Empirical studies of non‐cognitive skills often rely on the assumption that these skills are stable over time. We analyse the change in a specific non‐cognitive skill, that is locus of control, to assess the validity of this assumption directly. We find that short‐ and medium‐run changes in locus of control are modest on average, are concentrated among the young or very old, are unlikely to be economically meaningful and are not related to demographic, labour market and health events. Still, there is no evidence that locus of control is truly time‐invariant and the use of lagged measures could result in a substantial attenuation bias.