with an existing belief, and to interpret and recall evidence in a way that leans towards it, is what the definition of Confirmation Bias explains. In our estimate the two can occur together: a first impression produced by resemblance, then questions that look for whatever supports it.
• The halo effect. A single positive impression of a person carries over into the assessment of all their other qualities, according to the account of the Halo and Horn Effect. The Rand and Wexley abstract reports that similarity raised ratings of the candidate’s other characteristics, and we think a shared background thereby comes close to being the source of a halo.
• Culture add. Readers who want a selection criterion that asks what a candidate would bring to the team, instead of how closely the candidate resembles it, can turn to Culture Add, which separates a criterion written as behaviour observable at work from familiarity, which cannot be written down.
• Leader member exchange. In the entry on Leader Member Exchange Theory, the subject is how the quality of the relationship between a leader and each member varies inside a single team. Similarity is not part of its description of the theory and comes up only in passing, where it points to rater bias. We found no link between the two concepts in the sources we read that we could rely on, and the closeness of their topics is our judgement.

A worked hypothetical example: ten candidates and one rater

Suppose a hypothetical rater interviewed 10 candidates for one job, 4 of whom graduated from the rater’s own university and 6 from other universities. After each interview the rater recorded an “overall impression” scored out of 5, then scored the answers again against a written scale for each question.

• Overall impression, the four: 4, 5, 4 and 4, which total 17, an average of 17 ÷ 4 = 4.25.
• Overall impression, the six: 3, 4, 3, 4, 3 and 4, which total 21, an average of 21 ÷ 6 = 3.5.
• Written scale, the four: 4, 3, 4 and 3, which total 14, an average of 14 ÷ 4 = 3.5.
• Written scale, the six: 3, 4, 4, 3, 3 and 4, which total 21, an average of 21 ÷ 6 = 3.5.

The gap in overall impression is 4.25 minus 3.5, which equals 0.75 points, or 0.75 ÷ 5 = 15% of the whole scale, while the gap on the written scale is zero. We invented the figures to show the method of comparison, and they represent no rater and no study.

The gap does not prove bias. The sample is 10 candidates, the written scale itself still involves the rater’s judgement, and equal averages do not mean equal candidates. What the example raises is a question: does the gap between the two groups widen when candidates are judged on an overall impression and narrow when they are judged on a written scale?

What measures are proposed to narrow the bias?

The measures we list are our own reasoning, based on the logic of each measure. We have not seen a study we rely on that measures the effect of any of them on this particular bias.

• Standard questions and a set scale. In the format known as a Structured Interview, every candidate is asked the same questions in the same order and the answers are scored against a scale defined in advance. That entry reports that a common question set prevents the interview from being tailored around one candidate.
• Dropping the overall impression box. Or separating it from the score that counts, so that every judgement goes back to a specific answer.
• Criteria written before the candidates are seen. A criterion written after the interview easily ends up describing whoever impressed the interviewer.
• Raters from different backgrounds. Each rates alone before the panel confers, so that no single resemblance is shared by everyone on it.
• An aggregated comparison. Average scores grouped by the background a candidate shares with the rater, across several hiring rounds rather than one.

Common misreadings of similarity bias

• Inferring it from a single case. Hiring a candidate who resembles the rater does not by itself show bias; the pattern shows in the aggregate.
• Correcting it by favouring difference. Choosing a candidate because they do not resemble the rater is another unwritten criterion.
• Assuming that awareness removes it. We have seen no study that establishes this, and we think a written procedure is more useful than a general warning.
• Confusing it with discrimination. Similarity is a tendency in rating, and the legal treatment of discrimination is a separate matter to be read from its own sources.

What is the role of HR?

HR can use the concept to frame questions for reviewing interview and appraisal forms. Does the form have a box for an overall impression that cannot be traced to a question? Were the criteria written before the vacancy was opened? Are average scores compared according to what a candidate shares with the rater? These questions are of our own making, not taken from a source. Decisions on hiring and promotion remain with whoever holds that authority in management, and HR contributes to them by designing the instrument, not by classifying raters.

What does the evidence allow us to say?

We do not say that similarity between rater and candidate lowers the quality of hiring in any particular organisation, nor that any particular measure removes it. All we have are abstracts of some of the articles, sentences from the abstracts of others, and the title of an article whose abstract we did not read. We attribute no effect to any source beyond what we reported from its abstract, under that source’s name.

Since we read the text of no study, the size of the effect in organisations is unknown to us, and we report no figure beyond those in the abstracts. In the references we reviewed we found no Saudi study we could rely on about similarity bias in Saudi organisations.

In our view the concept is better treated as a question put to every selection criterion: does it describe what the job needs, or whom the rater feels at ease with? What is decided after that differs with the job, the team and who does the rating.

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