Insight network

From Opasnet
Jump to: navigation, search

Insight networks are graphical representations of a particular situation, where the objects described are causally related to each other. In addition, the diagrams contain non-causal elements such as value judgements or inferences based on data. Insight networks utilise the ideas of directed acyclic graphs, but they have additional features.


What notation is simple and flexible enough so that it can be used to represent all major issues related to a policy situation? It must be usable in both graphical and data formats.


+ Show code

Legend for extended causal diagrams.svg



Insight networks have been described in a scientific article manuscript From open assessment to shared understanding: practical experiences#Insight networks. Objects and their relations used in open policy practice are described on page Open policy ontology.

There is a need for methods facilitating the flow of information and understanding between science and policy. The principle is to describe a risk situation in a formal manner. Insight networks contain items along a causal pathway (or network) from e.g. abatement strategies to emissions to dispersion to exposure to effects. They have been designed to describe also non-causal connections such as non-causal reasoning, values, preferences, and arguments.

These diagrams use graph theory with vertices (or nodes) and arcs (or arrows). They are used to describe and define all the pieces needed for a description of the situation under scrutiny. Diagrams may be produced with any graphics software, providing that calculation functions are not required. If calculations are needed, we recommend the use of R software and OpasnetUtils package.

This is the process how data flows into insight diagrams:

  • List of data tables of different insight diagrams is found from It has the following columns:
    • Ilmio: Name of the phenomenon. This will become the name of a csv data file.
    • Id: Identifier of the phenomenon. This will be used in Oldid of the items and relations.
    • Tyyppi: Type of the table. In practice, it defines the columns that the data table has. Different types are listed on #Types of insight network tables.
    • URL: Location of the data table. If the URL contains "", it is assumed to be a google sheet. If the type (Tyyppi) is "keskustelu", it is assumed to be an Opasnet page with discussions. Otherwise, it is assumed to be a table on a web page that can be scraped with read_html() function.
    • Taulu: If the data is a table on a web page, it is the number of the table on that page. If the data is a discussion, it is the number of discussion; missing value means that all discussions on that page are read.
    • Alkurivi: In case of google sheets, it is the first row with actual data.
    • Kuvaus: Description of the table, with possible links to relevant description page.

All data tables and discussions are listed, formatted and saved as csv files in a zip file called op_fi:File:Näkemysverkkojen From there, the data can be accessed from within Opasnet Rtools. (The code scraping web pages does not work in Opasnet, although it is stored there.) Little formatting is done here, mainly the column titles are standardised. But the number and type of columns is not changed.

In the next phase, each csv file is opened, interpreted, and defined as items and relations. This is done in code Op_fi5810/graphs on page op_fi:Ympäristöterveysindikaattori. All these are saved as a DiagrammeR graph, and each topic may be separately selected as a subgraph.


Graphical properties of objects and relations

Graphical properties of objects and relations(-)
1defaultdefaultnode.shapecircleDefault values unless something else is specified
14defaultdefaultedge.fontsize10Not currently used
18defaultdefaultedge.arrowsize1Not currently used
19typeunknownnode.fillcoloryellowThis formatting is used if there are undefined objects
21typesubstancenode.shapecircleSubstantive type object
22typesubstancenode.fillcolorskyblue2Substantive type object
23typeknowledge crystalnode.colorgoldKnowledge crystal type object (including ovariables and key ovariables)
24typeoptionnode.colorpalevioletredOption for a decision
25typeoptionnode.fillcolorwhiteOption for a decision
26typeindexnode.shapepolygonIndex or other classifying determinant
31typegraphnode.shapepolygonIndex or other classifying determinant
37typestakeholdernode.shapehexagonStakeholder type object
38typestakeholdernode.fillcolorkhaki1Stakeholder type object
39typestakeholdernode.width0.8Stakeholder type object
40typemethodnode.shapepolygonMethod type object
41typemethodnode.sides6Method type object
42typemethodnode.fillcolorpurple1Method type object
43typeprocessnode.shapepentagonProcess type object
44typeprocessnode.fillcolorpurple1Process type object
45typeactionnode.fillcolor#009246Process type object, dark green (0,146,70)
46typeactionnode.shaperectangleDecision type object
47typetask 1node.colorbrownIllustration of the responsible organisation of the task
48typetask 2node.coloryellowIllustration of the responsible organisation of the task
49typetask 3node.colorblueIllustration of the responsible organisation of the task
50typetask 4node.colorgreenIllustration of the responsible organisation of the task
51typetask 5node.colorredIllustration of the responsible organisation of the task
52typedecisionnode.fillcolorredDecision type object
53typedatanode.shaperectangleData type object
54typedatanode.fillcolorgoldData type object
55typeobjectivenode.shapediamondObjective type object
56typeobjectivenode.fillcoloryellowObjective type object
57typeobjectivenode.width0.8Objective type object
58typepublicationnode.fillcolorgrayPublication type object
59typestatementnode.shapepolygonArgument type object
60typestatementnode.sides4Argument type object
61typestatementnode.width0.8Argument type object
62typestatementnode.distortion-0.5Argument type object
63typetrue statementnode.fillcolorgoldArgument type object
64typefalse statementnode.fillcolorgrayArgument type object
65typefact opening statementnode.fillcolorlightskyblue1Argument type object. Discussion start
66typevalue opening statementnode.fillcolorpalegreen1Argument type object
67typefact closing statementnode.fillcolorskyblueArgument type object. Discussion end
68typevalue closing statementnode.fillcolorspringgreenArgument type object.
69typefact discussionnode.fillcolorskyblueArgument type object. Not neede?
70typevalue discussionnode.fillcolorspringgreenValue judgement type object. Not needed?
71typerisk factornode.colorpinkAdditional information about object class
72typeindicatornode.colorbrownAdditional information about object class
73typeindicatornode.fillcolorgoldAdditional information about object class
74typeoperational indicatornode.fillcolor#00d7a7Additional information about object class light green (0,215,167)
75typetactical indicatornode.fillcolor#9fc9ebAdditional information about object class light blue (159,201,235)
76typestrategic indicatornode.fillcolor#0072c6Additional information about object class dark blue (0,114,198)
77typestrategic indicatornode.shapediamondAdditional information about object class
78typearviointikriteerinode.colororangeNot quite clear what criteria objects are: indicators or value statements, or something else
79typetasknode.colorgreenAdditional information about object class
80typedatanode.colororangeAdditional information about object class
81typehealth organisationnode.coloryellowAdditional information about object class
82Relationcausal linkedge.colorblackCausal link
83Relationcausal linkedge.stylesolidCausal link
84Relationpositive causal linkedge.fontcolor#009246Causal link, dark green (0,146,70)
85Relationincreasesedge.fontcolor#009246Causal link, dark green (0,146,70)
86Relationnegative causal linkedge.fontcolor#bd2719Causal link, red (189,39,25)
87Relationdecreasesedge.fontcolor#bd2719Causal link, red (189,39,25)
88Relationpart_ofedge.fontcolorgrayPart of (set theory link)
89Relationparticipatory linkedge.colorpurpleParticipatory link
90Relationparticipatory linkedge.styledashedParticipatory link
91Relationoperational linkedge.colorblackOperational link
92Relationoperational linkedge.styledashedOperational link
93Relationevaluative linkedge.colorgreenEvaluative link
94Relationrelevant attackedge.colorredAttacking argument
95Relationrelevant defenseedge.colorgreenDefending argument
96Relationrelevant commentedge.colorblueCommenting argument
97Relationirrelevant argumentedge.colorgrayInvalid argument
98Relationargumentative linkedge.styledottedArgumentative link
99Relationargumentative linkedge.penwidth4Argumentative link
100Relationreferential linkedge.colorredReferential link
101Relationreferential linkedge.styledashedReferential link


Insight network 2.0

An updated version should improve the

  • a) context sensitivity (referring to primarily to objects within own context but secondarily to those from another context),
  • b) making graphs by default from a single context rather than a full list of contexts from a meta table,
  • c) compatibility with cytoscape.js,
  • d) merging ready-made graphs meaningfully,
  • e) have a reasonable intermediate object format that contains all data needed, such as
    • tables for nodes and edges, compatible with Diagrammer, Cytoscape.js, AND Gephi.
    • metadata for display, such as seeds, steps, object types to ignore, whether to show labels etc. Or should these just be implemented on the graph?

What should be done?

  1. Fetch the data table by scrape or other function and with data about URL, table, and initial row.
  2. Use splizzeria and fillprev if needed.
  3. Interpret columns based on a vector of column numbers (with possibly 1+2 notation to paste columns) to create the standard columns. If this is done in an ovariable formula, there is no need for a specific function.
    • Context
    • Item
    • type
    • label
    • rel
    • Object
    • Description
    • Reldescription
    • URL
    • Result (dummy, always 0)
  4. Create missing node rows from objects. Do NOT assume context.
  5. Create URL from permanent resource location trunk and the identifier (where does the identifier come from?)
  6. Item ja label laitetaan pötköön ja haetaan mätsi. Tulos onrow-pötköstä.
  7. Create an ovariable from the table.
  8. Add meta to the ovariable with formatting data.
    • insightGraph:
      • seed
      • removenodes
      • formatting (character vector with possible entries: Hide node labels, Hide edge labels, Show legend nodes, Remove branches only)
      • ignoreobj
      • steps
  1. (NOT NEEDED? Create Oldid if does not exist from context and numbering)
  2. If a relation is presented as item, the formatting is applied to the ring.

Combine graph objects

  • Find items without context. Match them with items with the same Item (label) that do have a type.

Tuplarelaatiot, voidaanko kategorisesti poistaa?

Out <- rep(NA, length(find)) For(x in cond,) For(i in 1:length(find) Tmp<-id[context==contextfind(i))])[Match(find(i), df$cond(x)(df$context==contextfind(i))] pitää etsiä id alkuperäisestä taulukosta heti muuten ei toimi Out<- ifelse(isna(out). Tmp,out) )) Sitten sama ioman contekstirajoitusta.

Insight network 1.0

There are three different identifiers for a subject item.

  • Oldid: a technical identifier typically of format context.number, where number is a sequential number within a context.
  • Item: the actual name of the item, detailed enough to give a good understanding of its meaning.
  • label: a short name shown on insight networks. Does not exmplain everything, just enough to distinguish it from other items.

If Oldid is not given, it is created from the context and a number. If label is not given in data, it is truncated from Item.

Object item has one column Object that may contain any of these. The priority is Item > label > Oldid > Object. The last option means that it is assumed that Object refers to a new item that is not mentioned in the Item column.

An insight network is produced in this order (last object mentioned first).

  1. gr: a diagrammer graph with all data and formatting for an insight network. Produced by makeInsightGraph.
  2. makeInsightGraph

Making insight graphs

+ Show code

+ Show code

+ Show code

+ Show code

Function insightJSON fetches a JSON file of an insight network through a REST API. Works on own computer only.

+ Show code

Format tables

+ Show code

+ Show code

Shiny server

+ Show code

+ Show code

Scrape functions

These functions were be placed in the OpasnetUtils package, which is maintained in Github. To use the code, install a new version of the package by running R code


Codes Op_en3861/scrape.discussion, Op_en3861/scrape.functions, and Op_en3861/scrape.assessment on this page are outdated.

Copy descriptions to ovariables

The function assessmentDescriptions scans through an assessment ovarible that has all relevant assessment objects as parents. Dependencies slot may also have additional information, such as the following.

  • Name: name of parent (obligatory)
  • Ident: Opasnet page identifier and code name where the parent object can be loaded (e.g. Op_en7748/hia). Note: This is typically the code for the whole assessment, not the individual codes for the objects.
  • Token: Token for the model run where the parent object can be loaded (e.g. xxNsLw5hWdM6xyYp)
  • Description: A short description about what the object is. This is typically shown when cursor hovers over the object on an online insight diagram.
  • Page: Opasnet page identifier for the object's knowledge crystal page, which contains the research question, answer, and description of the object, together with discussion, if any. Typically this is empty for ovariables, because this information can be found from ovariable@meta slot and there is no need to duplicate it here.
  • Child: An object to which this object links. This is typically needed for objects such as graphs and data.frames that do not contain this information in their own structure, unlike ovariables. The direction of a relation is away from this object because then this object is the subject in triple sentences and can be given other parameters as well in other columns. A typical sentence is "graph describes ovariable", but for illustrative purposes this is inversed on insight networks so that the arrow points from an ovariable to a graph ("ovariable is described by graph").
  • Other columns are allowed.

+ Show code

Old notation

⇤--#: . Look at the table below together with Open policy ontology and merge. Decide which things should be on this page and which should be on the other. --Jouni (talk) 06:55, 24 April 2018 (UTC) (type: truth; paradigms: science: attack)

Node type Object Colour code in Analytica Comments
General variable.png General variable 8R3B (automatic) This is a deterministic function of the quantities it depends on.
Chance variable.png Chance variable 11L4B (autom) This is a variable which is uncertain and uncontrollable (in a direct sense).
Data-driven variable.png Data-driven variable 3R1B A general variable where the result is mostly driven by data (observations or literature).
Author judgement variable.png Author judgement variable 4R2B A general variable where the result is mainly driven by author judgement (estimates with poor or no data).
Decision variable.png Decision variable 9L3B This is the variable that a decision-maker has the power to control. The decision variable should always be at the top of the chain of causality, even if this is a subchain i.e. it should not have any parent variables. Essentially the decision variable should be regarded as a decision that has to be made; since many factors affect all decisions it is not (in the case of INTARESE) an efficient use of resources to attempt to model what leads a decision-maker to make his decision.
Objective variable.png Indicator 1R3B (autom) This is a variable of special interest. One of the indicators in an assessment may be the quantitative criterion that you are trying to optimize.

A particularly important variable in relation to the interests of the intended users of the assessment output (i.e. it must be a means of effective communication of assessment results).

  • It must be in causal connection to the endpoints of the assessment and thus address causality throughout the full chain.
  • It should reflect the use/purpose of the assessment.
  • It should address and be adapted according to the target audience.
  • It should be the ‘leading component’ in the assessment process.
Value judgement variable.png Value judgement variable 8L4B A preference or value that a person or a group assigns to a particular condition or state of the world.
Index variable.png Index (or dimension) 5R2B (autom) This identifies the dimensions of the variable to which it is linked. Note that these dimensions do not have to be numeric, but can also be classes etc.
Risk assessment node.png Risk assessment 8R3B (autom)
Scope node.png Scope 6R1B The scope of the object
Conclusion node.png Conclusion 6L3B A conclusion of the risk assessment (Result/Conclusion attribute).
Module node.png Module 6R3B (autom) A group of variables that are put together for illustrative or other practical reasons.
Proxy variable.png Data 2L3B (autom) Contents of the Definition/Data attribute of a variable. If the Result attribute of a variable is used as Data for another variable, the first variable is called a proxy, and this node is used in the diagram. If an arrow or line is drawn between these objects, it must be noticed that this is NOT a causal link but an inference link. The direction of the arrow would be from the proxy to the variable.
Argument node.png Argument 8R2B A piece of argumentation related to an object (variable, risk assessment, or class)
Formula node.png Formula 9L3B Contents of the Definition/Formula attribute of a variable.
Class node.png Class 1L2B A class object (a set of objects that share a particular property).
Function node.png Function 4R2B (autom) A special kind of class. The particular property that is shared contains a full description of the Scope and the Definition attributes with given parameters.
Causal arrow.png Causal arrow This states a causal relationship (or influence) of one variable onto another. Note that causal arrows can only exist between two arrows; any arrows to or from non-causal objects are non-causal inference arrows.
Non-causal arrow.png Non-causal arrow This states an inference relationship between two objects. This means that the object where the arrow starts from is in the Data attribute of the other object. It is thus used to infer something about the value of the result of the latter object. Either object can be a variable or a non-variable. Note that Analytica is only able to show one kind of arrows, so in some cases the nature of the arrow (causal or inference) must be concluded from the context.

Previous notations

Previous notation for insight networks. This version was optimised for Analytica use.

Insight networks have previously been called pyrkilo diagrams, extended causal diagrams, and factor-effect-value networks. These names are no longer in active use. An archived version of the notation can be found from an earlier version of this page.

See also

  • Arhived version 15.1.2019 with several functionalities that are now depreciated and removed.
    • T2b table Table types for different kinds of input tables.
    • Code for function grspec. This is no longer needed as a generic formatted data.frame is used for formatting of all resources.
    • Code for makeInsightGraph. This is replaced by makeGraph that has a better work flow.
    • Code for makeInsightTables. Insighttables are no longer produced as they are replaced by context-specific ovariables that are on their respective knowledge crystal pages.
    • Code for ovariable insightNetwork, which is an ovariable collecting all objects needed. Because of major updates, this is no longer useful.
    • Code server: function chooseGr was updated and moved to an own code.