Published on Let's Talk Development

Is there systemic bias in the classroom?

This page in:
Kindergarten students from Bertolt Brecht School walk through the Plaza de Armas in Lima, Peru on June 28, 2013 Kindergarten students from Bertolt Brecht School walk through the Plaza de Armas in Lima, Peru on June 28, 2013

The events of this last year have not only kept us inside and isolated, but they have also forced us to reckon with how our biases can perpetuate disadvantage among historically marginalized populations. West Indian cricketer Michael Holding has called for the end of “brainwashing” in schools where children are not taught about the achievements of black people. In India, cosmetics companies have announced that they will stop marketing skin lightening creams, while a popular online matchmaking site has removed its skin tone filter.  More recently, in the United States, we have started to question  why events like the Tulsa Race Massacre or Native American experiences after 1900 do not appear in schools’ history curricula. A more obvious concern, though, is how disadvantaged children are treated in schools. Can systemic bias affect the way teachers assess their aptitude and potential?

In a recent working paper, we investigate whether poor children are assessed differently than more affluent children. Getting an answer to this question is not straightforward; we cannot just ask teachers, “Are you biased against poor children,” as (i) most people are unaware of the biases they harbor and (ii) who on earth would answer, “Yes” to that kind of question? We also cannot compare how teachers grade poor and affluent students in their own classes as any differences observed among students from different socio-economic groups could reflect true differences in study-time or human capital investment more broadly, as well as the effects of any biases (good or bad) experienced in school. 

Previous experimental research in lab settings tried to isolate bias among university students. Psychologists John Darley and Paget Gross showed the students a video of a girl, Hannah, answering exam questions and asked them to report on her grade level. Some students first saw an introductory video in which Hannah was portrayed as a poor child, while other students saw an introductory video that implied she came from a middle-class background. The university students were randomly assigned to the different introductory videos. Which introductory video they saw made a difference in their assessments of Hannah; they rated her academic performance worse when they thought she was poor.   

While this result implies that the students used Hannah’s background to assess her grade level, university students are not teachers. They have little experience assessing students. To see whether teachers would exhibit the same bias, we had to bring the lab to the classroom. We went to Metropolitan Lima and recruited teachers in public primary schools. We changed Hannah to Diego, and in addition to the two different introductory videos (depicting backgrounds that approximately corresponded to the second and third quintiles of Peru’s income distribution), we created two variants of the examination video. In one video, Diego’s performance is a little inconsistent: he gets some easy questions wrong and some difficult questions correct. Sometimes he fidgets, sometimes he’s focused. In the second video, he’s a model student: he gets most questions correct and maintains his focus most of the time. We introduced these two variants to check whether clearer signals of aptitude decrease bias. We randomly assigned teachers to each combination of variants of the introductory and examination videos.  

Bias in the classroom

Our results suggest that teachers do use information on socio-economic status when assessing the aptitude, behavior, and potential of children. When Diego’s exam performance was inconsistent, teachers first observing the introductory video depicting him as poor were 14 percentage points more likely to rate him as performing below grade-level than teachers who had first observed the introductory video portraying the more middle-class background. Teachers who thought they were observing the performance of a poorer child predicted he would go on to tertiary education 19 percentage points less often than teachers who thought they were rating a middle-class child. The introductory video depicting a poor Diego had no impact on teachers’ assessment of his behavior. On the other hand, when the exam video showed a high-achieving student, teachers’ relative assessments changed. This time, they thought that poorer Diego was 15 percentage points more likely to be performing above grade level. Nevertheless, they still thought that he was 15 percentage points less likely to advance to tertiary education compared to teachers who saw the introductory video portraying Diego with a middle-class background. When observing the model student in the exam, teachers were also harsher in their assessment of poorer Diego’s behavior: they thought he was less motivated and less emotionally mature than the teachers who thought they were assessing a child from a middle-class background.                  

Thus, it seems as though knowing a child’s socio-economic background affects how teachers will assess his or her performance for both children who might struggle with curricular content and children with high aptitude. Children from poorer backgrounds will be underestimated.  When teachers thought Diego was poor, they fixated on his wrong answers and judged his scholastic aptitude as much lower than when they thought he came from a middle-class background, even though all teachers saw the same exam video. When, however, Diego got most of the questions correct in the exam video, this made it very difficult for the teachers to penalize poorer Diego on the exam questions, so they fixated on his behavior and judged him more harshly on that dimension.

Given evidence from other contexts that teachers’ gender  and racial biases affect students’ academic performance and progression, this difference we uncovered along the lines of social class most likely hinders the scholastic performance (and possibly educational attainment) of poorer children. Teacher’s assessments may also have detrimental and lasting effects on children’s own perception of self. This means that common policy tools to address inequities in human capital like conditional cash transfers or means-tested scholarships may be insufficient to overcome the disadvantage that poorer children face in the classroom. Instead, we may have to work on teachers’ mindsets. Consistent with our experimental results, a recent study of 20,000 teachers across nine countries by our colleagues Shwetlena Sabarwal, Malek Abu-Jawdeh, and Radhika Kapoor found that 43 percent of teachers reported a belief that “there is little they can do to help a student learn” if his or her parents are uneducated.

Changing mindsets may be difficult, however, particularly if bias is systemic - learned as part of social norms from an early age and reinforced in curricula, arts, and media. In these cases, we may need to obscure a child’s identity or use external assessments to mitigate the effects of bias, just as blind auditions increased the proportion of women in symphony orchestras and universal screening with IQ tests increased the proportion of economically disadvantaged students placed in gifted education programs in the United States. Making teachers aware of their biases or targeting self-affirming interventions to students could be promising ways to further support a permanent shift in mindset. Our study uncovered a problem: biased assessments of the scholastic aptitude and behavior of poor children. We hope that our results stimulate future research in classrooms to shed much needed light on potential solutions.  


Alaka Holla

SIEF Program Manager

Join the Conversation

The content of this field is kept private and will not be shown publicly
Remaining characters: 1000