5 Modelling land reclamation in the Odra river catchment
5.4 Agent‐based social simulation setup for the case of land reclamation
5.4.3 The SoNARe model
Farmer agents use their individual profit to appraise their respective behaviour of the past simulation year. For this purpose they assess their profit as ‐1 (“too low”) or +1 (“ok”) with respect to an aspiration level (Simon, 1955) that reflects the expected minimum profit (a fixed threshold value is used for all farmers). This profit appraisal is memorised with a retention time in years which is fixed per farmer agent but heterogeneous among agents.
The memory structure of an agent is composed of a set of tokens that are triples of profit appraisal, LRS strategy, and retention time. For instance if agent A has a retention of 5 years then a memorised profit appraisal will persist for 5 years and after that be removed from the memory. When evaluating a past behaviour (pro/con LRS) farmers consult the relevant tokens stored in their memory, sum up the associated profit appraisal values and normalise to the codomain [0,1] such that values below 0.5 represent “negative economic success“
while values equal to or above 0.5 reflect “positive economic success“. During decision‐
making, this calculated economic appraisal (economicAppraisal) of an agent’s past behaviour is used to quantify its attitude towards its past behaviour (see section 5.4.3.3).
The following table summarises the relevant parameters and the values used.
Variable name Value used
Description
marketPriceOfAttainedCropYield Provided by SHAM; between 0.0 and 10.0 units productionCost 8.0 Investment in farming activities (excluding LRS)
per year.
The value was estimated in relation to the maximum attainable yield (market price is 10 units). Maximum possible profit is 20% (=2 units) of the market price of the maximum yield on a field.
costLRS 0.5 Amount (or work equivalent) per year that has to be invested in the LRS if it is maintained profitThresholdFarmers 0.5 If a farmer’s profit in a year drops below
profitThresholdFarmers then the profit (of that year) is considered “too low” otherwise “ok”
minRetentionTimeFarmers 3 Individual retention times are distributed heterogeneously among agents. Retention times are assigned to farmers from a uniform
random distribution from
[minRetentionTimeFarmers, maxRetentionTimeFarmers]
The random seed of the distribution is set to memRngSeed.
maxRetentionTimeFarmers 7
memRngSeed
Table 9. Economic success: SoNARe parameters, values, and descriptions.
5.4.3.2 Social networks
The agents’ social environment is modelled as a network. We investigate the exertion and perception of social influence as ways of “acting in” and “perceiving” a given social environment. In order to clearly isolate the effects of possible topological network dynamics
(adding or removing edges) we use a one‐layer and static social network. This network only serves as the infrastructure for perceiving and exerting social influence.
Farmer agent and the LRS initiator differ distinctly in the ways they are embedded in their social and physical environments. In the model, both agent types are embedded in a common acquaintances network. The evidence that an LRS initiator has a high degree of social network integration is covered by the fact that the agent is linked to all farmer agents (in a star‐like manner) whereas farmer agents possess direct social links (bidirectional) only to a fraction of other farmer agents (but the social links could span a number of hydrologically independent channels of the LRS). As the Odra case study suggests, most LRS initiator actors are not farmers themselves since they are e.g. village mayors or external advisors. Therefore, in the model these agents do not directly interact with the simulated physical environment. In contrast farmer agents continuously interact with the simulated environment by performing (or neglecting) local LRS maintenance and by obtaining feedback about attained profits from crop yields.
A farmer agent’s perception of social support is a function of the agreement or disagreement concerning LRS maintenance with its social network acquaintances. An agent receives a signal of support from each acquaintance that shares its opinion regarding LRS maintenance in that year, whereas it receives a pressure signal from each agent that has the opposite opinion. The exertion of social influence is strictly symmetrical in the sense that a signal of support and a pressure signal sent by the same farmer agent are identical in magnitude. Furthermore, the magnitude of the social signals sent by farmer agents is set to 1.0 for all agents.
The LRS initiator agent being embedded in the social acquaintance network participates in the general opinion dynamics as regards LRS maintenance only in that it exerts social support to farmer agents maintaining their LRS and sending pressure signals to those neglecting the LRS. While farmer agents continuously exchange social messages an initiator agent only becomes active when certain conditions are met, see section 5.4.3.3. Signals of social support or pressure sent by LRS initiators are higher in magnitude than those sent by farmers. The strength of social influence exerted by an initiator is set in relation to that of
farmers to a fixed value of 3.0, i.e. with the setting used the initiator is three times as influential as a farmer.
Like for economic success, the final indicator of an agent’s perceived social support is calculated as a normalised sum of all social influences such that values below 0.5 represent
“negative social support“ while values equal to or above 0.5 reflect “positive social support“.
The social support an agent perceives for his behaviour as regards LRS maintenance may be seen as an agent’s social appraisal (socialAppraisal) of his past behaviour. During decision‐
making this value is used as a proxy for the agent’s perceived subjective norm (see section 5.4.3.3).
The table below describes the parameters used and documents the respective settings.
Variable name Value
used
Description
networkType WS
Watts‐Strogatz network, small world network (ring substrate) generated by the RePast network factory with the following parameters:
rewiringProbability=0.1
connectRadius = avgAcquaintancesDegree / 2 RandomSeed is the seed of the RePast random number generator that is used during random rewiring
avgAcquaintancesDegree 10
RandomSeed
relativeInfluenceLevelInitiator 3 Strength of a social influence exerted by an initiator in relation to that of farmers, i.e. with the used setting the initiator is three times as influential as a farmer.
Table 10. Social networks: SoNARe parameters, values, and descriptions.
5.4.3.3 Decision making
The process for farmer agent decision‐making is illustrated in Figure 38. Based on their
(step 2). The definitions given in the previous two sections provide the quantifications of a farmer’s social and economic appraisal of his past behaviour. The resulting opinion of an agent on his past behaviour is formed as a weighted sum of the two appraisal values (step 3). We implement this balancing by adding a parameter that reflects the (socio‐economic) decision bias that a farmer agent has. The decisionBias of a farmer agent is represented as a value in the range of [0,1] where values above 0.5 stress the economic influence on decision making, values below 0.5 stress the social dimension. Since the two appraisal dimensions are normalised, the combined appraisal of a farmer agent is calculated as a weighted sum in which economicAppraisal is weighted with decisionBias and socialAppraisal is weighted with (1‐decisionBias). This combined appraisal is calculated as follows:
combinedAppraisal = economicAppraisal * decisionBiasFarmer + socialAppraisal * (1 – decisionBiasFarmers)
The formation of a behavioural intention is abstracted in the form of a Win‐Stay, Lose‐Shift heuristics (Nowak & Sigmund, 1993), i.e. an individual intends to keep to his previous behaviour (contributing to the collective action or not) if the combined appraisal is sufficiently high with respect to an aspiration threshold, otherwise it intends to shift to the opposite behaviour. For the Win‐Stay, Lose‐Shift heuristics (step 4) we use an aspiration threshold of 0.5, i.e. a farmer agent keeps to its previous behaviour (maintaining or not maintaining the LRS) if combinedAppraisal is above 0.5, otherwise the farmer agent shifts to the opposite behaviour. Farmer agents are assumed to follow their formed intention and change their behaviour if the conditions are met. Therefore, the selected behaviour is executed by forwarding it to SHAM where the taken LRS decision persists for the following simulation year for a farmer’s respective land parcel.
The LRS initiator is assumed to possess information about the farming success of its social network neighbours and decides to exert its social influence in favour of LRS maintenance whenever it perceives a minimum number of farmers who have big losses. The LRS initiator does not exert any influence otherwise.
Figure 38. Formalisation of farmer decision‐making. In step 1 an individual’s perceptions of his social and biophysical environment are updated. Based on the perception in step 2 attitude and social norm are formed. In step 3 attitude and social norm are weighted according to an individual’s socio‐economic orientation which results in a combined appraisal of past behaviour. In step 4 this subjective opinion of an individual on his past behaviour is evaluated in relation to an aspiration threshold resulting in an intention regarding subsequent behaviour. In step 5 the intended behaviour is executed in the biophysical environment.
Variable name Value used
Description
decisionBiasFarmers 0.5 Socio‐economic orientation of the farmer agents, values above 0.5 stress the economic influence on decision making, values below 0.5 stress the social dimension.
bigLossesThreshold 0 At the end of a simulation year the LRS initiator observes the profit of each farmer and counts those farmers whose profit has dropped below bigLossesThreshold. The initiator keeps a track of these counts over the past 6 years, if the average of this track rises above initiatorActivationThreshold then the initiator becomes active and exerts social influence. In all other cases he remains passive.
initiatorActivationThreshold 10
Table 11. Decision parameters and values.