Investigating Low Confidence Among Girls

Simon Woodhead

In a report on gender equality in education the OECD concluded that girls lack of self-confidence meant they achieved lower results than they were capable of (OECD, 2015). They argue that more confident pupils feel free to fail: not being afraid to proffer an answer and learn from their mistakes. More confident pupils are more likely to benefit from the hypercorrection effect (Butterfield and Metcalfe, 2001) which says that errors made with high-confidence are more likely to be corrected than those made in low-confidence. The OECD found that boys outperform girls in science and maths, but when the results were grouped by confidence the gap disappeared or even reversed.

We decided to investigate whether this relationship held true for our bank of diagnostic questions. We were particularly interested in how confidence varies with ability. So at the beginning of 2019 we added a feature to Eedi to allow students to record their confidence in their answer by picking an emoji. 

We collected a data set of 178805 answers by 5645 girls and 169698 answers by 5307 boys, the table below provides some summary statistics.

Notice how similar the number of answers and the number of correct answers is, and gulf between the confidence for girls and boys. 

For each student we calculated their facility as the percentage of questions they answered correctly. By grouping students by facility we can examine the hypothesis that boys generally outperform girls. In the barchart below boys dominate in 4 out of 5 of the groups above 50%, and girls dominate in 3 out of 5 of the groups below 50%. 

For each of these facility groups we can plot a boxplot to show the distribution of confidence. In the graph below we can see that boys are more confident than girls at every facility level. This is a significant conclusion, from the least to the most able a lack of confidence is holding girls back.

It is illuminating to look at the same data grouped by confidence. The barchart below shows the percentage of students in each confidence group for boys and girls. A far greater proportion of boys than girls are found in the 80-90% and 90-100% confidence groups, with a greater proportions of girls in all the lower confidence groups.

Compare this to the distribution of facility for each of these confidence groups shown in the paired boxplots below. In almost every confidence group girls facility is greater than that of boys. Thus our results concur with the OECD result that when grouped by confidence girls outperform boys. In fact this is a direct consequence of the gulf in confidence between girls and boys and could be predicted from our earlier conclusions, because more able girls find themselves in lower confidence groups.

In conclusion, we found evidence that girls have lower confidence than boys when answering diagnostic questions, and this relationship holds for all ability levels. It is not so much a glass ceiling limiting the most able, but a glass slope working against progress at all ability levels. A way forward is offered by the alternative grouping of the same data. At a given confidence level girls outperform boys...maybe we should tell them?


Butterfield and Metcalfe (2001) Errors Committed with High Confidence Are Hypercorrected. Journal of experimental psychology, Learning, memory, and cognition, 27, 1491-4, 10.1037/0278-7393.27.6.1491. 

OECD (2015) The ABC of Gender Equality in Education: Aptitude, Behaviour, Confidence. PISA, OECD Publishing, Paris,

Written by
Simon Woodhead
Head of Research