Microscopic simulation models in social sciences nowadays have to meet high demands if they want to compete with data-based approaches like machine learning. The main reason for that is the rich availability of data for almost all human-centred questions. Unfortunately, features like reproducibility, validation, verification, calibration and sensitivity of microscopic simulation models raise problems, which make these approaches hardly applicable for todays quantitative research problems. Consequently, we dedicate this work to the improvement of methods for mathematical analysis of microscopic simulation models. Only by thoroughly investigating microscopic models on a formal base we are able to fully understand and characterise their behaviour and finally find reasonable answers to mentioned problems. We use so-called mean-field analysis for investigating the aggregated numbers of these models. ^Originating from statistical physics, this method is basically used to describe the aggregated behaviour of physical models with a large number of interacting physical sub-molecular particles. Hereby, the aggregated numbers of the model are approximated by a simpler macroscopic model, a mean-field model, usually an ordinary or partial differential equation. Most importantly for this work, the concept can partially be used to describe the aggregated numbers in microscopic simulation models for completely different approaches as well. Hereby, interacting individuals like persons, cars or animals take the place of the particles. Clearly, the usage of mean-field theory for social science applications requires slightly modified methods, as physical particles behave different than entities in microscopic models in social science. Therefore, we present a couple of mean-field theorems, specifically developed for these applications, in this work. ^We finally propose a new classification concept for microscopic models: A series of attributive adjectives according to the models time-update, state-space, randomness and interaction do not only convey a unique picture of specific parts of the model, but also give ideas on possible challenges involved with model, simulation, parametrisation, sensitivity, and finally its mean-field behaviour.