Modeling in the Social Sciences: Interdisciplinary Comparison

Building energy models result from interdisciplinary expertise and collaboration. In order to understand this, models are best seen as narrative devices, capable of integrating various ingredients and to address both research questions and policy initiatives. Shipworth's account of models as sausage machines that can potentially mix ingredients challenges us to reevaluate the epistemological consequences of the use of models as interdisciplinary tools. Models tell stories to different audiences, and through stories, they integrate available expertise to highlight the key findings or outcomes.

Model building in applied fields functions both as a mode of knowledge production and as a tool to inform policy. Building energy models are in use within and outside research communities. David Shipworth's article explores these models, their vernacular architecture and implications for interdisciplinary collaboration. By contrasting building energy models with epidemiological modeling, their similarities and differences will be clarified. Interdisciplinarity is an integral part of model architecture and a facilitator for research teams. Shipworth's notion of folk ontology, which favors "material, visual, physical and causal" approach to model building can be read in parallel with the account of models as narrative devices¹. I will discuss the nature of these models in relation to other cultures of modeling.

1. Motley Epistemology of Building Energy Models

Shipworth's study focuses on Bayesian Network models that estimate household energy consumption. These building energy models are constructed by elicitation and learning. By combining them elicitation is used to determine the structure of the networks, and learning to extract the probabilities from datasets. Elicitation relies on the use of experts from a particular domain. The variety of expertise (e.g. from physics, economy, sociology) present in modeling energy consumption can be problematic. Domain-specific expertise is required in the process of variable selection and measurement and model selection, for example. This heterogeneity reflects epistemological and ontological aspects of modeling.

Motley ontology refers to the way in which models "draw in a wide variety of sources," extending from theoretical assumptions to the modeling practice with "blood, sweat and tears of much trial and error" (Winsberg, 2001, 2009). Shipworth studies how motley ontology manifests in building energy models. He calls these models epistemic sausage machines, which "combine inputs of all qualities and types into outputs of homogeneous and indeterminate quality and type." This is in agreement with a view that modeling is baking a cake without a recipe (Boumans, 1999). Models are constellations of ingredients, such as theoretical assumptions, policy views, data, mathematization, metaphors and empirical facts (Boumans 1999). Both accounts try to capture the messiness of model-building process, yet lacking an aspect that is proven integral for application-driven modeling: intentionality or directedness. This is expressed as tailoring a model, which means building models for particular uses, often motivated by a policy call (Mattila 2006b). What, then, happens in building energy models, if they turn into epistemic sausage machines? I would argue that, as appealing as the metaphor seems to be, it may hide some of the concerns Shipworth raises when he analyses the vernacular architecture of these models.

Shipworth's main concern is the poor representation of people as users of energy. As he argues, "energy use is driven by occupants' needs and desires for energy services." He continues that "folk ontology" of the dominant modeling community in physics is an underlying factor for that. The modelers have a strong mental image of the energy use, and they see buildings as model systems, as "imagined concrete things." This prevents them actually seeing the occupants' needs as heterogeneity of energy use. Added with the difficulty of gaining data from the behavioral and social aspects of energy consumption, the vernacular architecture captures successfully the physical and the causal of the model systems, not the invisible or variable, such as human behavior.

Interestingly, representing human behavior by modeling seems to be a challenge to various communities, not only to energy modelers. Infectious disease modeling faces a similar challenge of representing fairly unpredictable human behavior in transmission models. Simple applications of heterogeneous mixing assume that humans contact each other as if they were gas molecules...