Methods · Structural analysis
ISM (Interpretive Structural Modelling)
ISM is a structural analysis method that collects the "which factor affects which" relationship among a group of factors from a panel of experts and separates them into levels, drawing a hierarchy from root cause to outcome.
Base method's data type: Classical
What Is the Method?
ISM is a structural analysis method for when you hold a large number of interacting factors (barriers, criteria, causes) and want to see which triggers which. Unlike TOPSIS or AHP, it does not answer "which is best" or "which is most important"; it answers "which factor sits below which, which sits above whom." Its output is not a ranking or a score but a hierarchy diagram that separates factors into levels: the bottom level holds the root causes that trigger everything else, and the top level holds the visible effects that result from those causes. Warfield proposed it in 1973, and it has since found wide application, from systems engineering to management science, from supply-chain barrier analysis to sustainability studies.
The Philosophy Behind It
The idea behind ISM is to resolve a complex problem by reducing it to the question of which factor gives rise to which. For every pair of factors, a panel of experts is asked whether one affects the other. Working from these pairwise answers, the method also accounts for indirect effects (if A affects B and B affects C, then A also affects C) and separates the factors into levels. The factors at the top level affect no factor beyond themselves and are only affected; those at the bottom are roots affected by no factor, only affecting others.
This idea carries a philosophical consequence: consistency. ISM does not search for a "best" option; it makes a system's internal logic visible. It breaks the scattered causal knowledge in experts' minds into individual pairwise questions, then combines those answers into a coherent whole. The method does not produce a right or a wrong hierarchy; it produces the logical consequence of the pairwise answers the experts gave. Change the input, the pairwise relationship answers, and the output, the hierarchy, changes with it.
How It Works
The method proceeds through five steps.
First, building the pairwise relationship matrix. For every pair of factors, experts are asked whether one directly affects the other, and the answers are combined into a table consisting only of "yes" or "no" (1 or 0) values.
Second, accounting for indirect effects. Once direct relationships are entered, cases where A affects B and B affects C are added as an indirect effect of A on C. This is repeated until no new indirect link remains, yielding a stable table known as the reachability matrix.
Third, finding reachability and antecedent sets. For every factor, the set of factors it affects, directly or indirectly, and the set of factors that affect it are each extracted separately.
Fourth, separating into levels. When a factor's affected set intersects with its affecting set and this intersection equals the whole affected set, that factor is placed at the top level. These factors are removed from the table and the process is repeated on the remainder; each round settles one further level.
Fifth, drawing the hierarchy diagram. Once the levels are set, which factor directly affects which is shown on a diagram; factors at the same level are not ranked against one another, only the direction between different levels is shown.
The formulas behind each step, the intermediate tables and the citation formats are given on the DecisionMind method page; this card carries no formulas.
How to Read the Output
The output is not a score but a hierarchy that separates factors into levels, together with the arrow directions between those levels. Factors at the top level are the outcomes most easily observed in the system yet the least useful to intervene on directly; factors at the bottom level are harder to see but produce a cascading effect once worked on. Two factors placed at the same level are not "equally important"; it means only that they do not affect one another, directly or indirectly; they are independent.
Thus instead of writing:
"ISM found the most important factor"
the report should read:
"With these expert opinions, these factors are root causes and these are their consequences; intervention should begin with the root causes"
Data Type and Inputs
ISM works with crisp data, but this data is not a numerical measurement; it is a binary (yes/no) expert judgement. DecisionMind carries no extension of this method; the base method stands on its own. You need a complete list of the factors under study and, for every pair of factors, an expert judgement answering whether one affects the other. As the number of factors grows, the number of pairs to compare grows rapidly; three to twelve factors remain manageable in practice, and going beyond that can lead to expert fatigue and inconsistent answers. ISM does not produce or require weights; what it produces is not a rank or a score but a level structure.
When to Use It, When Not To
If you hold a large number of interacting factors and your goal is not to rank or score them but to see which triggers which, ISM is appropriate. It is generally used to uncover the root causes of a problem, be it a barrier, a risk, or a cause of failure. If your goal is instead to rank alternatives or allocate a resource, ISM is the wrong tool; ranking methods such as TOPSIS and VIKOR, or portfolio methods such as HF-TRADEOFF-PORT, should be considered instead. Where the factors do not genuinely interact, where every factor is independent, ISM produces an empty table and becomes redundant.
Interaction among factors, a search for hierarchy or root cause → ISM
Ranking alternatives, finding the best one → ranking methods such as TOPSIS, VIKOR
Allocating a resource among alternatives → portfolio methods such as HF-TRADEOFF-PORT
Weighting criteria → subjective weighting methods such as AHP, ANP
Strengths
ISM's greatest strength is reducing a complex web of relationships to individually simple pairwise questions and presenting the result as a visual hierarchy. It converts the scattered causal intuition inside an expert panel's collective mind into a systematic table that can be discussed and reviewed. It accounts for indirect effects, chains of causes, not by hand but in a mathematically consistent way. Because it shows which factor, if intervened upon, would produce the widest cascading effect, it points to where to focus first under resource constraints.
Weaknesses
ISM's core limitation is that it treats relationships only as binary, present or absent; it does not distinguish how strong a relationship is (Sushil, 2012). Second, its input rests entirely on expert judgement; a different panel of experts may arrive at a different binary table, and hence a different hierarchy. Third, it draws no priority distinction among factors at the same level; which of these to focus on first falls outside the method. Fourth, as the number of factors grows the number of pairs to compare grows rapidly, and it becomes harder for experts to give consistent answers (Watson, 1978; Attri, 2017).
Common Mistakes
The most common mistake is trying to rank factors at the same level against one another; ISM does not provide this information, since factors within a level are only independent of each other. A second mistake is putting the pairwise relationship questions to a single expert and presenting the result as the whole panel's view; different experts' answers must be reconciled, not merely combined. A third mistake is choosing the top-level factors, the outcomes, as the point of intervention; a cascading effect requires focusing on the root causes at the bottom level. A fourth mistake is marking indirect effects as though they were direct when building the pairwise relationship matrix; indirect effects are computed by the method itself and should not be added to the input by hand. The governing principle is this:
The hierarchy ISM produces is the logical consequence of the binary answers the expert panel gave; if the answers change, the hierarchy changes too, and this does not invalidate the result: it shows that the result depends on its input.
Cases
Each case opens with a pairwise relationship table, describes in words what the method does to it, and shows how the hierarchy should be read. In this card, the example in manifest block J is a placeholder numerical table used for method auditing and does not show ISM's real structure. All three cases are therefore illustrative constructions; the figures were computed for the writing of this card in Python, following ISM's own steps, the reachability matrix and level separation.
1. Agriculture: Factors holding back a cooperative's move to a digital sales platform
An agricultural cooperative's board wants to understand why so few members register their produce on a digital sales platform. Four factors have been identified: weak management support for the platform, insufficient training and promotional activity for members, members' mistrust of the platform, and the low product-registration rate. The board has answered, for every pair of factors, whether one directly affects the other: weak management support leads both to insufficient training activity and to member mistrust, and both insufficient training and mistrust in turn produce the low registration rate.
| Factor | Abbreviation | Factor(s) directly affected |
|---|---|---|
| Weak management support | E1 | E2, E3 |
| Insufficient training and promotion | E2 | E4 |
| Members' mistrust of the platform | E3 | E4 |
| Low product-registration rate | E4 | - |
The method first enters these direct effects into a table, then adds the indirect effects: E1 also indirectly affects E4, both via E2 and via E3. Comparing each factor's affected and affecting sets settles the levels.
| Level | Factor(s) | Position |
|---|---|---|
| 1 (top) | E4 | Outcome |
| 2 (middle) | E2, E3 | Intermediate layer |
| 3 (bottom) | E1 | Root cause |
The result reads as follows. The low product-registration rate (E4) affects no factor, only being affected; it therefore sits at the top of the hierarchy, in the position of an observed outcome. Insufficient training (E2) and mistrust (E3) sit at the same level; both are affected by weak management support and both affect the low registration rate, but they do not affect each other. Weak management support (E1) is affected by no factor, only affecting others, and therefore sits at the bottom, in the position of a root cause.
The board hesitates here: because E2 and E3 appear at the same level, ISM does not answer which of them should be addressed first; that is left to the board's own priorities or to a further cost-benefit assessment. Intervening on E1 (management support) would, moreover, cascade through both E2 and E3, so if resources are limited, focusing on E1 first produces the widest effect.
In the report: "The root cause of the low product-registration rate is weak management support; insufficient training and members' mistrust are two independent intermediate factors that result from this root cause."
2. Healthcare: Barriers to hand-hygiene practice in a hospital
A hospital's infection-control committee wants to map the reasons why hand-hygiene practice among healthcare workers falls short of its target level. The factors include a heavy workload, an inadequate number of poorly placed hand-sanitiser stations, a lack of awareness on the subject, and the low hand-hygiene practice rate. The committee has judged that a heavy workload makes station use harder and also leaves less time for awareness training, and that both station shortage and lack of awareness lead directly to the low practice rate.
The method combines the direct and indirect effects and derives the levels. Suppose the result places the low hand-hygiene practice rate at the top level, station shortage and lack of awareness at the same intermediate level, and the heavy workload at the bottom level.
The committee hesitates here: increasing the number of stations, a physical intervention, and providing awareness training, a behavioural one, correspond to two different factors at the same level, and ISM gives no priority order between them; which one receives resources first is left to the clinical team's own judgement. The fact that workload appears as the root cause may also push the committee towards concrete steps at the intermediate level, since workload itself may not be changeable on its own.
In the report: "The root cause of the low hand-hygiene practice rate is the heavy workload; station shortage and lack of awareness are two independent factors that result from this root cause."
3. Disaster Management: Factors delaying post-earthquake relief distribution
A disaster coordination centre is examining why it takes so long for emergency relief supplies to reach an area after an earthquake. The factors include weak information-sharing between regions, outdated inventory records, centralised and slow allocation of logistics vehicles, and the late arrival of supplies. The centre has judged that weak information-sharing harms both the currency of inventory records and the speed of vehicle allocation, and that both of these lead directly to delay.
The method processes these relationships and derives the levels. Suppose the result places the late arrival of supplies at the top level, outdated inventory records and slow vehicle allocation at the same intermediate level, and weak information-sharing at the bottom level.
The coordination centre hesitates here: improving the inventory system requires a technical investment, while speeding up vehicle allocation requires an organisational change; although the two sit at the same level, their costs and implementation times differ. ISM does not show this cost difference, only the structural position; resource planning must be carried out separately.
In the report: "The root cause of the late arrival of relief supplies is weak information-sharing between regions; outdated inventory records and slow vehicle allocation are two independent factors that result from this root cause."
4. What Not to Do
Three concrete errors can be shown using the Case 1 table. The first is ranking E2 and E3, which sit at the same level, by saying "E2 matters more than E3"; ISM gives no priority order between factors at the same level, it shows only that they are independent. The second is noticing that E1 indirectly affects E4 and then adding this to the input table by hand as "E1 directly affects E4"; indirect effects are computed by the method itself and must not be mixed into the input. The third is assuming that intervening directly on E4, the outcome at the top of the hierarchy, will solve the problem; E4 is an outcome, and a lasting solution is aimed at the root cause, E1.
Sources
For the formulas behind each step, the intermediate tables and citation formats, see the DecisionMind method page: decisionmind.app/library/ism
Warfield, J. N. (1973). On arranging elements of a hierarchy in graphic form. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(2), 121–132. DOI: 10.1109/TSMC.1973.5408493
Watson, R. H. (1978). Interpretive structural modeling—A useful tool for technology assessment? Technological Forecasting and Social Change, 11(2), 165–185. DOI: 10.1016/0040-1625(78)90028-8
Sushil (2012). Interpreting the Interpretive Structural Model. Global Journal of Flexible Systems Management, 13(2), 87–106. DOI: 10.1007/s40171-012-0008-3
Attri, R. (2017). Interpretive structural modelling: a comprehensive literature review on applications. International Journal of Six Sigma and Competitive Advantage, 10(3–4), 258–331. DOI: 10.1504/ijssca.2017.086597