MandE NEWS .
Edited by Rick Davies, Cambridge, UK. | Email the Editor | Last Edited: April 2006 | Home Page

PLEASE NOTE: The MandE NEWS website has recently undergone a major reconstruction. The new (>May 2008) version of this page is now located  here. The page below will no longer be updated.
Developing network models of development projects: An introduction                   
Contents (a work in progress, last updated  28/04/06)
  1. The purpose of this webpage
  2. Where are network models likely to be useful?
  3. What is a network? 
  4. Quick examples: Some networks seen in development projects
  5. Network models and the Logical Framework
  6. Two complimentary ways of representing networks: diagrams and matrices
  7. Three ways to develop network models
  8. Describing relationships within a network matrix or network diagram
  9. Using one and two mode networks
  10. Summarising the data available within a network matrix
  11. Analysing the structure of simple networks
  12. Thinking about types of network processes
  13. Developed examples of network models (plus links to example pages)
  14. Questioning network models about the wider context
  15. Collecting network data
  16. Using Social Network Analysis software
  17. People's participation?
  18. Available datasets
  19. Structure and agency in network models
  20. Complexity theory and network models
  21. Useful online resources on network analysis
See also the NetworkEvaluation emailing list


1. The purpose of this webpage to explain how network models can be used to describe development projects. Descriptions can be useful for documenting an intended set of activities and outcomes, and for documenting the actual activities and the outcomes that eventuate. An evaluation will normally need to develop and compare both types of description. 

There are some related webpages that are relevant:

  • Network perspective on development projects: More than a metaphor. A 2004 conference paper by Rick Davies. This provides an overview of why network perspective is relevant, and some of the wider implications of taking a network perspective. 
    • Also relevant: "Scale, Complexity and the Representation of Theories of Change Part  I and Part II", in Evaluation, Vol 10(1):101-121. (2004) and Vol 11(2):133-149, (2005), Sage Publications, London. Request a copy
  • Evaluation and analysis of networks, a section of the MandE NEWS website. This includes a link to "Networks and Evaluation" emailing list with approximately 100+ members
  • Working with the Logical Framework, a section of  the MandE NEWS website. Network models can provide a complement to, and an alternative to, Logical Framework descriptions of development projects.

2. Where are network models likely to be a useful means of description? (return to Contents)

Network models are likely to be useful in any of the following kinds of settings:
  1. Where there are many actors (people and / or organisations) who are fairly autonomous and where there is no central authority able to direct the behavior of all the other actors.

  2. In large projects with many stakeholders, rather than small projects with few, where a single authority is less likely to be found. National and international level development activities are likely to have larger numbers of actors, without there being one over-riding authority.

  3. In projects where there is no single objective, but many alternative and/or competing objectives. These may be symptomatic of the absence of a central authority, and / or the intention that the participating parties have considerable independence. 

  4. In projects where a given output may be used by many actors and a given actor may use many outputs. In other words, where there is a complex web of relationships, not simple one-to-one connections.

  5. In projects deliberately designed to function as networks. These will vary in the extent that they are designed to be centrally managed  ( /coordinated / facilitated) or not.
Network models can also be useful as a means of describing projects that are more "traditional", where there is a single agreed objective, clear lines of authority, clearly differentiated responsibilities, and which can be describe by a temporal logical model (e.g. a Logical Framework). Here network models can provide supplementary detail that cannot be easily represented otherwise. See section 5 below for more on this topic.

3. What is a network? (return to Contents)

A network is a collection of people and / or things that are connected to each other by some kind of relationship. Many kinds of entities can be part of a network: people, projects, documents, events, organisations, cities, countries, etc. And there are many kinds of relationships that can link such entities, involving transmission or exchange of information, money, goods, affection, influence, infection, etc.

Note: A network does not need to be labelled or formally named as a network to be a network. Such networks are part of a much larger set of networks, some of which may be recognised as de facto or informal networks, and many others may not be.

Note: Because this webpage is concerned with the use of network models for planning and evaluation purposes the focus will be on networks of  actors, objects and events that are observable. That can be interviewed or that can be read, or that can be read about. So network models of abstract processes will not be discussed here. (PS: Here is an example of a network of abstract processes that was part of the ToR for an evaluation of a complex "project" (multi-donor budget support). The challenge with this network would be to establish linkages between events which, as described in this example, are themselves not readily observable)

Network analysis is the analysis of the structure of relationships within a network. Social Network Analysis (SNA) is especially relevant to the development of network models of development aid programs, because development aid is about people and their institutions. It is the main intellectual influence on the contents of this webpage. Network analysis is also done of biological systems, physical systems and economies, but those usages are not discussed here.

4. Examples of networks that can be seen in development projects (return to Contents)
  • A network of international donors supporting various NGOs within a particular country

  • A network of NGOs within that country, who have contact with each other, work with each other and who may also compete with each other

  • Within an NGO, a network of staff, who are connected formally and informally

  • Within an NGO, a network of activities which form different kinds of business processes, that generate different types of services. Such as workshops, training events and email newsletters

  • A network of donors and NGOs linked by common policy concerns, such as specific objectives within the national poverty reduction strategy

  • A network of policy documents, linked by over-lapping sets of  indicators of achievement.
See section 12 below for a wider range of examples of network models
5. Network models and the Logical Framework (return to Contents)

 A Logical Framework is kind of temporal logic model, that describes development projects in terms of a chain of "if we do this...and this ...assumption holds, then this ....will happen, and if this happens..." statements

There are two reasons for looking at the relationship between temporal logic models and network models. Firstly, for readers who are more familiar with the use of the Logical Framework as a means of describing a project, it may be useful to use the Logical Framework as a starting point for developing a simple example network model, which can then be elaborated into more detail.  Secondly, inter-operability. There are many circumstances where a Logical Framework description of a project will be required by donors and or senior management, and many situations where Logical Framework descriptions are not sufficient to capture and help manage the complexity of a project. It should therefore be possible to convert a Logical Framework description of project into a Network Model of a project, and convert a Network Model of a project into a Logical Framework description of project. 

As argued on the Logical Framework page on this website, a Logical Framework description of a development project can be improved upon by making it more "actor oriented". That is, by clearly identifying who is involved at each level of the Logical Framework's description of a project. Then by identifying how these actors are expected to interact with each other. Doing this makes it easier for everyone to understand the overall storyline that should be connecting Activities to Outputs to Purpose to Goal. Doing this also helps make it possible to develop a network model of the same project design, as will be shown further below.

Here is a very simple actor-oriented interpretation of a Logical Framework, representing an imaginary development project that can be made more complex, and realistic, later on. Read from the bottom of the table upwards.

The Actors involved And how they relate to the levels of a Logical Framework  
Other village members Who interact with the VDCs, and this leads to changes in their lives, in the longer term. (Goal level  outcomes / impact)
Village Development Committees (VDCs) Use the goods or services provided by the NGO (Outputs) and this results in medium term changes (Purpose level outcomes)
NGO Undertake various Activities, some of which generate some Outputs: goods and services that are usable by the VDCs
Donors Provide Inputs to the NGO (these are sometimes listed in OVI column next to Activities)

In the Logical Framework the Assumptions column also has an important role to play, especially at the Output and Purpose level. Assumptions, statements here usually relate to external influences that could effect the linkage between events at Output and Purpose level, and between the Purpose and Goal level. While often described as abstract processes , it is possible and useful to describe these Assumptions in more actor-oriented terms, by identifying specific organisations, groups or individuals who could have a positive or negative influence on the actors in the main narrative column (as in the table above). Doing so brings us a further step closer to a network view of development projects. See more on this below...

Temporal logic models, such as the Logical Framework, vary in the number of levels or stages they have within their model. While the Logical Framework has four, there is no reason why an actor oriented interpretation of the Logical Framework could not have more, if the chain of actors linking the project managers with the final intended beneficiaries was longer.

[For more on more actor-oriented Logical Frameworks, see comments on the Logical Framework]

6. Two complimentary ways of representing networks: diagrams and matrices (return to Contents)

In Social Network Analysis (SNA) a network can be represented in matrix form and in the form of a network diagram. A network diagram can be converted into a matrix, and vice versa. The network diagram and network matrix shown below both represent the relationships between the actors involved in the table above. Note that the convention with such matrices is that the cell entries always show the relationship that exist from the row actor to the column actor. In this example the relationships between all the actors in the network are two way, so the matrix is symmetrical. But this will not be the case if the focus is on funding relationships or on influence processes, for example (see more on this below).

Network diagrams are good for providing a overview of what is happening. Network matrices are good for providing detail that can be systematically analysed. More on this below.

logical framework model

7. Three ways to develop network models (return to Contents)
  1. Top-down: Start with broad categories of actors and map out the expected linkages between them. Then break these categories down into smaller categories of actors. Then map the relationships between them, and with others. This can be done using network diagrams to start with, or by matrices

    The network model shown above is clearly a very simple view of a development project. It is a useful starting point, but needs more detail if it is to be of operational use for planning and evaluation purposes.  Looking at each of the cells in the matrix above, it is easy to see that each cell could be developed into a matrix of its own.  Two examples:

    A. The cell showing the linkage between the NGO and the VDCs could be developed into a new more detailed matrix, where the left column listed the NGO Community Development Workers (CDW) and the top row listed the various VDCs the NGO was working with. The cell values could indicate the percentage of time each CDW expected to spend working with each VDC. Many of the CDWs might be expected to work with multiple villages,  but addressing different development tasks.

    B. The cell showing the NGO linked to the NGO could be developed into a new more detailed matrix, where the left column and the top row both listed the NGO staff (CDWs and other staff). Cell values could describe the relationships between all the staff. Cell values could code types of working relationships, or percentages of the row actor's time that will be spent working with each other actor.

    Note: Not all the relationships in the original matrix above would need to be expanded in detail, in this way. Such a process would be time consuming and unnecessary. The choice of which cells to expand into new more detailed matrices should be a strategic choice, reflecting a sense of what are the most important relationships within the whole model, which need more detailed planning and description.

    A matrix that is used to represent a whole set of more detailed matrices, has been called a "meta-matrix" by Krackhardt and Carley, 1998  This was in the context of developing quite complex computerised models of organisational processes, which are not relevant to task being addressed here by this webpage. The most immediate use of a meta-matrix is as a means of developing a simple network model of a development project to start with, then deciding where to selectively develop the model in more detail. 

    It is also possible to scale up as well as down, using the same technique, and place the project in a wider context. The network matrix shown above could be seen as one cell in a larger matrix, such as one showing a range of projects in a given country and their interactions with each other. Use this way,a meta-matrix can provide a visible set of links between network models developed at different scales and locations. Contra the Logical Framework, network models are scalable!

  2. Over time: Start with the actors who will be involved at the beginning, and map out the expected relationships between them. Then add in the new actors that will get involved, each x period of time. And map the relationships they will have with the existing actors. There are two options here, to develop: (a) A cumulative model, that shows all actors and relationships that have existed up to the final period of concern, (b) A consecutive model, in the form of a series of "slides", showing pictures of the network at different points in time. This could require change or removal of old relationships and actors, as well as addition of new actors and relationships.  

  3.    Opportunistically: Document the relationships that exist at this moment according to whatever data is available. Then ask the participants (a) How this network structure relates to their original plan, of  what they intended to see happen when they first got involved, (b) How well does the network structure represent what is actually happening and the moment, (c) How they expect this network structure to change by the end of x period.  Make sure they can comment not only the structure of the relationships that are mapped, but on alternative types of relationships (not shown in the network diagram) that might be more important to them, and which should be mapped.  Many of the example networks shown in section 12 below have been opportunistically developed.

8. Describing relationships within a network matrix or network diagram (return to Contents)

The matrix that has been shown above has a very simple description of the relationships involved. It simply states whether a relationship exists or not, not more. For the purpose of project planning and evaluation a more detailed description would be needed. In Social Network Analysis cells in a matrix can be used to describe many aspects of relationships:

  • Existence of a relationship: Described by using a 1 or 0 in a matrix (as above), or the presence or absence of a link in a network diagram

  • Type of relationship: Described by using numbers, such as 1, 2, 3,... to indicate the presence of different types of relationships that have been pre-coded.

  • Frequency of interaction: Described by using numbers to indicate frequency per period or in total. Or by indicating the relative proportion of an actor's time spent on each relationship.

  • Value of the relationship: Described by using numbers to signify a rating or ranking of the relative value of different relationships

  • Sequence of the relationships: Described by using numbers (e.g. 1, 2, 3) to representing a sequence of events over time, or dates representing actual times

  • Details of a relationship: In small matrices the cells can contain text descriptions of the relationship
The same information can be shown in network diagrams by colour and size coding the links between the actors

The choice of what aspects of a relationship to map is a strategic choice, of one amongts many, that needs to be informed by a theory, or view, of what is most important in the network being modeled

 Cross and Parker (2004) have focused on information flows within organisations, and how to inquire about them. Consistent with the thrust of this webpage, they have commented that "We think that one trend in network analysis  will be towards mapping different, theoretically important dimensions of relationships". Some of the candidate dimensions of relationships they have suggested need exploration are:

Who do people contact for task purposes - people who provide information , resources or diretcion that helps us get work done

Who do people contact for career development (learning) - people who give feedback that is helpful for our career development

Who do people contact for career support (political support) - people in influential positions who are advocates and provide political support

Who do people contact for sense making - people who help make sense of rumours, events, or gossip

Who do people contact for personal support - people who help us cope with and recover from troubling situations at work or personal dilemmas.

Who do people contact for purpose - people who make use feel that what we do at work matters, that our work has meaning

9. Using one and two mode networks

There are two main kinds of matrices that can be used to describe networks:

  • One-mode (symmetric or adjacency) matrices: For example a matrix showing how 10 NGOs were related to each other. The same list of NGOs would be shown down the left column of a matrix, and across the top row. The matrix shown above is a one-mode matrix. The same set of actors are shown in the top row and left column. The matrix shown in section 6 above is a one-mode matrix.
  • Two-mode (asymmmetric or affiliation) matrices: For example, a matrix showing which of the 20 objectives in a government poverty reduction strategy that each of the 10 NGOs was addressing through its research and advocacy work. Here the NGOs would be shown down the left column of a matrix, and the policy objectives would be shown across the top row. Each cell could contain a number signifying the relative importance of a given policy objective to a given NGO. Two-mode networks can be more interesting, and useful as development interventions, because participants' knowledge of the whole network is usually less complete than in one-mode networks. Constructing and then then sharing the network model can be a development intervention in itself.

Within the matrix version of the Logical Framework shown in section 6 above, we can see a number of potential one-mode and two-mode matrices. The cells in the diagonal (upper left to bottom right) show relationships between the same kinds of actors (NGOs with NGOs, Donors with Donors, VDCs with VDCs, etc). As noted above, each of these could be developed into a matrix of its own. These would be one mode-matrices, because the same actors are listed on both sides of the matrix. All the other individual cells could also be developed into matrices, but these would be two-mode matrices, because the actors on the side of the matrix would be different from those listed across the top.

"Multiplex relationships" is a term to describe where two actors might have multiple kinds of relationships with each other. For example, in aid agencies people will be connected through formal organisational structures, and informal through friendship and other ties. Multiplex relationships can be shown in two mode matrix format, that has a one-actor x multiple-other-actors structure. Each row will then show a particular kind of relationship that actor has with all the other actors. For example, an NGO may produce many kinds of information products, for a range of other organisations. The left column could list those products, and the top row could list the organisations using them, or expected to use them. Cell entries in each row could indicate whether a given actor uses, or is excpected to use a given product.  Multiplex relationships can also be shown in network diagrams by color coding different types of linkages.

9. Summarising the data available within a network matrix (return to Contents)

There is a third strand of network analysis, in addiiton to matrices and diagrams. These are mathematical measures of the structure of networks. Many are far too complex to be of day to day use in the development of network models of development projects. There are however some simple measures which can be useful, and which are outlined here.

1. Using summary rows and columns
A matrix full of numbers can be daunting, especially a large matrix with many actors. How do you make sense of all that data? One simple way forward is to make use of  a summary row (at the bottom), and a summary column (to the right).  Here below is another version of the matrix already shown above. The cells in the summary column count the number of links each row actor has with all the column actors. The cells in the summary row count the number of links each column actor has from all the row actors.

PS: Because each relationship in this matrix is a two-way relationship, the row and column totals are the same. But if the matrix described funding relationships between the actors (a one way relationship) this would not be the case. 

2. The same matrix could include valued relationships with each cell, describing the relative importance of the column actor to the row actor. In this case we can use two types of summary rows. These are shown in the matrix below. 1 = highest priority, 5 = lowest.

Here the links to the VDC are the most important links , and the links to the Others (Output Assumptions) are the least important. Remember that in this example, low rank numbers = high importance.

In network diagram versions of the same matrix, it would be common to show the relative importance of the different links, by varying the thickness of lines signifying a link. And the size of the node representing each actor could vary according to the number of links it has with others.

3. Introducing actor attributes, to weight the importance of relationships

We can make the picture more detailed (and more realistic), by introducing another category of information into the picture. So far all the data inside the matrix describe the relationships between the actors involved. This is the traditional focus of social network analysis. However, we can also  introduce some data about the attributes of  each of the actors involved. These attributes could be the size of the group (organisation staff or group members), the resources they have available, their relative status or importance,  their willingness to become involved in a project, or any other measure relevant to the expected success of the project.  These attributes may make a big difference to what happens within a given relationship (as specified within the matrix).

The matrix below includes some imaginary attributes of each of the actors, on the far left. The cells combine these with the relationship values taken from the matrix above.  The second summary row then shows the combined "effects" of these weighted relationships on each column actor. 

The introduction of weightings has changed the picture, with the linkages to "Other Villagers" now having the highest average importance, compared to the linkages to the VDCs which were more important previously. Again, remember  that in this example, low rank numbers = high importance, and vice versa!

The wider relevance of  including actor attributes in network models

In the Logical Framework, and other Logic Models, there is a causal chain of expected events: Activities + Assumptions = Outputs, Outputs + Assumptions = Purpose, Purpose + Assumptions = Goal.

In network models the equivalent is a process of expected influence in the form of:  Actor + Relationship + Actor + Relationship + Actor... [but one involving multiple interacting actors, not a simple linear sequence]

Giving actors numerically valued attributes, and combining them with relationship values, allows us to convert this text description into a quantified description. If the same actors are also shown in other matrices, with links to other actors, the effects of their complex interactions can be traced over longer distances.

In the simple example above, the actor attributes were judgements about their relative importance. Other significant attributes could be their size (in staff or budget), if they are organisations.

10. Analysing the structure of simple networks (return to Contents)

There are a number of measures of network structure that have been developed within the field of Social Network Analysis that may be helpful when analysing the network structure of development projects. Some of these relate to the position of individuals in a network, and some relate to the structure of the network as a whole. Some of these can be directly observed in network diagrams, others can be identified using social network analysis software described further below.

The position of individuals within networks

Centrality:  There are different ways of describing how central an actor is in a network, and being central may be a good or a bad thing.

A simple measure of centrality is called Degree Centrality This is the number of links an actor has with other actors. In the matrix above, the VDC has the highest Degree Centrality (4) and  the Donor has the lowest (1).

There are two kinds of degree centrality: In-Degree and Out-Degree. These are shown in the summary row (# of links in to an actor) and in the summary column (# of links out from an actor), respectively. These measures are easy to calculate, and can be shown in summary rows under a matrix (using the Count function). They can be very important where the network represents (expected or actual)  influence relationships. An actor that influences many others is likely to be seen as powerful. On the other hand, an actor that is influenced by many other actors, is likely to be seen as much less so.  Most actors in networks will have varying combinations of In-Degree and Out-Degree centrality.  The same measure is also relevant when looking at networks of project activities, and their immediate effects on peoples' lives. Project activities that influence many other activities may need careful planning. Changes in peoples lives that are influenced by many project activities will warrant more intensive monitoring.

Related to In-Degree and Out-Degree is the extent to which links between actors are reciprocated. This is important to attend to when information on linkages is collected from the actors themselves. One actor may say they are working with another, but that second actor may not report working with the first. This may be a simple measurement error or it may be symptomatic of the relative status of the two actors, with the lower status actor wanting to report a relationship with a higher status actor, but not vice-versa.  In some circumstances it may be worth analysing the extent to which each actors outgoing links are reciprocated by incoming links

Betweenness Centrality: An actor might not have many connections with others (i.e. low Degree Centrality), but those they have might still be very important. Betweenness Centrality  describes the extent to which an actor is situated between two groups, and is a necessary route between those groups.  Such people can act as mediators between those two groups, or they can become unintentional bottlenecks, or they can be deliberate obstacles to communications between  those two groups. Which of these roles they take on will, in part at least, depend on their relative power and status, compared to the two groups they are linking. It is not easy to show Betweenness Centrality in a summary columnn of a network matrix. However, in relatively small and /or simple networks it is often possible to identify which actors have high Betweenesss Centrality by looking at network diagrams. In the network diagram in section 5 above the Village development Committee has the highest Betweeness Centrality.   For larger and more complex networks, social network analysis software can be used to identify Betweenness Centrality, and many other measures, for all the actors involved.  Actors with high Betweeness Centrality clearly have the potential to have a major influence on what happens in a development project, and therefore need to be identified and monitored.

Closeness Centrality is a measure of  the average distance between an actor and all other actors in the network. These actors are likely to be most "in the know" about what is happening. Finding these actors and making use of them through M&E activities would make sense.

Peripheral actors:  These are the converse of the above. They may have few connections with other actors, not in any key brokerage roles, and at a higher average distance to others. But they are worth knowing about. They may be more independent minded,  because they are not part of a group which has self-reinforcing beliefs. They may have links with other networks, which take up more of their time, but which could provide useful new knowedge and resources to the network being analysed. Or they may be truly marginal, unconnected to other networks, and at risk of neglect by the actors within the project network
Note: It is important to remember a general point: that all network models are incomplete. Many if not all the actors in the network of concern will have connections with others outside the network. And even within this network, there will be multiple other kinds of  relationships between the actors. A network model will always be a purposeful simplification of reality.

Structural Equivalence: Two or more actors who have the same structure of relationships with other actors are described as being "structurally equivalent". For example, two northern donor NGOs may have relationships with a very smilar set of recipient southern NGOs. In this setting two different questions could be asked. Firstly, in what way can each northern NGO be differentiated from the other, when they are funding the same set of southern NGOs? This is all about differentiating roles, and identifying niches. For example, one might focus more on funding support, and the other on technical support. Secondly, how these northern NGOs  work together to maximise the value of their support to their common southern NGOs. This is all about coordination and harmonisation.

A summary, using Krackhardt's Kite network example
Node 7 has the highest Degree Centrality
Node 8 has the highest Betweenness Centrality
Nodes 4 and 5 have the highest Closeness Centrality
Node 10 is the most peripheral, having the least connections of all
Nodes 4 and 5 are "structurally equivalent"

The structure of networks

Network Density: This is a measure of  how inter-connected a network is. A network where all the actors are connected to all the other actors is said to have a density of 1.0 This is the maximum possible. In the example matrix in section 5 above, that network has a density of  0.55 (that is, there are 20 of the 36 possible links).  Such calculations can be easily built into spreadsheet versions of network matrices.

As with other measures above, network density may be good or bad. A high density network will be less vulnerable to the breakdown of any of the links between the actors, but this will be at a cost. All the actors will be having to manage multiple relationships, and they may not do this as well as they might if managing a smaller number. Within highly hierarchical organisations the density of formal linkages will be quite low, but in organisations using ad hoc teams or a form of matrix management, density with be relatively high.

When developing network models of development projects it is important not to design networks that are too dense. If everything seems to be connected to everything else, then it is hard to see what are the important linkages that are central to the project design. There are two ways around this problem. One is to omit the least important linkages, as has been done in the matrix above. The other is to put a value on the relative importance of each linkage, so it is possible to selectively focus in on the most important linkages. Doing both is even better.

Clusters: Many networks will display some form of clustering, a cluster being a groups of actors with many inter-connections between each other, but few with others. It is sometimes possible to see clusters in network matrices, if the contents have been sorted by row and column beforehand. And by visual inspection of network diagrams. But with larger / more complex networks it is helpful to use statistical functions in social network analysis software, to identify clusters at different scales of connectedness, and the overall degree of clustering.

The identification of clusters in development projects is important. This aspect of a network structure is likely influence the flow of information of concern to a project. Information will normally flow  better within clusters than between clusters. And conversely, the relative availability of information of concern to a project, from what seem to be different groups actors, may tell us about the overall structure of their relationships.

Important Caveat: The significance of all the measures of network position and structure introduced above has to be interpreted in the light of the intentions of the actors involved. Either of the person who has developed a network model as a plan. Or, the actors in the network, if the network has been developed as a description of current relationships. None of the network attributes introduced above are by definition good, or bad.  

11. Thinking about types of network processes  (return to Contents)

Steve Borgatti (in a technical paper) has made some useful distinctions about types of processes that can taken place within networks, according to:
  • How things move within any relationship: 
    • By replication (like information),  or transfer (like physical goods)
      • And if by replication: serially (in one-to-one discussions) or in parallel (like email, or public announcements)
  • How things move through networks of relationships
    • One way only, never retracing its steps. For example, used goods, gossip and diseases (where immunity can be aquired)
    • Both ways, like money. 
These are useful distinctions, because they have consequences for how we interpret the significance of a particular network structure.  Things that can spread by parrallel replication that can retrace their steps are likely to be least constrained by network structure.  And vice versa, things that have to be moved, and which cant retrace their steps, will be most constrained by network structure. Humanitarian emergencies are more likely to involve movement of physical goods than development programs, and in this respect more likely to to be constrained by network structure. But the introduction of cash grants to victims of disasters can be one way around those constraints.

In development projects it may be more useful to think about different kinds of information flow, and how they are affected by network structures

Private information: being information that can only flow from person to person (serially). The content of "private" information that a person can report be more symptomatic of  their position in a network than public information, which can spread in by more mass means.

Public information is less constrained. Allowing information to be put in the public domain is one way of empowering people, because they will no longer be as dependent on network structures for access to that information

Attitudes: Spread not through one-off contacts, but through repeated contacts. As such they may be symptomatic of the duration or strength of relationships. Or the absence of contacts, if we know they are ill-founded.

News: Like other types of information, news can be replicated and spread in parallel. But it may move through some parts of a network faster than other parts. Where news first becomes available may be symptomatic of the structure of the sorrounding network.

Money: Is partly like information and partly like a good. It can be replicated (digitally) but under strict constraints (conservation of quantity). It can be split (to a finite degree) and re-combined as it is transferred through networks of actors. Budgets are means to capture what is received from whom and what is passed on to whom. But the idea of "fungibility" captures the idea that we cannot trace what happens to individual units of money, when they are received by one actor from different sources, and then sent off to different receipients. Budget transparency is all about making the pathways that money follows more visible. That requires readable budgets, but also budgets that have visible onward connections, to people (or organisations), their activities and objectives.

"....when one takes into account timing of the flow of things through a network, ones perspective on key structural features of the network fundamentally changes. For example, the most central node in a network, e.g., as measured by betweenness, can look peripheral if one takes into account the timing of flow. If the most central node gets information slowly, information will flow around that node. A dynamic picture of flow in a network thus can produce a fundamentally different understanding of the structure of the network than a static picture. (I should note that a dynamic picture does not mean necessarily that the network is changing—just that there may be a natural sequence in communication, which may be a long standing structural feature of the network. That is, there is a difference between saying that networks are dynamic and that networks evolve over time.)"

David Lazer, on the Complexity and Social Networks Blog, 1st May 2006
, refering to a point made  by Jim Moody of Duke University.

12. Analysed examples of developed network models (return to Contents)

Most of the examples here are not network models of project intentions that have been deliberately developed by project managers or designers. Rather they are network representations of aspects of development projects, developed opportunistically, when useful data became available. As such it is useful to ask two types of questions about these models:
(a) To what extent are events actually happening as described or implied by these network models?
(b) To what extent to these models represent what was expected to happen?

Please note that the example page links are still being developed. Links in bold with * are now active. The list of examples is not in an especially meaningful order

Networks within organisations

*Strategic objectives and organisational structure. Three examples are shown of administrative units within organisations and how they are linked to each other by their shared strategic objectives (or not). Tow questions: 1. Does this organisational structure reflect how strategic objectives are meant to relate to each other? 2.  Does information flow between these linked sections because they  are talking about  their shared strategic objectives?

Budgets and their relationships to project activities
. The structure of budget categories and sub-categories usually has its own history, independent and predating the planning of activities in a new project. There are also often real constraints on what changes can be made to budget categories to make them have a better "fit" with categories used to describe proposed activities in a new project. The alternative is to construct a budget lines x project activities (or outputs) matrix. That is, a network model of how these two sets of entities are expected to relate. One benefit is that the total cost of each output becomes visible, through the use of summary rows (as described above).

Communications strategies and actual practice

The interaction of communications  products with audiences, and audiences with each other An organisation  may produce multiple types of communications products. These will be expected to be used by  by different audiences, in different combinations. These audiences may be expected to subsequently interact with each other, in varying ways. These relationships can be shown in a series of linked matrices (products x audiences, audiences x audiences)

The same organisation may also have plans for how to engage their audiences with some communications products initially, then with other products later on.  A network diagram can developed to show how users will be expected to move from one product to another. Actual use of those products can then be  monitored to see what is actually happen, and how it fits with initial plans.

*Networks of participants and events. An NGO organises a series of workshops, over a period of years. Quite a few people attend a number of these workshops. They are potentially connected by their co-participation. Overlaps in groups of participants can be intentional as well as fortuitious. Creating and reinforcing relationships may be an important outcome of the workshop series, as well as the more fleeting exchange of information during a particular workshop.  New section (27/04/06): A research funding mechanism in Vietnam has organised a series of workshops to publicise its partners' research findings. These events are linked by co-participants. A series of useful questions can be asked about the structure of the network that results.

Funding relationships

*A research funder and its network of grantees. (PETRRA in Bangladesh). This is a network of contractual relationships between five different types of organisations within Bangladesh and beyond.  Interesting questions can be asked about the project's original intentions and how its longer term impact could be assessed. It is also possible to develop two levels of models, a simple one showing relationships between types of organisations, and a more detailed model showing relationships between specific organisations that belong to these types.

Bangladesh NGOs and their multiple northern donors. A survey of Bangladeshi NGOs in 1992 showed that they were connected to each other via an overlapping set of funding relationships with northern donors. This was not a planned development, though there may have been  local coordination activity amongst some donors. It is clear that some donors have very similar sets of linkages with NGOs, which raises two questions. (a) What sort of differences exist between them, that could justify their separate existence, but shared relationships with NGOs? (i.e. what is their niche?), (b) How can they cooperate, to maximise the value of their support, and minimise the costs to NGOs of having to deal with two separate donors?

Relationships funded NGOs have with others. G-Rap has provided core funding for 12 Ghanaian NGOs, all of whom are engaged in some form of research and advocacy activities. They have some working relationships with each other, and with others. Some these other relationships they share, and some are unique to each NGO. These relationships include other NGOs, community based organisations, central and local government bodies, donors, and others. This complex network can be analysed using the idea of social capital as involving bonding and bridging capital. The  former is all abou the strength of links between the funded RAOs. The second is all about the unique connections each NGO has with other parties, which might be valued by the other NGOs.
*Networks of organisations and projects. Research projects can be connected by the overlapping participation of different institutions. These provide potential channels whereby ideas and practices from one organisation may influence another. As a network they provide a number of potential impact pathways that might be realised in the relatively short term, versus the impact on a wider range of more distantly connected organisations. Which of these pathways is most desirable, and which is most likely to be realised?

Relationships arising via shared objectives

*INGO networks in Vietnam. International NGOs can be potentially linked together by working in the same locations, or in the same sector. The VUFO- NGO Resource Centre in Vietnam produces an Annual INGO Directory. The Directory includes two extensive cross-tabulations, showing which INGOs are working in which provinces and in which sectors. These can be used to generate two-mode networks.

 Global networks of projects linked to global program objectives : The CIAT Water and Food Challenge Program has 33+ projects, in 15 river basins, in x countries, on three continents, with more than x participating  institutions. Where is the structure and coherence? Is the structure we can see what was intended? And what are the expected consequences of this structure? Where should information and influence be flowing? 

*Networks of donors and policy objectives: A network perspective on donor harmonisation in Ghana, in the form of  a matrix showing what donors will be providing funding support to what Growth and Poverty Reduction Strategy policy objectives. Qustions can be asked  about who needs to be talking to whom, about what? And if you are an NGO wanting to do advocacy work on poverty related policies, what are the implications for who you work with?

Networks over time

Networks of influence over time: Fisheries research in Australia. Research projects can influence multiple other research projects, in the short term and in the longer term. The result: a complex genealogy, a network spreading through time rather than a simple branching family tree. Who seems to have been the most influential over the longer term may be relevant to analysis of the long term impact of the research funding mechanism.

Document networks

Policy document networks.  Development policy documents often include lists of indicators, and overlaps between these indicators can sometimes be easily identified. The 2003 M&E Plan for the Ghanaian Poverty Reduction Strategy included a table showing which indicators were being used on three related policy documents (HIPC, GPRS, MDG). Treated as a two-mode matrix the results can be converted to a network diagram which readily shows in some detail how the policy documents are related.

*Documents can be linked by common themes, and sometimes researchers are paid to identify these themes and their prevalence.  This was done by Brocklesby and Crawford's  a report on "Why Local Drivers of Change Matter: Reviewing "Local" in DFID Drivers of Change Studies" (2005). A network analysis of the results can highlight connections that were not given attention in the text description. 

Causal networks  Many large projects have multiple objectives. Many of these objectives can be expected to feed into each other, with the achievements of one contributing to the achievement of another, or multiple other objectives.  This means there will not be a simple one-to-one correspondence between the amount invested in an objective and the degree of its achievement. Some objectives may require additional investment, others less than expected. The network of causal linkages will also have implications for where M&E efforts should be concentrated.

13. Questioning network models about the wider context  (return to Contents)

All network models are incomplete, both by necessity and by intention. They show particular relationships between a set of actors, but not other kinds of relationships that are seen as less important. They show relationships between a specific set of actors of concern, but not their relationships with the many others who sorround them. They show relationships within a given period of time, but not the relationships that preceded them, or which may follow later.

Where network models are being used to represent the design and / or achievements of a project it can be of value to ask questions about this wider context.. For example:

Re wider networks

What are the most important wider connections that each actor has with other actors outside the network?

Which of these may provide the network with opportunities?

And which of these may function as significant constraints?

PS: The focus on each actor's  unique connections with others outside the immediate network is a way of describing their "bridging" social capital
Re pre-existing and subsequent relationships

Which of these relationships pre-existed the project, and which do not?

Which of the relationships in the network model are expected to continue to exist after the project finishes?

14. Collecting network data (return to Contents)

Potential sources
  • Applications
    • By organisations seeking funding (application forms) 
    • By individuals seeking employment (CV/ Resumes) 
  • Records of participation: people and organisations who were participants in various project "outputs": workshops,  training events, mailing lists, websites, etc

  • Reports:
    • Lists of cross-references to other documents and other authors
    • Tables of objectives, sub-objectives and indicators
    • Tables of data in  annexes (where more data, in more detail, can be allowed)
  • Online sources
    • Websites: Logs of visitor activities, help by all hosts of websites.
    • Google searches: Websites, and documents, deliberately linked, and linked by shared use of same key words
  • Special purpose surveys 
    • Face to face.
    • In workshops
    • Online
  • Content analysis of documents
    • Coding of content into different categories of actors, events, etc, then treating their co-occurence in a  document, or section of a document as a network linkage... 
Survey methods [to be developed]

15. Using Social Network Analysis software  (return to Contents)

There are three specialist software packages that can be used, as listed below. But note that for small networks, all that is needed is an Excel spreadsheet (for matrices) and the Draw function in Excel (for network diagrams)
  1. UCINET, at This includes the NetDraw program for producing network diagrams, using data input into UCINET.Good feature: You can easily cut and past matrices of data into a spreadsheet within UCINET. And there is an online Users Group that you can join, and get help from
  2. Visualyzer, at Good feature: You can draw network diagrams on the screen, which are then converted into matrix form.
  3. Visone, at  Fairly easy to use. See also the users group mailing list 
Caveat: The main problem with all three is that  there are lots of "Bells and Whistles" in these programs that you need to ignore at the beginning, while you are still struggling to learn how to import and export data, and to create and manipulate network diagrams. Be patient and persistent.

16. People's participation? (return to Contents)


Because social network data is increasingly amenable to analysis using social network analysis software there is a risk that analysis of networks will be reduced to a 21st century version of "number crunching". If the data is there, and the software is there, why not crunch the numbers?

The answer is that this will only take us so far. Analysis needs to be informed by a theory of what is supposed to be happening. While our own theories may be interesting and relevant, the theories of the actual participants in these networks will probably be even more so.  This is especially the case in development projects, where there will usually be some intention to seek change in the behavior of at least some of the actors in the network.

Social network analysis has come from a fairly academic background, so there is still plenty of room for innovation relating to practical applications. Especially in participatory approaches to the description, analysis and planning of networks. If you have any ideas, have already tested some methods, or know of others who have, please use the Participatory Network Models wiki page to share these with others.


People can be involved in the development of network models in different ways:
  • Planning the changes needed, in the form of a network model
  • Describing what the current situation is (before, during or after plans are implemented), in the form of a network description
  • Making predictions of the expected results of a survey of network structure, shortly before they are analysed and presented
People can participate via the use of various tools
  • Individually, by being survey respondents and providing data on their place in networks. This is the most limited form of participation but the most widely used. 
    • But prediction questions could be built in to such surveys, about the likely response of others. For example: After answering a particular question about teir relationships with others, then ask "Who is likely to name you in response to this question? I think this takes us in the direction of cognitive social networks, tracking what people know about others' views of networks (Readers: Correct me if I am wrong)
  • Collectively, by discussing and analysing data as they provide it, after it has been analysed and fed back.
    • Both matrices and network diagrams can be used in workshop settings and can be the focus of intense and lively debate
      • I have outlined my own experience with both of these in the wiki mentioned above
    • In outdoor situations Venn diagrams can be drawn on the ground, in a small group setting, to capture relationships between groups of people.
  • Dynamically, by involving people in what are called "social simulations" of networks, whereby they take the part of a given actor (person, organisation or group) in a network, and make a relationship with another in that network, to pursue an objective. This process is then repeated over a number of iterations, or generations, allowing people's subsequent choices to be informed by their knowledge of the choices made by other actors in the previous iteration. 
    • A "Getting to know people" exercise example: Round 1: individual participants pick another participant they dont know, and find out about them. They also find out who else they know in the room, and what they know about them (perhaps some pre-specified information). Round 2: They pick another actor,  and find about them. They also find out who else they know in the room, and what they know about them. Round 3. As above, and so on with subsequent rounds until the first person claims they have all the (relevant information) about all participants. Or until a defined number of rounds are completed, at which stage you map who knows whom, and who they do not know.  As well as identifying who is most known, this process may also help identify those who are unknown, and therefore in most need to be known!

17. Available data sets 

This section will contain a list of links to files of data on a range of networks. In most cases they will relate to the network model examples accessible via section 12 above

The purpose is to involve more people in the analysis of this data. A: To give people data to use for practice purposes, to develop their skills in network analysis, and B: To bring more brains to bear on the same data, and perhaps generate some more creative and or complete analyses, building on whatever has been done so far. So, if you down load this data, and use it for network analysis, please do so with the intent of sharing the results of your analyses with Rick Davies (the author of this web page) and others who are on the Network Evaluation email list

The data sets that will be listed here are (provisonally):
  • 30 NGOs in Bangladesh interviewed in 1992 x the 50+ donors they reported as funding them, with some categorisation of  their status as donors (primary versus other, years as a donor). Click here for the example analysis of this data on this website.
  • A 13 year time series on international NGOs in Bangladesh x 30+ sectors they were working in each year, and international NGOs in Bangladesh x 60+ privinces they were working in each year.  Click here for the example analysis of this data on this website..

18. Structure and Agency in network models   (return to Contents)

With their emphasis on structures of relationships, network models may seem to biased towards deterministic view of human activities, and may appear to neglect the significance of human agency, the ability of individuals to make a differences. This should not be the case, if network models are used wisely.

Network structures can not only influence human actions (both by constraining and enabling), but they can also be the outcome of human actions. In other words, we can see network structures as independent variables, and as dependent variables. Both perspectives are important. At the planning stage of a development project we may need to focus more on how existing network structures can enable and constrain proposed developments. At the end of a project, an evaluation may need to attend to changes that have arisen in the structure of relationships between project stakeholders, and their causes. And looking further forward, beyond the project lifespan, the evaluators may also need to ask how sustainable various relationships will be, and what factors will affect their sustainability.

Network structures only have significance in as much as they are seen and interpreted by the actors in those networks. A link between two actors will be of little significance, if for some reason those two  actors (and others) have forgotten about the link. Differences in their perception of the value of that link will have consequences, regardless of how an independent observer might  assess the link. As already argued above, we not only need data on what relationships seem to exist, but also the actors's views of those relationships: their relative significance, and how they relate to actors' prior plans and intentions.  Narrative methods, such as Most Significant Change (MSC) monitoring may be a useful means of collecting and analysing such information.

For more on managing  the issue of structure and agency in research on networks, see A network approach for researching partnerships in health. by Jenny M Lewis 

19. Complexity theory and network models (return to Contents

One characteristic of complexity is unpredictability. One source of unpreditability is feedback. Beautifull and unexpected fractals can be produced when the results of a calculation are fed back as an input into the next iteration of that calculation. In Conway's artificial life program (Life), a diversity of life forms emerges as a population of "live" cells follow a simple set of rules about how to respond to their immediate environment. And where in the process their environment is changed by the results of following those rules.

Complex systems are almost by definition dynamic, concerned with change over time. In contrast, network models have frequently been criticised as being overly static, focused on the state of a social or other system at a point in time.  Network models can however generate  dynamic and complex processes, if the "outputs" generated by a network model are fed back to inform the "inputs" of the next "generation" of this model. The inputs in a network model are the attributes of the actors and the structure of the connections they have with each other (both the existence of specific connections, and their strengths, as given in the cell values). The outputs  can be seen in the summary rows and summary columns shown in the example in section 9 above.  These can be used to re-adjust the input values in the next version of the network model.

This updating process can happen in  in two ways: (a) through a social process, such as a workshop. (b) mathematically, by creating a formula that uses the output values to recalculate the input values (especially the actor attributes).

In a workshop setting participants can complete a network matrix, by identifying and agreeing on appropriate cell values. These can be inserted in an matrix presented in an Excel file, projected onto a screen visible to all. The summary row can be constructed to automatically generate summary values, as soon as the cell values are entered on the screen. The facilitator can then ask participants to reflect on the summary values, and their acceptability, relative to their views up to then. If the values are not acceptable, participants then have three choices, which would cause changes in these values:
(a) Change the characteristics of the actors in the left hand column
(b) Change the characteristics  of the actors in the top row
(c) Change the linkages between them, by changing the cell values within the matrix

Complex systems can exhibit different kinds of behavior, according to  the structure of the connections between the actors involved. They can (a) be chaotic (i.e. summary values will vary unpredictably over a  long series of iterations), (b) tend towards a stable state (i.e. summary values will stablise, and not change any further, despite further iterations), or (c) exhibit complex behavior (where particular summary values appear, disappear, and re-appear in cycles).

Most organisations using network models will be interested in models that will generate stable sets of values, and seek to avoid those which generate chaotic or complex behavior. In workshop settings there is usually limited time available to develop network models, and then run them through one or more iterations. So participants are likely to have an interest in coming to agreement about the acceptability of summary values relatively quickly. Where outputs are linked back to inputs mathematically, through the use of formula in a spreadsheet, there is no such "inertia", and reaching a stable state may be more difficult. In these circumstances simplifying the structure of the linkages may help. Evidence seems to suggests that networks with high degrees of connectivity are more likely to show chaotic behavior [Readers: Please correct  this statement if it is wrong]

Example  of stabilisation of attributes values for  actors, when matrix summary values are fed back as inputs

Postscript: What was intersting about this simulation was that the initial values that were generated through the first iteration of the mdeol were quite different to what the final values settled down to, by the ninth iteration.  See actors 2, 3, 4 above, especially.  This suggests that running iterated versions of causal network models may be worthwhile. The Excel spreadsheet that was used to generate these values can be found here.

20. Useful online resources  (return to Contents

See Robert A. Hanneman and Mark Riddle's excellent online "Introduction to social network methods" at

And Valdis Krebs' more application oriented web pages at See the 24 examples. Every one of them is interesting.

Boundary partners: "Those individuals, groups, and organisations with whom the programme interacts directly and with whom the programme anticipates opportunities for influence" A term borrowed from Outcome Mapping

Visit counterCounter started: 03/01/06