While Agent-Based Modelling has often been associated with quantitative and formal approaches, qualitative data offers indispensable insights, particularly when modelling social complexity, context-dependent behaviours, and meaning-making processes. This work surveys the use of qualitative data as an input to Agent-Based Modelling, emphasizing its role across varying modelling purposes and at different stages of the modelling cycle.

1 Introduction

Understanding how human societies adapt to environmental and social pressures is a question of ongoing relevance, particularly considering contemporary challenges such as climate change, resource scarcity, and institutional transformation. Historical cases provide long-term records of how communities sustained themselves through compounding crises involving shifts in fiscal regimes, demographic shocks, and climatic variability [1, 2].

The first millennium CE Mediterranean is particularly well suited to exploring this challenge. Communities across this region experienced successive imperial transitions: the decline and fragmentation of the Western Roman Empire, the continuation and development of the Eastern Roman (Byzantine) Empire, and the expansion of the Islamic Caliphates. Each political regime brought distinct modes of economic organisation, fiscal demands, and social regulation that shaped how local populations managed their resources and responded to institutional change.

Debates about this period often contrast environmentally driven explanations [4] with interpretations that emphasise structural and institutional factors [5]. Both positions have been constructed at the level of narrative historical interpretation, without systematic formalisation of the assumptions embedded in either approach. This presentation reports on work undertaken through the ERC-funded project SSE1K: Science, Society and Environmental Change in the First Millennium CE, which uses agent-based modelling (ABM) to understand the socio-ecological history of the Mediterranean. ABM allows us to move beyond both of these traditional interpretive frameworks by modelling the active agency of rural households, specifically the choices they made about crop selection, market exchange, and resource storage, as strategies shaped simultaneously by accumulated social knowledge, institutional context, and environmental conditions.

2 Data and Methods

A previously developed generic ABM of rural household behaviour in the Mediterranean during the first millennium CE [6] provides the basis for this study. Here, we apply this model to Egypt, selected as our first case study for its exceptional corpus of primary textual records, including papyrus archives, fiscal documents, administrative correspondence, and agronomic manuals describing agricultural practices and resource management. This documentary corpus enables comparative analysis across different regions of the Nile Valley (the Delta, Middle Egypt, and the Fayum) throughout the first millennium CE.

The ABM represents rural households as agents making annual decisions about crop and livestock portfolios, market exchange, and resource storage, subject to a range of constraints and rules. Environmental constraints include precipitation, yield variability, and temperature change. Fiscal obligations are defined by period-specific tax rates. Technological factors include the availability of irrigation infrastructure and assumptions around agricultural productivity. Political disruptions such as conflicts and epidemic events are incorporated as stochastic shocks. Social rules derived from historical sources, including market prices and exchange norms, govern how agents interact with one another and with the broader economic environment. The landscape is represented as a spatially explicit grid calibrated to the environmental characteristics of the Nile Valley.

The model integrates textual data, palaeoenvironmental reconstructions, and modern agronomic databases. Written sources are often focused on administrative centres and elites, and do not provide systematic coverage of rural communities across all regions and periods. Modern agronomic data define ideal current conditions under which resources can be produced, and establish the potential yield of each resource depending on those conditions. However, special care must be taken to avoid extrapolating them directly to pre-modern contexts, given the absence of equivalent technologies, the lack of modern inputs such as fertilisers or pesticides, and the substantial changes in cultivars and crop productivity over time. Working between past and present datasets requires a process of translation, as much in computational settings as in historical ones.

ABM is not commonly used in archaeology, and very rarely used by historians, particularly in the premodern era [7]. In building the model, we have developed a series of novel protocols and practices for integrating the complex historical evidence of this period [6]. Where documentary sources provided direct, quantifiable evidence, these were encoded directly as behavioural parameters in the model. In this way, the model prioritises historically documented behaviour over theoretically optimal behaviour, which is essential for capturing the actual decision-making logic of past communities rather than projecting modern assumptions onto them. Where historical evidence was directional but not quantitative, modern agronomic databases calibrated to Mediterranean conditions were used to define plausible parameter ranges consistent with both the available textual evidence and the ecological context, in some cases with off-sets designed to capture changes in productivity through time.

Sources were cross-checked within the selected corpus of primary textual data to identify consistent trends and detect apparent outliers. Importantly, all modelling decisions were documented explicitly, distinguishing parameters directly grounded in historical evidence from those requiring interpretive inference, in order to ensure transparency and reproducibility. This dual-source strategy was designed not to eliminate uncertainty, but to make it tractable: for each parameter, we specify what is known, how it is known, and what assumptions are required to fill evidential gaps.

Parameter values are period-specific, reflecting documented differences in fiscal pressure, administrative organisation, and agricultural diversity across Egypt. Higher fiscal extraction in the Byzantine and Islamic periods is reflected in increased taxation parameters [8]. Productivity constraints associated with the expansion of irrigation infrastructure and increased labour demands during the Byzantine period are captured through yield-change and irrigation coverage parameters [9]. The agricultural diversification of the Islamic period, including the introduction of rice, sugarcane, and cotton, is represented through an expanded crop portfolio with corresponding agronomic parameters [10]. Together, these period-specific parameter sets allow the model to simulate how the distinct political-economic contexts of each regime shaped the conditions within which rural households made decisions and pursued agricultural strategies.

The model is used to run systematic scenario analyses, varying parameters most relevant to competing historiographical interpretations, including taxation intensity, climate stress, epidemic shock frequency, and the availability of market infrastructure. Each scenario produces time-series outputs for population dynamics, and production strategy (measured via a crop diversification index).

3 Preliminary Results and Conclusion

At this stage, the project is focused on refining the historical parameterisation of the model. We are currently conducting a systematic cross‑checking of all parameters derived from the textual corpus in order to calibrate fiscal, agronomic, and behavioural values for the different regions of Egypt across the first millennium CE. This process involves reconciling heterogeneous documentary evidence, identifying consistent patterns, and excluding outliers to ensure that the model reflects historically grounded ranges rather than modern assumptions.

Once this calibration phase is complete, the next step will be to develop a series of simulation experiments exploring how rural households in the Delta, Middle Egypt, and the Fayum may have responded to varying combinations of fiscal pressure, environmental variability, and socio‑political disruption. These experiments will allow us to test competing historiographical interpretations and to evaluate how institutional, environmental, and economic factors interacted to shape resilience trajectories across different ecological and administrative contexts.

Although full results are forthcoming, early exploratory runs suggest that resilience dynamics are unlikely to be driven by a single factor alone. Instead, they appear to emerge from the interaction between household‑level adaptive strategies, regional ecological constraints, and the broader institutional environment. The ongoing calibration and sensitivity analysis will enable us to formalise these observations and present robust comparative results in the next phase of the project.

This work engages directly with current historiographical debates regarding the relative importance of environmental versus institutional factors in shaping the local practices of Mediterranean rural communities in the first millennium CE [11]. By formalising assumptions within a simulation framework, the model enables us to test how fiscal pressure, market access, environmental variability, and household-level strategies may have interacted to produce different resilience trajectories across regions and periods. Rather than framing the analysis in terms of either environmental determinism or purely institutional explanations, this approach allows us to explore how macro-level patterns could emerge from interplay between social knowledge, economic constraints, and ecological conditions. Beyond the specific case of Egypt, this work illustrates how ABM can move beyond binary interpretive frameworks to model the social dimensions of adaptive behaviour. By grounding household decision rules in documented historical evidence, the model captures how accumulated social knowledge and economic strategies shaped rural outcomes in interaction with, rather than simply in response to, environmental and socio-political conditions. Perhaps most importantly, the procedure developed here for translating heterogeneous qualitative evidence into formal ABM parameters is transferable to other historical and social contexts. The challenges involved in bridging empirical evidence and modelling assumptions, and in making that process explicit, reproducible, and critically examined, extend well beyond Mediterranean history and represent a broader methodological contribution to social simulation research.

Regular physical activity (PA) reduces the risk of non-communicable diseases, maintains physical and cognitive functions, and improves mental wellbeing as well as weight management. PA is not determined by a single factor; rather, it is shaped by a complex interplay of factors at individual, social, and environmental levels. Increasing attention has been given to the built environment as a key determinant of PA. Parks can play an important role in facilitating PA by providing opportunities for a wide range of activities, from walking to running, as well as supporting social interactions and relaxation. While research has examined associations between park elements and park-based PA, these studies tend to concentrate on a limited set of factors and often examine them in a straightforward, cause-and-effect way. However, the interaction between park elements and park-based PA is much more complex, influenced by a mix of environmental, individual, and social factors. To address this complexity, a systems approach is required. Agent-Based Modelling (ABM) offers a method to simulate the interactions between individuals and their environment, accounting for heterogeneity in behaviours and decision-making. ABM is particularly useful for exploring how populations respond to interventions, and can overcome common limitations of natural experiments, such as short follow-up periods and small sample sizes. This study uses ABM to simulate park visits and their influence on park-based PA among adults in two Dutch cities: Utrecht and Dordrecht. Both cities are implementing interventions to improve local parks, aiming to promote PA. Agents represent adult residents who are heterogeneous and characterised at individual, social, and physical environment levels. The model simulates daily park visits, park-based PA, and social interactions, and allows testing of various interventions. The ABM is parameterised using multiple data sources: a scoping review, Group Model Building (GMB) workshops, pre- and post-intervention surveys, and systematic park observations (SOPARC). Model development is ongoing. The next steps are to analyse the survey and observation data, implement these data into the ABMs, and assess the models’ ability to estimate the effects of future interventions tailored to different neighbourhoods and population groups.

Life course processes such as inequality, family formations, social mobility unfold over decades and across interacting life domains, which makes them difficult to study, both for quantitative methods (because of their complexity) and qualitative methods (because of long time frames and lack of causal mechanisms). Agent based models are a natural candidate, but are rarely deployed at biographical timescales. We present ABMLS, a multi-layer agent architecture that extends the ASSOCC needs-model to biographical scale by using biographical events to change an agent's motivation structure as its circumstances and priorities evolve. We illustrate it through a model of single motherhood trajectories.

Agent-based modelling (ABM) has evolved through successive phases, from early `toy models' with simple behavioural rules, through a period of increasing cognitive sophistication, to a more recent emphasis on the structured contexts in which agents act. This paper reviews and synthesises this `contextual turn', arguing that it represents a substantive reorientation of ABM towards its core strength: the analysis of emergent social dynamics. The paper argues that further progress in ABM depends on deepening this engagement with social processes, particularly around structure, power, and meaning.