1 Introduction

Teachers’ academic judgements and interactions can be distorted by their students’ backgrounds, with research suggesting that characteristics such as socioeconomic status (SES) (Batruch et al., 2017, 2019, 2023; Doyle et al., 2023; Goudeau et al., 2023; Pit-ten Cate & Glock, 2018), ethnicity (Anderson-Clark et al., 2008; Burgess & Greaves, 2013; Campbell, 2015; Connolly et al., 2019; Glock, 2016; Starck et al., 2020), gender (Glock, 2016; Lievore & Triventi, 2023), immigration status (van den Bergh et al., 2010), and physical attractiveness (Longobardi et al., 2022) may all influence teachers’ perceptions. However, rather than engaging with the issue of overcoming these stereotypes and biases, teachers—like many humans—may act defensively and be dismissive of evidence relating to their prejudices (Clark & Zygmunt, 2014; Solomona et al., 2005). The current study aims to address this issue by exploring how teachers engage with potentially threatening information about biases, and testing whether a brief self-affirmation manipulation could make teachers accept greater responsibility for their role in the issue.

1.1 Theoretical background

Biases are tendencies in our behaviours or thoughts that are often automatic reactions to the stereotypes or predispositions we hold about different groups of people (Starck et al., 2020). They reflect what we expect people to be like, based on our own experiences, as well as things we learn from others, the media, and society as a whole (Tatum, 2017). Biases are often widespread and can be useful mental shortcuts; however, despite the best intentions of teachers, these unconscious and automatic biases may also bear a significant cost for students from certain backgrounds.

The prevalence and levels of stereotypes and bias in education may differ by geographical context. For example, research suggests that racial bias is commonplace in the US (Okonofua & Eberhardt, 2016; Starck et al., 2020; Warikoo et al., 2016), whereas in mainland Europe, bias against those from an immigrant background is also an issue within education (De Benedetto & De Paola, 2022; Glock et al., 2013; van den Bergh et al., 2010). Despite these forms of bias also being concerns in the UK, where the current study took place (Burgess & Greaves, 2013; Campbell, 2015; Connolly et al., 2019; Gillborn et al., 2012), some of the largest inequalities in educational outcomes fall along the lines of SES (Easterbrook et al., 2022, 2023). Yet, teachers may be less aware of SES bias compared to racial biases (Doyle et al., 2023). Therefore, SES bias was chosen as the most contextually relevant form of bias to explore in our UK sample, but it is also possible that the findings will be generalisable to other forms of teacher bias.

In England, students who are eligible for free school meals (FSM)—often used as a proxy for lower SES (Taylor, 2018)—are only just over half as likely to achieve a strong pass in GCSEFootnote 1 level English and Mathematics compared to students who are not FSM-eligible (Department for Education, 2022a). Moreover, data from 2019 suggests that by the end of secondary school the ‘attainment gap’ between children from lower and higher income backgrounds was equivalent to around 18 months of learning (Hutchinson et al., 2020). Early indications suggest that this discrepancy has increased as a result of the Covid-19 pandemic (Goudeau et al., 2021; Haelermans et al., 2022; Renaissance Learning and Education Policy Institute, 2021), thereby providing a clear rationale for research into this area. There is a plethora of factors that may contribute to these inequalities, including – but by no means limited to – unequal access to resources and support (Easterbrook et al., 2023), a school system that advantages those with middle or upper-class values, language and knowledge (Bourdieu & Passeron, 1990; Croizet et al., 2017; Goudeau & Croizet, 2017; Jæger & Møllegaard, 2017), and psychological barriers such as stereotype threat, whereby individuals fear that their performance on a task will confirm a negative stereotype about their group (Croizet & Claire, 1998; Hadden et al., 2020; Steele et al., 2002). However, an increasing body of research has shown that biases within the education system may be also contributing to these unequal outcomes. While many biases play out at an institutional level (Murphy et al., 2018), the focus of this study is on individual teachers’ understanding and acceptance of their biases.

1.1.1 Teachers’ biases

Biases that negatively impact students from lower socioeconomic backgrounds may start to take effect very early in life (Campbell, 2015; Goudeau et al., 2023). This is highly problematic as students who gain a negative reputation at a young age are unlikely to shake it throughout their schooling (Childs & McKay, 2001). These biases subsequently appear to affect teachers’ academic judgements: In one recent UK-based study, teachers were given an identical piece of work to assess, believing that it had been written by either a student from a higher or lower socioeconomic background. Despite reading the same piece of work, teachers who believed it had been written by a lower SES student awarded them a significantly worse grade, assigned them to a lower ability group, and perceived their overall level to be inferior compared with those in the higher SES condition (Doyle et al., 2023). Further studies have found that, even when student performance was equal, teachers were more likely to deem a higher academic track more suitable for high-SES children than low-SES and vice-versa (Batruch et al., 2017, 2019, 2023).

In addition to grading and tracking, the quality and quantity of teacher-student interactions have been found to vary with socioeconomic background. Goudeau and colleagues (2023) found that pre-school teachers were less likely to call on lower SES students to participate, and that even when they did, the interactions were shorter. This suggests that at times educators may be blind to both their own biases and the varied needs of students from different groups. Furthermore, Dunne and Gazeley (2008) found that teachers were reluctant to explicitly acknowledge the social class of their students, but that it nevertheless influenced their academic judgements about them. This creates an issue akin to the ‘colourblind’ approach to educating children from diverse ethnic backgrounds, whereby educators may at times overlook the unique experiences of their students (e.g., Apfelbaum et al., 2008, 2012).

1.1.2 Distancing oneself from biased transgressions

Devine (1989) likened prejudice and stereotypes to a bad habit that must be broken with motivation and hard work. Specifically, the individual must have an awareness of their biases, concern about their consequences, and knowledge of how best to replace them (Devine et al., 2012). Therefore, in order to increase teachers’ motivation to work through the discomfort of learning about unconscious bias, we began the current investigation by attempting to raise teachers’ awareness and acceptance of the biases they may exhibit.

Unfortunately, people tend to exhibit a ‘bias blind spot’, whereby they perceive bias in others but not in themselves (Ehrlinger et al., 2005; Pronin, 2008; Pronin et al., 2002, 2004). Such feelings of immunity to bias are akin to the better-than-average and self-enhancement effects, whereby individuals rate themselves more favourably than others on aspects such as popularity (Zuckerman & Jost, 2001) and driving ability (Svenson, 1981). As such, when confronted with notions of our own biases, rather than expressing concern and showing motivation to change, people are typically quick to deny or distance themselves from their biases and prejudices (Howell & Ratliff, 2017; Knowles et al., 2014; O’Brien et al., 2010). This defensive “shield” in the face of personal threats was demonstrated by Clark and Zygmunt (2014), who presented teachers with Implicit Association Test results as evidence of their unconscious racial and skin-tone biases. Teachers’ written reactions to this conscious-raising “encounter” (Gay, 1985), were analysed and categorised into five distinct groups: Disregard, Disbelief, Acceptance, Discomfort and Distress. Most reactions fell into the first two categories, indicating that most teachers are likely to be resistant to acknowledging and addressing their biases. Further research by Solomona and colleagues (2005) shows that in the domain of unconscious racial bias, White teachers use a number of ‘coping mechanisms’ to avoid the dissonance of ideological incongruence (a mismatch between one’s ideologies and behaviour). These include re-centring the focus of the discussion, trivialising sources of information and focusing instead on the perceived difficulties that White people experience. This focus on the self and the personal feelings of sadness, anger, discomfort and guilt mean that there is little space to address the needs of other groups who genuinely are oppressed (Solomona et al., 2005). In this way, attempts to highlight bias in highly defensive individuals may backfire unless they are able to process the information in a non-threatening fashion. One potential avenue for achieving such information processing is through the use of self-affirmation.

1.1.3 Self-affirmation

Self-affirmation theory (Cohen & Sherman, 2014; Steele, 1988) posits that defensive reactions may arise in part because we are motivated to maintain a sense of our own self-integrity – the notion that overall, we are good, moral, and appropriate in our conduct. When self-integrity is threatened—for example through a challenge to our identity or morality—our so-called ‘psychological immune system’ compels us to restore a positive self-image by either acting defensively, drawing on personal resources, or taking advantage of affirmational opportunities available to us (Gilbert et al., 1998; Steele, 1988). These affirmational opportunities often manifest themselves at the expense of other people and include downward social comparison, gossiping, and derogation of outgroups (Sherman & Cohen, 2006). However, self-affirmations—broadly defined as acts that affirm one’s sense of being morally and adaptively adequate (Cohen & Sherman, 2014)—can also be experimentally induced, via writing tasks that encourage participants to reflect on their most important values. These values manipulations are considered to be pivotal reminders of what one values and stands for and so act as powerful ways of inducing a sense of being worthy. Typical manipulations aim to dilute the impact of threats by reminding the individual that they have self-integrity, thereby broadening one’s perspective to a point where threats can be viewed independently and rendered less important for one’s overall self-evaluation (Cohen & Sherman, 2014; Sherman, 2013).

Such affirmations have already proven effective in the health domain, where various studies have shown the potential of self-affirmation as a method of reducing defensiveness arising from threatening messages. For example, Harris and Napper (2005) demonstrated that among participants with the highest levels of alcohol consumption (i.e., those at greatest risk), completing a self-affirmation before reading an article that linked alcohol use with breast cancer reduced defensive processing. Affirmations have also been shown to increase personal acceptance and perceived relevance of threatening messages (Sherman et al., 2000; Sparks et al., 2010; Wenzel et al., 2020), reduce derogation of threatening messages (Armitage et al., 2011; van Koningsbruggen & Das, 2009), reduce psychological discomfort associated with threats (Steele & Liu, 1983), and increase willingness to change behaviours (Armitage et al., 2011; Graham-Rowe et al., 2019). Finally, self-affirmations have also been effective at lessening the impact of negative psychological barriers experienced by students in certain educational contexts (Borman et al., 2018, 2022; Easterbrook et al., 2021; Hadden et al., 2020; Sherman et al., 2021; Turetsky et al., 2020; Wu et al., 2021), but it is possible that they may also be useful tools in reducing teachers’ defensiveness about personal bias.

Indeed, being made aware of how one may – albeit unintentionally – contribute to inequalities could threaten the moral self and thus generate defensive reactions. In a recent study, participants watched a video about animal cruelty before conversing with a purported fellow participant – who was in reality a computerised script – about their dietary choices. During this exchange, one group was subject to social rejection for their animal cruelty-related transgressions (thus creating a threat to the self) and the other group was socially accepted (no threat condition). Threatened participants who then had an opportunity to reaffirm the violated value were more likely to accept their wrongdoings, feel guilty, morally engage with the issue, and restore their self-worth than those who had not reaffirmed (Wenzel et al., 2020). This suggests that self-affirmation may also help to reduce the defensiveness of teachers who are confronted with the moral transgression of their own role in educational inequality. Once the defensive shield has been lowered, it is predicted that teachers would be more likely to engage in repair strategies to reduce their biases.

1.2 Current study

The current pre-registered studyFootnote 2 sought to answer two key questions. Firstly, we wanted to know if teachers would demonstrate a ‘bias blind spot’ (Ehrlinger et al., 2005; Pronin, 2008; Pronin et al., 2002, 2004) whereby they perceive bias in other teachers but fail to acknowledge it in themselves. Secondly, we asked whether a brief self-affirmation would make teachers feel less defensive about the prospect of being biased and therefore more likely to acknowledge their personal biases.

1.2.1 Hypotheses

We predicted that participating teachers would more readily perceive other teachers as biased than they would perceive themselves as biased. In addition, we pre-registered the following hypotheses about the effect of self-affirmation on participants’ willingness to accept and address personal biases:

Compared with participants in the control condition, those in the self-affirmation (treatment) condition would feel less threatened by the message (e.g., Harris and Napper, 2005), and therefore:

  1. 1.

    Report higher levels of message acceptance.

  2. 2.

    Report lower levels of message derogation.

  3. 3.

    Report lower levels of psychological discomfort.

  4. 4.

    Report that the article’s content is more personally relevant.

  5. 5.

    Demonstrate a higher level of readiness to take part in bias-focused Continuing Professional Development (CPD) activities about overcoming bias.

  6. 6.

    Be more likely to agree to discuss their own biases with other educators.

2 Methods

Ethical approval was obtained from the host university’s ethical review board.

2.1 Procedure

Participants were all teachers based in the UK and were recruited via social media posts on teacher groups during the summer of 2022. As the effects of self-affirmation may be diminished when participants are aware of its purpose (Sherman et al., 2009), participants consented to take part in two ostensibly separate studies. The first of these was about “personal values in [their] everyday life”. Using simple randomisation on the Qualtrics platform, participants were allocated to complete either a self-affirmation or control exercise involving a short writing task. The second study was about “issues in education”, for which participants began by reading an article about teacher bias that was designed to be mildly threatening. This procedure was similar to that of past self-affirmation research (e.g., Harris & Napper, 2005), whereby the manipulation (i.e., choosing a value and writing about it) precedes a threat (in this case, the article about issues in education). Following this, participants completed a short questionnaire asking them about their reactions to the article, support for CPD practices to eliminate bias, and willingness to discuss their own biases with trainee teachers. Finally, participants provided their demographic details, read a full debrief and were eligible to enter a prize draw for a £75 voucher (see Fig. 1 for the full sequence).

Fig. 1
figure 1

Study flowchart

Full size image

2.2 Sample

We used the software program G*Power to conduct an a priori power analysis. In order to obtain 0.80 power to detect a small effect size of d = 0.20 at the standard 0.05 alpha error probability, our minimum target sample was n = 192. The effect size was based upon previous literature investigating the role of self-affirmation on message acceptance and perceptions of prejudice (Adams et al., 2006; Sparks et al., 2010). However, as we were using opportunity sampling via social media posts on teacher groups, we over-recruited to account for potential exclusions due to disengagement and drop-outs.

In total 320 UK-based teachers completed a 15-minute online study. Our pre-registered exclusion criteria meant that cases were excluded for withdrawing their data after learning the study’s true aims (n = 3), failing an attention check (n = 15), failing one of two manipulation checks (n = 15), only writing one word in the self-affirmation writing task (n = 1) or for a combination of these factorsFootnote 3. The final sample consisted of 288 teachers (MAge = 41.19, SD = 10.63; MExperience = 13.64 years, SD = 9.68; 94% female; 89% White British; 44% in a leadership position; 33% had been eligible for free school meals as a child (see Table S2 in the Supplementary Materials for a breakdown of baseline characteristics by condition).

2.3 Design

For the main hypotheses, we employed a one-way independent measures design with condition (control vs. self-affirmation) as the between-subjects independent variable. The dependent variables were participants’ message acceptance, message derogation, psychological discomfort, perceptions of personal relevance and relevance to others, support for CPD practices, and commitment to discuss their own biases.

2.4 Materials

Self-affirmation manipulation. The self-affirmation condition involved a similar activity to that used in various studies (e.g., Harris & Napper, 2005; Sherman et al., 2000). Participants were shown a list of twelve values (i.e., honesty, kindness, creativity) and were asked to write down a value – either from the list or one of their own choosing – that was the most important to them and give three reasons for this. To complete the affirmation exercise, participants completed a short writing task in which they were asked to write – in as much detail as possible – an example of something they had done to demonstrate this value in their lives. Self-affirmation theory posits that such writing activities bring to mind important values that affirm one’s sense of self-integrity (Cohen & Sherman, 2014). In the control condition, participants identified their least important value, gave three reasons why someone else may find it important and wrote an example of what someone else might do to demonstrate that value.

Article about teacher bias. All participants read the same short article (374 words) entitled “Unconscious bias in education” which outlined what biases – in general – are, and genuine research (i.e., Doyle et al., 2023; Goudeau et al., 2023.) suggesting that teachers’ unconscious biases may negatively affect the outcomes of children from lower socioeconomic status backgrounds (see Figure S1 in Supplementary Materials).

Message acceptance. Participants’ acceptance of the article’s message was measured using two items adapted from previous research (Cohen et al., 2000; Harris & Napper, 2005): “Overall, how believable did you find the content of the article?” (1 = Extremely unbelievable to 7 = Extremely believable) and “How convincing did you find the content of the article” (1 = Extremely unconvincing to 7 = Extremely convincing). The items had acceptable reliability (α = 0.74) and were combined into a composite score (M = 5.05, SD = 1.34).

Message derogation. Participants’ derogation of the article’s message was measured using two items adapted from previous literature (e.g., Sparks et al., 2010): “The article was biased/exaggerated” (1 = Strongly disagree to 7 = Strongly agree). The reliability of these items was acceptable (α = 0.77) and they were combined into a composite score (M = 4.03, SD = 1.16).

Psychological discomfort. We measured participants’ level of psychological discomfort from reading the article using two items: “The article made me feel uncomfortable about my classroom practices” and “I felt guilty whilst reading the article” (1 = Strongly disagree to 7 = Strongly agree). Reliability of these items was very strong (α = 0.90), so they were combined into a composite score (M = 3.12, SD = 1.54).

Personal relevance. The extent to which the article’s content was deemed personally relevant to the participants was measured using three new items: “How relevant to you is the information in the article” (1. Completely irrelevant to 5. Extremely relevant); “How likely do you think it is that you have unintentionally interacted differently with different students due to their socioeconomic backgrounds?” (1 = Extremely unlikely to 5 = Extremely likely); “How likely do you think it is that you have graded work unfairly due to unconscious biases?” (1 = Extremely unlikely to 5 = Extremely likely); and one item adapted from Jones et al. (2022): “To what extent do you think it is necessary for you to work towards improving your interactions with students from lower socioeconomic backgrounds?” (1 = Not at all to 5 = A great deal). The items had good reliability (α = 0.84) and were combined into a composite score (M = 3.03, SD = 1.01).

Relevance to others. The extent to which the article’s content was deemed relevant to other teachers was measured with adapted versions of the same four items used for the personal relevance measures, e.g., “How likely do you think it is that other teachers have graded work unfairly due to unconscious biases?” (1. Extremely unlikely to 5. Extremely likely). The items had acceptable reliability (α = 0.72) and were combined into a composite score (M = 3.77, SD = 0.75).

CPD support. Participants’ desire to take part in CPD programmes targeting bias reduction were measured using two new items: “I would be happy to have my teaching video recorded to see if/where my interactions with certain pupil groups could be improved” (M = 3.31, SD = 1.26) and “I would be happy to participate in a monthly session at my school to discuss methods of overcoming unconscious biases” (M = 3.53, SD = 1.17; 1 = Strongly disagree to 7 =  Strongly agree). This two-item scale had poor reliability (α = 0.60, r = .43) so items were analysed separately.

Commitment to discuss personal biases. We took a behavioural measure of participants’ willingness to share personal reflections and discuss their own biases, thereby indicating a commitment to acknowledge and act upon them. As a cover story, participants read a sentence suggesting that sometimes hearing fellow teachers open up about their experiences is the best way to encourage us to reflect on our own biases and start the process of positive change. They were then asked if they would be willing to share some personal reflections and discuss their own biases with future trainee teachers. We chose the cover story about sharing reflections with trainee teachers because we expected participants to be aware that hearing current teachers openly share their experiences can be highly beneficial for trainees (Hudson, 2016) and that the hierarchical imbalance would put the [participant] teacher in the position of expert. Commitment to this was given by clicking on a box in which ostensibly they would enter their email address. Those who clicked on the box were coded as willing and the survey informed them that the email address would be collected later, though in reality, to preserve anonymity it was not.

Professional beliefs about diversity. To assess teachers’ beliefs about representation, diversity and teaching practices, the study included an adapted version of Pohan and Aguilar’s (2001) professional beliefs about diversity scale. The 13-item scale included items about supporting students from ethnic minority and lower SES backgrounds, such as “the traditional classroom has been set up to support the middle-class lifestyle” (1 = Strongly disagree to 5 = Strongly agree). After removing three items that had brought Cronbach’s α below an acceptable level, the 10-item scale had acceptable reliability (α = 0.70, M = 3.50, SD = 0.54).

Demographics. Finally, we collected information about participants’ age, gender (“Which gender best describes you?” - Male, Female, Non-binary, Transgender, Intersex, and an open-text response to self-disclose), ethnicity (“Which ethnicity best describes you?”), socioeconomic status during childhood (“During your time as a school child, were you at any point eligible for free school meals?”), teaching experience (“How many years have you been a teacher?”) and leadership responsibilities (“Do you hold a management or leadership position at your school?”). Using a sliding scale of 0-100%, we also asked participants to report the percentage of students in their current schools who were (a) eligible for free school meals, and (b) from ethnic minority backgrounds. These demographics were used to check the representative nature of our sample, ensure that the random group allocation was balanced, and to run exploratory analyses.

3 Results

3.1 Writing length and randomisation check

Participants in the self-affirmation condition (Mwords = 36.73, SD = 27.42) wrote significantly more in their writing about values than those in the control condition (Mwords = 17.20, SD = 14.87), F(1, 285) = 57.17, p < .001. A series of one-way ANOVA tests revealed that there were no significant differences between conditions with respect to participants’ teaching experience, leadership responsibilities, free school meal eligibility as a school child or percentage of children from ethnic minority backgrounds in their current school (all ps > 0.2). However, participants in the self-affirmation group were significantly older, F(1, 284) = 4.93, p = .027, and reported teaching in schools with significantly more students who were eligible for free school meals, F(1, 270) = 6.59, p = .011. Therefore, contrary to our pre-registered analysis plan, we also controlled for these variables in each model listed belowFootnote 4.

Overall means and those by treatment condition for all outcome variables are presented in Table 1. These descriptive statistics reveal two key things: firstly, there may be a difference between means of the personal relevance and relevance to others variables, and secondly, only the support for CPD video variable yielded a noticeable difference in means between conditions.

Table 1 Means and standard deviations for key outcome variables
Full size table

3.2 Analysis plan

As per our pre-registered analysis plan, we tested the effect of the self-affirmation by regressing each outcome on condition (0 = control, 1 = self-affirmation) using a linear model in R. However, as noted above, we also included in the model the control variables participant age and percentage of children on free school meals in their current school (see Table 2 for full output). As the behavioural measure of commitment to discuss personal biases was measured by either clicking to enter an email address or taking no action, we deviated from our original analysis plan to test the effect of the treatment condition on the behavioural measure by using a binomial logit model in R, the results of which are reported in the text below.

Table 2 Regression of key outcome variables on self-affirmation condition and control variables
Full size table

3.3 Personal relevance and relevance to others

Firstly, to establish whether participants believed the article’s main message was equally relevant to themselves and other teachers, we ran a one-way ANOVA on participating teachers’ perceptions of relevance with self-other as the predictor. Teachers were significantly more likely to believe the article was relevant to other teachers (M = 3.77, SD = 0.75) than to themselves (M = 3.03, SD = 1.01), F(1, 572) = 101.11, p < .001, d = 0.84 (see Fig. 2). This suggests that whilst they accepted that bias is an issue within education in general, it was less considered to be their personal responsibility. We then tested the hypothesis that self-affirmation would make participants accept that the article’s content was more personally relevant, thereby potentially narrowing this personal-other discrepancy. However, our linear regression revealed that there were no significant differences in perceptions of personal (p = .853, d = 0.01), or other (p = .892, d = -0.03)Footnote 5 relevance between conditions. Analysis of the control variables in the models suggested that perceptions of personal relevance decreased with greater age and when working in a school with a higher percentage of children eligible for free school meals. Being older was also associated lower perceptions of relevance to others.

Fig. 2
figure 2

Participants’ perceptions of article relevance to themselves and other teachers. Error bars represent 95% confidence intervals

Full size image

3.4 Support for anti-bias CPD

In general, participants were not highly supportive of having their teaching videoed to explore where their interactions with particular groups could be improved. However, there was a significant effect of condition, whereby affirmed individuals (M = 3.46, SD = 1.19) were more supportive of the video proposal than those in the control group (M = 3.16, SD = 1.30), B = 0.31 (0.00, 0.62), SE = 0.16, t = 1.99, p = .047, d = 0.24 (see Fig. 3), thereby offering support to hypothesis 5. Interestingly, the percentage of children eligible for FSM in the participants’ schools also significantly and positively predicted support for this initiative, suggesting that the more contact teachers have with children from lower SES backgrounds, the more supportive they are of engaging with potentially threatening CPD exercises.

Fig. 3
figure 3

The effect of self-affirmation on support for being video recorded. Error bars represent 95% confidence intervals

Full size image

Support for attending monthly meetings to discuss methods of overcoming bias was generally higher than that for the video initiative. However, there was no effect of condition on support for meetings, (p = .340, d = 0.05). The model output also revealed a significant and negative association between age and support for meetings, indicating that the older participants tended to be less supportive of the initiative.

3.5 Message acceptance and derogation

Overall message acceptance was above the mid-point of the scale (M = 5.05 on a 7-point scale) indicating a general acceptance of the article’s message. However, there was no significant difference in acceptance levels between conditions (p = .454, d = -0.07). Conversely, overall message derogation was around the mid-point (M = 4.03 on a 7-point scale), but there was once again no significant effect of self-affirmation on derogation (p = .950, d = -0.01). As such, being self-affirmed made no difference to participants’ acceptance or rejection of the article’s message.

3.6 Psychological discomfort

In general, participants reported that the article had not made them feel guilty or uncomfortable about their practices. Moreover, there was no significant difference in levels of psychological discomfort between conditions (p = .974, d = -0.02).

3.7 Commitment to discuss biases with trainee teachers

The behavioural measure asked respondents to share their email address if they would be willing to discuss their biases with trainee teachers. In the control condition, 30% of participants agreed to enter their email address, compared with 33% of self-affirmed participants. However, a binomial logit regression revealed that the odds of participants agreeing to share their email addresses were not significantly different between conditions (OR = 1.17, p = .551).

3.8 Professional beliefs about diversity

We made further use of our data by running a number of exploratory analyses to examine the relationships between teachers’ beliefs about equity enhancing teaching practices and their engagement with the issue of bias in education. There had been no significant effect of condition on participants’ responses to the professional beliefs about diversity (PBD) scale (p = .770) so we were confident that the PBD measure could be used as a predictor. We used this scale to predict the same key outcomes listed previously. Regression output is shown in Table 3 and reveals that the extent to which participating teachers supported teaching practices to accommodate the needs of diverse learners significantly and positively predicted their acceptance of the article’s key message, feelings of psychological discomfort after reading the article, perceptions of personal relevance, perceptions of relevance to others, and support for monthly meetings in school to discuss overcoming biases (all ps < .001). It also positively but marginally predicted support for being videoed (p = .065). By contrast, higher support for these inclusive teaching practices was associated with significantly lower message derogation (p < .001). Finally, there was no association between these teaching practices and commitment to discussing biases (p = .452).

Table 3 Professional beliefs about diversity as a predictor of key outcomes
Full size table

Further exploratory data analysis investigated whether self-affirmation moderated the relationships outlined above. The interaction between responses to the PBD scale and condition on message derogation approached significance, B = 0.46 (-0.02, 0.94), SE = 0.24, t = 1.91, p = .058, whereby self-affirmation trended towards attenuating the strength of the relationship. Similarly, the strength of the relationship between PBD responses and support for monthly CPD meetings to discuss methods of overcoming bias was marginally attenuated by self-affirmation, B = -0.42 (-0.91, − 0.07), SE = 0.25, t = -1.70, p = .091. There were no other significant interactions (ps > 0.1).

4 Discussion

Research has shown that teachers’ academic judgements may be biased by the backgrounds of the students they teach, but that they may respond defensively when presented with evidence of their own biases. This study employed a sample of current teachers to explore three key themes: Firstly, we investigated how educators engage with psychologically threatening discourse about teachers’ unconscious biases in the classroom. Secondly, we tested whether a brief self-affirmation manipulation could make teachers less defensive when confronted with notions of their own bias, more accepting of their role in the problem, and more willing to put in the hard work required to overcome biases. Finally, we ran exploratory analyses to assess whether teachers’ support for equity-enhancing teaching practices predicted the same key outcomes measured in the pre-registered hypotheses, and whether this varied by condition.

For the first key theme, the results offer a clear indication of a ‘bias blind spot’ (Ehrlinger et al., 2005; Pronin, 2008; Pronin et al., 2002, 2004), whereby teachers accept that bias is an issue for other teachers whilst denying that they bear similar levels of personal responsibility. Devine and colleagues (2012) argued that in order to overcome prejudice, individuals must be aware of and concerned about their own biases. Our findings therefore suggest that breaking the ‘habit’ (Devine, 1989) of prejudice and bias within education may be challenging, as educators – in our sample, at least – were effectively blinkered, perhaps by a desire to maintain a view of themselves as good and moral.

For the second part of the study, we explored whether a self-affirmation manipulation would effectively uncouple the threat of information about biases from participants’ perceptions of their moral self, thereby making them more accepting of their biases and willing to make a change. We found that the self-affirmation manipulation did not alter participants’ acceptance or derogation of the article’s key message, feelings of psychological discomfort after reading the article, perceptions of personal relevance, perceptions of relevance to others, support for monthly meetings in school to discuss overcoming biases, or willingness to discuss personal biases with trainee teachers. After reading the evidence-based article about biases within education, participants in the self-affirmation condition were, however, more willing to have their teaching video recorded to highlight where their interactions with certain groups could be improved. Video observations can be a powerful tool to enable teachers to better understand their own classroom interactions with students (Goudeau et al., 2023; Perry et al., 2020), so this finding suggests that if harnessed correctly, self-affirmations could be applied to help schools engage teachers with this potential avenue for overcoming bias. We did not correct for multiple comparisons as the outcome variables relate to fundamentally different concepts (e.g., psychological acceptance of personal bias vs. the intention the change that comes with support for teaching-related CPD). Nevertheless, the use of multiple tests and the absence of other significant effects indicate that this finding should be interpreted with caution.

The largely null effects of the self-affirmation manipulation raise questions about the efficacy of the manipulation itself and the context in which it was delivered (Ferrer & Cohen, 2019). Unlike much of the original self-affirmation literature, which has yielded encouraging results (e.g., Cohen et al., 2000; Harris & Napper, 2005; Sherman et al., 2000; Steele, 1988), the current study was carried out online. Although some research has suggested that online environments provide a reliable complement to in-person studies (Arechar et al., 2018), much research has pointed to issues with careless responding in online studies due to participant (dis)engagement, environmental distraction, and lack of social contact with the researcher (Francavilla et al., 2019; Meade & Craig, 2012; Nayak & Narayan, 2019). As self-affirmation manipulations require a degree of engagement on the part of the participant, it is possible that the online environment for the manipulation was not suitable. Recent research has indicated that modes of self-affirmation such as watching videos (Shuman et al., 2022) or creating computer passwords that encourage reflection about the embodiment of personal values (Li et al., 2022) could be engaging and effective, so future research could employ these modes in the context of educational bias. Moreover, self-affirmation manipulations tend to work best when the affirmed value is in a different domain to the one under threat, as this enables the individual to see that their self-worth is not entirely contingent on their evaluation of the immediate situation (Sherman & Cohen, 2006). However, when writing about their most important value, many teachers gave teaching-related examples of how they embody it, which may have rendered the affirmed and threatened values to be insufficiently removed from one another for the manipulation to take effect. Finally, self-affirmation research in applied settings (e.g., counteracting stereotype threat in schools) has shown that the long-term benefits of such interventions can be stronger than the short-term gains due to recursive cycles of increasingly positive interactions that participants have with their environments (Cohen et al., 2009; 2017; Goyer et al., 2017). Although this study was focused more on individuals’ immediate responses to threatening messages and did not afford the same student-teacher recursive processes found in the stereotype threat literature, it is possible that the single-hit “intervention” and measurement were insufficient to capture the full potential of self-affirmation.

Participants in the self-affirmation condition were, however, significantly more willing to have their teaching recorded to explore where biases may exist. Having one’s teaching observed may also be perceived as psychologically threatening (Aubusson et al., 2007; Carter, 2008; O’Leary, 2013). As such, it is plausible that if the self-affirmation manipulation did indeed work, it was more effective on the threats associated with being filmed than the threat of being (perceived as being) biased.

Previous studies have demonstrated that following a self-affirmation manipulation, individuals for whom a threatening message is of greatest personal relevance may become the most accepting of its content and express greater intentions to adhere to its advocated behaviour (Harris & Napper, 2005; Sherman et al., 2000; Sparks et al., 2010). In our study, teachers who were unsupportive of equity enhancing teaching practices were precisely the individuals for whom message acceptance may be most important. However, our exploratory analyses revealed that, regardless of condition, participants who were less supportive of equity enhancing teaching practices were also less accepting (and more dismissive) of the message about unconscious biases, perceived it to be less relevant to both themselves and other teachers, and were less supportive of initiatives to address these biases. These findings suggest that high-relevance participants may have engaged in ‘motivated scepticism’, whereby they became less likely to believe counter-attitudinal information (Ditto & Lopez, 1992; Kunda, 1987; Liberman & Chaiken, 1992). Further exploratory analyses hinted that self-affirmation may weaken the strength of these relationships for some of the outcomes, thereby potentially uncoupling the threat from the self (Sherman, 2013). Future research should target these “high-risk” groups as a priority and explore the role of self-affirmation in breaking the link between harmful beliefs and message denial.

4.1 Implications and further avenues for future research

SES-based outcome differences are of great concern in education (Department for Education, 2022a; Easterbrook et al., 2022) and this may be compounded by teachers’ biases (Batruch et al., 2017, 2019, 2023; Doyle et al., 2023) that—when brought to their attention—may threaten educators’ sense of self integrity (Sherman & Cohen, 2006; Steele, 1988). More research is needed to fully probe the potential for self-affirmation in this context, but the largely null results from this study suggest that a brief online self-affirmation may not be a panacea to overcoming bias-threat. The findings do, however, clearly show that like many humans, teachers tend to downplay their own susceptibility to bias. Becoming aware of personal bias may indeed threaten one’s self-integrity and give rise to anxiety, but it can be argued that this discomfort is incomparable to the unjust inequality caused by a discriminatory education system (Solomona et al., 2005). Overcoming these issues requires the work of institutions as well as individuals (Stephens et al., 2020) as focusing discussions of prejudice on the transgressions of a few individual ‘bad apples’ may enable those who deny their own biases to absolve themselves of all responsibility for the issue (Murphy et al., 2018). Although the threatening article presented to participants in the current study included a suggestion that it was the responsibility of both schools and individuals to de-bias the education system, the research evidence included focused only on individual bias. Future research could explore whether improving teachers’ awareness of how biased places can lead to biased individuals could make them less defensive and more supportive of initiatives for change (Murphy et al., 2018).

Furthermore, our threatening article focused on issues surrounding socioeconomic/class bias, as research and government attainment data suggested that this was contextually relevant for the teachers in the UK. Unlike racial and gender bias, class bias is seldom included in organisations’ diversity training (Ingram, 2021; Williams et al., 2018), so it is possible that teachers had not had sufficient time to engage with the issue and consider their role within it. The findings from this study and the research on teacher bias that informed it (Batruch et al., 2017, 2019, 2023; Doyle et al., 2023; Dunne & Gazeley, 2008; Goudeau et al., 2023; Pit-ten Cate & Glock, 2018) offer a rationale for the broadening of diversity programs in both education and non-education settings to include a greater focus on socioeconomic/class bias.

4.2 Limitations

We recruited using a convenience sample through social media posts, which may have biased our sample towards teachers with an interest in research. Additionally, the sample was made up almost entirely of females and White British participants. Although these groups are over-represented in the English teaching workforce (Department of Education, 2022b), the over-reliance on these groups may have unintentionally discounted the views of more typically under-represented groups in education (Buchanan et al., 2021).

Our methodology was also limited by the absence of pre-manipulation and follow-up measures of teachers’ biases and views about bias. As a result, whilst we were able to draw conclusions about their self-reported perceptions of teachers’ roles in educational bias, we were unable to analyse changes in teachers’ perceptions and biases, or to control for baseline differences. It was therefore impossible to conclude whether teachers’ initial biases and perceptions of bias played a moderating role, or if the manipulation led to lasting change. To address this limitation, future research should, wherever possible, aim to take measures both before and after the main study. Moreover, as a method of keeping the study at a manageable length for unremunerated teacher-participants, several of our measures included only two items. Therefore, despite most of them having acceptable or good reliability, results should be taken with caution.

Finally, a number of our key findings strayed from those we initially sought to investigate in our pre-registration. Exploratory analyses can be highly advantageous and help researchers to credibly reach the best explanation for their data (Jebb et al., 2017; Rubin & Donkin, 2022). However, such analyses also necessitate confirmatory work to draw strong conclusions (Jebb et al., 2017), so future research should aim to replicate the exploratory aspects of this project in a pre-registered study.

5 Conclusions

Teachers tend to enter the profession to facilitate learning and often aim to enhance social equity (Richardson & Watt, 2006). However, many teachers may also carry a hidden set of unconscious biases which can have damaging consequences for typically low-achieving groups. This study showed that teachers are more likely to see bias as a problem for other teachers to grapple with rather than themselves, but that levels of acceptance of, and engagement with their biases, were not consistently enhanced by a brief online self-affirmation.