This paper investigates the extent to which test performance is affected by shocks to non-cognitive skills. 440 students took a low stakes mathematics test. About half of them were exposed to positive affirmation while being given test instructions, whereas the other half served as controls. The students were allocated to 14 tutorials and randomisation was conducted at the tutorial level. Mean comparisons suggest that test scores were raised by the intervention. In particular, students with low maths grades and with self-assessed difficulties in maths gained from the positive affirmation. Results suggest that teachers might increase their students' performance by interventions to their non-cognitive skills. Inference is obtained by four different methods that take into account that randomisation was clustered at the tutorial group level. These methods are evaluated in a Monte Carlo study for data generating processes which resemble actual data. We find that randomisation inference followed by the wild cluster bootstrap have superior size properties compared to conventional approaches.