When non-economic determinants, notably institutions, are considered by mainstream economics, they are analysed via modelling. The paper argues that there are intrinsic limitations to the modelling of the contributions of institutions to economic dynamics, in particular that modelling under-addresses the composite nature of causation. Economic dynamics undoubtedly exhibit strong regularities, which are clarified by models. Yet, there are elements that are inherently qualitative, and which involve causation mechanisms that cannot be explained by models. An ‘institution’ is not a unitary entity that has a stable reference across time and space and which can be a term in unambiguous causalities. Concepts may not be computable, and therefore causation in models differs from causation involving concepts. Equally, with modelling, agents behave according to the rules that have been assigned to them by the modeller. Yet in the ‘real world’, individuals may depart from social norms without any ‘reason’ (‘rationality’). These behaviours and the causation they involve are not predictable ex ante, there may be no appropriate measurement for them, and they may be typically the small ‘bifurcations’ that can induce large effects. In a second step, the article explains the cognitive dimensions that underlie such unpredictability. Institutions include forms and contents and imply the transformation of beliefs in deontic norms and subsequently their dissemination in a social group. Diversity of beliefs can be modelled, but the content of this diversity is more difficult to model. There are undoubtedly regularities, but there are also singularities.Read less <