Affordable Access

The knowledge management arena: agent-based modelling of the pig sector

Authors
  • Osinga, S.A.
Publication Date
Jan 01, 2015
Source
Wageningen University and Researchcenter Publications
Keywords
Language
English
License
Unknown
External links

Abstract

Abstract belonging to PhD thesis: The knowledge management arena: agent-based modelling of the pig sector Sjoukje A. Osinga Wageningen University, Information technology group To be defended on 22nd of April, 2015 Promotor: Em. Prof. ir AJM (Adrie) Beulens, Information technology group Co-promotor: Dr ir GJ (Gert Jan) Hofstede, Information technology group Complex adaptive systems are characterized by multiple levels of behaviour: the behaviour of individual components and the behaviour of the entire system. In this thesis we study this relationship by means of agent-based models. By modelling individuals (agents) and their behaviour only, and simulating this behaviour over time, we generate emerging patterns: we did not explicitly put them in. We try to understand these patterns by reasoning back to individual level (multi-level analysis). Our application domain is knowledge management in the pig sector. Through a series of cases, we study the relationship between farmers' decision outcomes and their implications for the sector (bottom-up), and, vice versa, the relationship between sector-wide interventions and their effect on farmers' decision outcomes (top-down). Farmers make decisions based on knowledge, which diffuses through the population. We develop our agent-based models and the representation of knowledge throughout the thesis. Our final model is applicable to not only the pig sector, but to any sector with autonomous suppliers who need to make decisions based on criteria to be matched. A secondary aim of this thesis is methodological: to convey the merits of applying agent-based modelling to this type of multi-level research problem. Our cases concern each farmer's decision of which quality market to supply his pigs to (agent level). As outcome, we observe the spectrum of emerging quality market shares (sector level). Knowledge is assumed to be a prerequisite for market entry, and defined as everything a farmer needs to know to match the entrance criteria set by a market segment, as perceived by that farmer. Knowledge management refers to both the individual farmer's activities to coordinate a market's criteria with his own options, and the activities at sector level to influence all farmers' decision behaviour. One case addresses reproducing a well-known sector-level phenomenon (the pork cycle) by modelling individuals only. Other cases study the effect on emerging market shares of experimenting with agent-level properties: the amount of available knowledge and the conditions under which knowledge can be exchanged, and knowledge quality. The last case investigates the effect of experimenting with sector-level properties on individual farmer behaviour: two different policy interventions, and variations in demand. We apply multi-level analysis to seek explanations for emergent patterns in terms of individual farmer behaviour. Expert validation is used to evaluate the plausibility of model outcomes and explanations with respect to the real world. Results show that (1) the presence of sufficient knowledge in the system is more important than the network structure between knowledge exchanging agents for emerging quality market shares; (2) efficient knowledge management increases quality, but there is a limit to that efficiency; and (3) imposing policies on a sector the hard way is not necessarily more effective than making gradual changes, while the latter is more friendly for the individuals. Multi-level analysis proves to give added value to the results: in two cases, an unexpected pattern in model outcomes occurred, for which multi-level analysis could provide an explanation in model terms. Judged by the experts, the explanation for one of the patterns was deemed plausible in reality. In conclusion we can say that both varying individual properties and varying system-level properties result in responsive behaviour that can be explained in model terms, and that is to some extent plausible in reality. Knowledge representation power appears to differ per model. Dependent on the aim of the model, representation power can be kept deliberately modest (as in the pork cycle model), or can be rich (as in the final model, that allows representing different types of knowledge). We believe that the representation power of agent-based models make them sufficiently suitable to represent a real-world case, as long as the model has a well-defined purpose. We recommend agent-based modelling as a method, with multi-level analysis providing added value. We believe that extending this line of research is promising for any discipline where complex adaptive systems are object of study, of which knowledge management is an example.

Report this publication

Statistics

Seen <100 times