NetLogo User Community Models
WHAT IS IT?
This is a simple model about landscape transformation by farmers. The model simulates on how farmers decide to practice particular land use system on their plots, given that they have 3 choices available. The basic premise of the model is that farmers adopt certain land use system based on their knowledge about profitability of the system and their ability to generate capitals for establishment.
Farmers’ knowledge about profitability of each land use system is shaped either by their individual experiences or information received from others. Information from others is received either by means of knowledge exchange among farmers or observing the performance of others’ plots directly.
Farmers exchange their knowledge via diffusion networks. Depending on farmers’ preferences in selecting partners, the network can be heterophilous, homophilous, centralized or decentralized. The network is homophilous if majority of farmers exchange their knowledge with partners of the same ethnic group or with relatively close neighbors. The network is centralized if majority of farmers attach to one opinion leader. Through these networks, farmers share or seek information. Based on ethnic similarity, relative distance and relative popularity of sources, partners’ credibility is judged by farmers. Farmers observe others’ plots within certain radius from their plots. Inaccuracy in observing performance of others’ plots is determined by “visibility” of information about performance of the system.
All collected information from individual experience, sharing, seeking and direct observation from others’ plots are stored in farmers’ “memory”. Farmers recall the data from their “memory” as the reference to update their knowledge. It depends on their memory-recall ability. At memory-recall ability equals 0, farmers use the most current collected information for the reference. The rate of adjustment depends on farmers’ progressiveness in trusting the reference. At progressiveness close to 0, farmers’ trust the reference at relatively very slow rate. Farmers’ knowledge is also enhanced by their exposures in extensions. Effectiveness of extensions in influencing farmers’ knowledge depends on farmers’ perception about credibility of the extension agents. Overall knowledge updating process in this model is called learning.
In connection with decision-making process, the model splits farmers’ knowledge into plot knowledge and general knowledge. Basically, plot knowledge is product of farmers’ book keeping about performance of existing land use systems currently practiced on their plots. It is pure knowledge of farmers from their plots without any influences from others. When the plots are renewed, the plot knowledge is reset into 0. General knowledge is farmers’ knowledge about all options of land use systems, which are available in the landscape. This type of knowledge is fusion of knowledge from their individual experiences and knowledge from others, including knowledge from extension and knowledge from interaction. General knowledge is more permanent than plot knowledge, in sense that physically, it is maintained in farmers’ “memory”. Although farmers do not practice a land use system anymore, it does not imply that they have no general knowledge on that system.
Solely, general knowledge is used as the basis to make strategic decision. It is a decision to select the best land use system when farmers need to renew their plots. Together with plot knowledge, general knowledge is used as the basis to make tactical decision. It is a decision to allocate labors to existing plots for harvesting. Both strategic and tactical decisions are determined by prioritization degree of farmers. At prioritization degree equals 1, farmers consider or allocate labors proportionally according to relative profitability of each land use system. At prioritization degree more than 1, farmers consider or allocate labors to the most profitable land use system and ignoring the less rewarding options. At prioritization degree close to 0, farmers consider or allocate labors to each land use system uniformly. It illustrates a situation where farmers are facing dilemma, have no ideas about what to do with available options.
In the model, farmers are decision makers. They are the heads of household. They are landlords of their farms. They have power to generate and to mobilise labor capital. They have power to generate financial capital and use it for renewing old plots. A farmer can occupy more than one parcel of farms. A farmer can adopt more than one land use system on their plots. A farmer can have more than one worker. Other than knowledge, labor and financial capitals, adoption of certain land use system is limited by convertibility of a land. In this model, convertibility of farmers’ plots is defined by age threshold. Farmers can renew their plots after the existing plots are not productive at all.
This model is developed as learning and/or research tool to explore barriers on adoption of a land use system that we want to promote. The barrier can be caused by unfavourable conditions found in information diffusion network, in farmers’ personal aptitudes (their mental properties), in extension strategy, in biophysical/economical performance of the land use system or due to lack of capitals. Thus, using this model, we can explore possible drivers that can help us in boosting the adoption process of better land use system option. It can be used to explore favourable condition where critical mass in adoption process can be achieved.
HOW IT WORKS
1. Scales of the model. In term of interaction captured by this model: decision-making is made at household level; spatial dynamics is at plot level; and temporal dynamics is yearly. In term of emergent behavior as results of individual interactions in this model: decision-making should be observed at community level, spatial dynamics should be observed at micro rural landscape; and temporal dynamics should be observed at longer time frame.
2. Number of farmers is generated based on user-defined input. They are composed from 3 ethnic groups, which fractions are user-defined input. In their neighborhood, farmers can be agglomerated based on ethnic group.
3. Diffusion network is formed according to attractiveness of partners, which is defined by equation:
· p is partners’ relative popularity (0<=p<=1)
Links among farmers in this network are randomized based on attractiveness values using random lottery algorithm borrowed from NetLogo. For initialisation, popular persons (or opinion leaders) are assigned by fraction. In the dynamic part, popularity is later judged by number of current attachments linked to farmers relative to maximum attachments found in the whole community. When w1=1, w2=0, w3=0, the network is centralized to opinion leader. When w1=0, w2=1, w3=0, the network is ethnical homophilous. When w1=0, w2=0, w3=1, the network is homophilous based on neighborhood. When the neighborhood is agglomerated based on ethnic group [No. 2], the last setting forms ethnical homophilous network.
4. Each farmer may have more than 1 parcel of farms at various sizes. Position of the farms can be agglomerated based on landowner.
5. Initial patterns of the landscape are generated based on user-defined input.
6. Number of farmers, number of plots, landholder of each plot and size of plots are static over the whole simulation. No population dynamics of farmers, no land exchange among farmers, no land expansion and no modification to plot size.
7. Three choices of land use systems are provided. All land use systems are assumed as tree-based systems. Productivity of each land use system is determined by age. To simplify, each land use system is stratified into 4 production stages: pioneer (stage 1), early production (stage 2), late production (stage 3) and post-production (stage 4). Age of each plot is always updated during simulation and reset into 0 when farmers renew their plots. Production stage of each plot is defined by time bound to reach each stage, based on user-defined look up. Productivity of each plot is randomized based on given statistic of each land use type at each production stage. Productivity should be expressed in attainable yield unit [yield-unit.person-1.day-1.area unit-1]. External factors affecting plots’ productivity (e.g. bad weather, pest and disease) are not yet included. But, seasonal variability of yields is covered by coefficient of variance in user-defined statistic of productivity. Causal relation between land use system and site quality are not yet considered.
8. Farmers exchange their knowledge with partners through diffusion network. Willingness of farmers to share or to seek information from their partners is driven by probability. At prisoner dilemma game situation, probability of sharing and probability of seeking equal 0.5. Credibility of information from others is defined by similar equation in No 3:
· p is partners’ relative popularity (0<=p<=1)
9. Farmers observe others’ plots within specified radius. It depends on relative distance of farmers observation range from their plots. Visibility of performance of each land use system determines information accuracy from observation.
10. Average of information collected from individual experience, sharing, seeking and direct observation from others’ plots are stored in farmers’ “memory”. Farmers recall the data from their “memory” as the reference to update their knowledge. It depends on their memory-recall ability. Moving average code borrowed from NetLogo is used in this procedure.
11. Plot knowledge is knowledge of farmers about performance of particular land use system currently practiced on their plots. When a plot is converted into other types, plot knowledge is reset into 0. When land use system is sustained on given plot, plot knowledge is updated using this equation:
Kp[i,j,t+dt] = Kp[i,,t] + a*[Ip[i,j,t]-Kp[i,j,t]]
· Kp[i,j,t+dt] is knowledge about land use system i currently practiced on plot j at time t + dt
12. General knowledge, expressing knowledge of each farmer about performance of all land use systems available in the landscape, is updated using this equation:
Kg[i,t+dt] = Kg[i,t] + a*[Ig[i,t]-Kg[i,t]] + c*[E[i,t]-Kg[i,t]]
· Kg[i,t+dt] is general knowledge about land use system i at time t + dt
13. Based on general knowledge and plot knowledge, farmers make tactical decision to allocate labors for harvesting. Labors [persons.days] are dynamically generated. Thus, labors capital may vary from one farmer to other farmers, from one year to the next years. Labor is allocated using following steps.
Firstly, based on general knowledge, farmers make macro labor allocation, allocating labor into each option of land use system, using this equation:
fLM[i,t] = ((o?[i]*(Kg[i,t]^p))) / ((o?*(Kg[i,t]^p)) + (o?*(Kg[i,t]^p)) + (o?*(Kg[i,t]^p)))
· fLM[i,t] is labor fraction allocated for land use system i at time t
In the case when farmers have no general idea about all systems or when Kg[1,t] = Kg[2,t] = Kg[3,t] = 0, farmers allocate their labors according to existing systems they are practicing currently. Thus:
fLM[i,t] = o?[i] / (o? + o? + o?)
Secondly, based on plot knowledge, farmers make micro labor allocation, allocating labor spatially into each plot, using similar equation:
fLm[i,j,t] = Kp[i,j,t] ^p /( Kp[i,1,t] ^p + ... + Kp[i,n,t] ^p)
· fLm[i,j,t] is labor fraction allocated for land use system i on plot j at time t
In the case when (Kp[i,1,t] ^p + ... + Kp[i,n,t] ^p) equals 0, farmers allocate their labors uniformly, depending on plot number (n) where land use system i is adopted. Finally, number of labors allocated to harvest land use system i on plot j at time t is fLM[i,t] * fLm[i,j,t]*LT[t]. LT is total labor capital available at time t.
14. Harvested yield is calculated from each plot using equation:
· L[j,t] is harvesting labor allocated for plot j
15. Profitability of each parcel of farms is calculated using equation:
(Y[i,j,t]*$1[i,t] - L[i,j,t]*$2[t] - $3[i,j,t]* S[i,j]) / (L[i,j,t]* S[i,j])
· Y[i,j,t] is harvested yield from plot j at time t, where land use system i is adopted
16. Based on general knowledge, farmers make strategic decision to adopt certain land use system to renew their plots. Equation to calculate attractiveness of each land use system to be adopted is similar to equation to calculate macro labor allocation [No. 13]:
fS[i,t] = ((c?[i]*(Kg[i,t]^p))) / ((c?*(Kg[i,t]^p)) + (c?*(Kg[i,t]^p)) + (c?*(Kg[i,t]^p)))
· fS[i,t] is fraction of consideration given for land use system i at time t
In the case when farmers have no general idea about all systems or Kg[1,t] = Kg[2,t] = Kg[3,t] = 0, farmers consider each system according to capital availability. Thus:
fS[i,t] = c?[i] / (c? + c? + c?)
This equation implies that at given time when farmers need to renew their old plots but they have no knowledge about any systems or when they have no ideas in selecting the best choice (when p = 0), farmers may adopt particular system when the capital to establish that system is available. Using this equation, the model can explain some cases where adoption of particular system at early stage works through incentives disbursement to establish that system as such, without any efforts related to knowledge diffusion. For example, project approaches that promote particular land use system usually start the promotion by disbursing subsidies to farmers.
17. Suppose that a farmer, in order to renew their plots, consider each system uniformly, thus the fractions are 1/3 each. A pot of numbers representing the system id according to consideration fraction given by the farmer is made. The pot contains 100 times number of plots to be renewed. Suppose that a farmer need and able to renew 2 plots. Thus, in this case, we will have a pot of numbers containing: 66 seeds with values equal 1, 66 seeds with values equal 2 and 66 seeds with values equal 3. This pot is shuffled, and a number is picked up randomly from the pot to define which type of land use system that the farmer finally adopts on given plot.
18. At last, dynamics of each plot is updated. Ages of renewed plots are reset to 0, and their land use type is adjusted to the new one. Ages of sustaining plots are increased by 1.
HOW TO USE IT
SETUP button: sets up the model by initialising farmers and their plots. Input parameters of the model are read from procedure read-input-data. In another version, the data are read from external file.
GO button -- runs the model.
SIMULATION-DURATION slider -- lets you determine the simulation duration.
ZOOM slider -- lets you scale the size of plots or farmers for visualization.
SHOW-NETWORK? switch -- ON switches the view into diffusion network of farmers. OFF switches the view into plot dynamics.
LAYOUT-NETWORK button -- attempts to move farmers around to make the structure of the network easier to see. Use the button before or after the simulation. This procedure is borrowed from preferential attachment model by NetLogo.
PLOT KNOWLEDGE plot -- .shows overall average of plot knowledge of farmers about each land use system.
GENERAL KNOWLEDGE plot --. shows overall average of general knowledge of farmers about each land use system.
LABOR plot --. shows overall average of allocated labors on each land use system.
LAND AREA plot --. shows total area of each land use system
THINGS TO NOTICE
During the simulation run, you can switch the view from patch dynamics into communication network dynamics. When you switch to communication network dynamics, you’ll see how farmers are interacting each other. Grey links illustrates no interaction. White links illustrates farmers are only sharing to their neighborhood. Yellow links illustrates farmers are only seeking from their neighborhood. Pink links illustrates farmers are exchanging knowledge with their neighborhood, both through sharing and seeking in one go. Size of farmers represents relative communication partners that they have. Bigger the farmers, more partners they have. Color of farmers illustrates their ethnic groups. Red is ethnic group 1, green is ethnic group 2, and blue is ethnic group 3. If you simulate many farmers, you may find difficult to see the network clearly. Use layout-network button to ease the visualisation. But, during the simulation, the layout is always reset into farmers’ original location for computation purpose. Thus, it’s better to layout the network before or after the simulation run.
When you switch to patch dynamics, you will see visualization of farmers’ plots. Random irregular shapes of each plot illustrate that this model consider spatial representation of plots as objects, not as raster. In this regard, size of the plots represents the relative size and the color represents their land use types. Red is land use system 1, green is land use system 2 and blue is land use system 3.
The purpose of this model is to observe emergent behavior at larger scale than individual interaction captured in the model. Thus, attributes of each individual in the simulation should not be parameterized at precise individual level. Instead, we use average of larger scale by still respecting individual variability. This model lets some parameters of each individual changing over time, with respects to uncertainty. For instance, prioritization degree or learning progressiveness of each individual may not be the same from time to time. In fact, there is no evidence that human being are having constant mental properties over time. Just look at us. We may find ourselves that one day we were fast learners, having high prioritization degree, but on another day we totally behaved contrastingly. Shortly, those properties are in fact uncertain. Therefore, let them being uncertain.
THINGS TO TRY
1. To get to know the model’s behaviour at the first time, modify input data in procedure read-input-data. Imagine that there are 10 farmers living on a neighborhood, composed by 3 ethnic groups uniformly. Suppose that initially, the landscape is only dominated by land use system 1 and all farmers are only able to generate capital to establish land use system 1 (with constant probability = 0.5). Let all farmers only learn from their own plots. Let them never exchange their knowledge each other, neither through sharing, seeking, direct observation nor extension. Now, just let other parameters random. Run the model using this setting several times, and notice the main results from 4 provided plots. You should see static pattern of the landscape. This example explains that without any interaction with each other and without any capabilities to generate capital to establish other options than land use system 1, farmers only learn about existing land use system from their own plots.
3. Now, allow farmers to have capability to generate capital to establish land use system 2 and 3 (with constant probability = 0.5). Imagine that a project is coming in to the village, promoting those two new systems by giving subsidies. Other parameters are maintained the same as the previous scenario. Run the model using this setting several times, and again, notice the main results from 4 provided plots. You should find some farmers experimenting land use system 2 or 3. But, land use system 1 still predominates the landscape. It seems hard for land use system 2 or 3 to totally replace land use system 1. This example explains that without manipulating farmers’ knowledge, adoption of new systems by early adopters can work through incentive, especially when farmers have prioritization degree less close to 0, a situation when farmers face a dilemma in decision-making.
4. Next, open the communication network, so that farmers can share, seek and observe each other. But, we do not bring extension agents into the world. We use prisoner dilemma game in this communication network by setting the probabilities to seek, share and observe into 0.5. You should find that it’s easier for land use system 2 and 3 to achieve their critical mass, although they are still competing with land use system 1 at their tipping points.
5. Finally, your mission is to promote land use system 2 into the landscape.
You should construct input parameters in a tab delimited text file to run the model using defined format. Simply use Adopt&Learn.xls file to parameterize the model. The file should have following input parameters:
EXPOSE in extension
INITIAL land use patterns
VISIBILITY by others
NON harvesting labor costs
EXTENDING THE MODEL
The model can be extended with regards to the scale of complexity that we want to explore. We can add modules to consider site quality dynamics and link it to productivity, elaborate the economical modules by tracing the flow of revenue into financial capital and link it back to capital generating ability to adopt new systems, add some toolboxes to assess environmental consequences of emergent landscape patterns, and so on. Although this model is developed using decision-making scale at household level, but the main purpose is to explore emergent behavior at larger scale, how the majority adopt certain land use system.
· This model uses breeds to represent farmers, their communication links and their plots. The model runs slow at number of fields close to 100.
The FALLOW Model (van Noordwijk, 2002), Preferential Attachment Model (Wilensky, 2005), Ethnocentrism (Wilensky, 2006).
CREDITS AND REFERENCES
This model is a NetLogo version of adoption and learning (Adopt&Learn Model), authored by Meine van Noordwijk and Desi Ariyadhi Suyamto, ICRAF Southeast Asia. Initially, it was developed as a system dynamics model using STELLA at a community scale. Recently, it is migrated to NetLogo due to some needs to capture individual interactions among farmers. The current conceptual framework of the model is inspired by:
· Ball, P. 2004. Critical Mass: How One Thing Leads to Another.
The development of Adopt&Learn Model is part of research activity funded by Common Fund for Commodities (CFC)-Improving the Productivity of Rubber Smallholdings through Rubber Agroforestry Systems (CFC/IRSG/11) Project. The task of this modelling exercise is to explore the barriers of adoption process of rubber-based system promoted by World Agroforestry Center in some areas.
Thank you to Uri Wilensky and the team by facilitating us with this powerful tool. Also thanks to Luis E. Garcia Barrios (firstname.lastname@example.org) from Alternativos-ECOSUR Mexico, who is the first person who share knowledge about NetLogo to us. It was on May 26, 2006 at the airport in West Sumatera, Indonesia.
Corresponding author for the Adopt&Learn Model: Desi Ariyadhi Suyamto. World Agroforestry Centre, ICRAF Southeast Asia Regional Office, Email: DSuyamto@cgiar.org. Tel: +62 251 625415 ext 791. Fax: +62 251 625416. http://www.worldagroforestrycentre.org/sea or http://www.icraf.cgiar.org/sea. A Future Harvest centre supported by the CGIAR.
Concept review. Latest update: August 18, 2006.
(back to the NetLogo User Community Models)