The Challenge of Validation

By Martin Neumann


In November 2021 Chattoe-Brown initiated a discussion at the SimSoc list on validation which generated quite some traffic on the list. The interest in this topic revealed that empirical validation provides a notorious challenge for agent-based modelling. The discussion raised many important points and questions which even motivated a “smalltalk about big things” at the Social Simulation Fest 2022. Many contributors highlighted that validation cannot be reduced to the comparison of numbers between simulated and empirical data. Without attempting a comprehensive review of this insightful discussion, it has been emphasized that different kinds of science call for different kinds of quality criteria. Prediction might be one criterium that is particularly important in statistics, but that is not sufficient for agent-based social simulation. For instance, agent-based modelling is specifically suited for studying complex systems and turbulent phenomena. Modelling also enables studying alternative and counterfactual scenarios which deviates from the paradigm of prediction as quality criterion. Besides output validation, other quality criteria for agent-based models include for instance input validation or process validation, reflecting the realism of the initialization and the mechanisms implemented in the model.

Qualitative validation procedures

The brief introduction is by no means an exhaustive summary of the broad discussion on validation. Already the measurement of empirical data can be put into question. Less discussed however, had been the role which qualitative methods potentially could play in this endeavor. In fact, there has been a long debate in the community of qualitative social research on this issue as well. Like agent-based social simulation also qualitative methods are challenged by the notion of validation. It has been noted that already the vocabulary that is used in attempts to ensure scientific rigor has a background in a positivist understanding of science whereas qualitative researcher often take up constructivist or poststructuralist positions (Cho and Trent 2006). For this reason, in qualitative research sometimes the notion of trustworthiness (Lincoln and Guba 1985) is preferred rather than speaking of validation. In an influential article (according to google scholar cited more than 17.000 times in May 2023) Creswell and Miller (2000) distinguish between a postpositivist, a constructivist, and a critical paradigm as well as between the lens of the researcher, the lens of the study participants, and the lens of external people and assign different validity procedures for qualitative research to the combinations of these different paradigms and lenses.

Paradigm/ lenspostpositivistconstructivistcritical
Lens of researchertriangulationDisconfirming evidenceReflexivity
Lens of study participantsMember checkingEngagement in the fieldCollaboration
Lens of external peopleAudit trialThick descriptionPeer debriefing
Table 1. validity procedures according to Creswell and Miller (2000).

While it remains contested if the validation procedure depends on the research design, this is at least a source of different accounts. Others differentiate between transactional and transformational validity (Cho and Trent 2006). The former concentrates on formal techniques in the research process for avoiding misunderstandings. Such procedures include for instance, techniques such as member checking. The latter account perceives research as an emancipatory process on behalf of the research subjects. This goes along with questioning the notion of absolute truth in the domain of human sciences which calls for alternative sources for the legitimacy of science such as emancipation of the researched subjects. This concept of emancipatory research resonates with participatory modelling approaches. In fact, in participatory modelling accounts some of these procedures are well-known even though they differ in terminology. The participatory approach originates from research on resource management (Pahl-Wostl 2002). For this purpose, integrated assessment models have been developed, inspired by the concept of post-normal science (Funtowicz and Ravetz 1993). Post-normal science emphasizes the communication of uncertainty, justification of practice, and complexity. This approach recognizes the legitimacy of multiple perspectives on an issue, both with respect to multiple scientific disciplines as well as lay men involved in the issue. For instance, Wynne (1992) analyzed the knowledge claims of sheep farmers in the interaction with scientists and authorities. In such an extended peer community of a citizen science (Stilgoe 2009), lay men of the affected communities play an active role in knowledge production, not only because of moral principles of fairness but to increase the quality of science (Fjelland 2016). One of the most well-known participatory approaches is the so-called companion modelling (ComMod) developed at CIRAD, a French agricultural research center for international development. The term companion modelling has been coined originally by (Barreteau et al 2003) and been further developed to a research paradigm for decision making in complex situations to support sustainable development (Étienne 2014). In fact, these approaches have a strong emancipatory component and rely on collaboration and member checking for ensuring resonance and practicality of the models (Tesfatsion 2021).

An interpretive validation procedure

While the participatory approaches show a convergence of methods between modelling and qualitative research even though they differ in terminology, in the following a further approach for examining the trustworthiness of simulation scenarios will be introduced that has not been considered so far, namely interpretive methods from qualitative research. A strong feature of agent-based modelling is that it allows for studying “what-if” questions. The ex-ante investigation of possible alternative futures enables identifying possible options of action alternatives but also detecting early warning signals of undesired developments. For this purpose, counterfactual scenarios are an important feature of agent-based modelling. It is important to note in this context that counterfactuals do not match empirical data. In the following it is suggested to examine the trustworthiness of counterfactual scenarios by using methods from objective hermeneutics (Oevermann 2002), the so-called sequence analysis (Kurt and Herbrik 2014). In terms of Creswell and Miller (2000) the examination of trustworthiness is from the lens of the researcher and a constructivist paradigm. For this purpose, simulation results have to be transformed into narrative scenarios, a method which is described in (Lotzmann and Neumann 2017).   

In the social sciences, sequence analysis is regarded as the central instrument of hermeneutic interpretation of meaning. It is “a method of interpretation that attempts to reconstruct the meaning of any kind of human action sequence by sequence, i.e. sense unit by sense unit […]. Sequence analysis is guided by the assumption that in the succession of actions […] contexts of meaning are realized …” (Kurt and Herbrik 2014: 281). A first important rule is the sequential procedure. The interpretation takes place in the sequence that the protocol to be analyzed itself specifies. It is assumed that each sequence point closes possibilities on the one hand and opens new possibilities on the other hand. This is done practically by sketching a series of stories in which the respective sequence passage would make sense. The basic question that can be asked of each sequence passage can be summarized as, “Consider who might have addressed this utterance to whom, under what conditions, with what justification, and what purpose?” (Schneider 1995: 140). The answers to these questions are the thought-experimentally designed stories. These stories are examined for commonalities and differences and condensed into readings. Through the generation of readings, certain possibilities of connection to the interpreted sequence passage become visible at the same time. In this sense, each step of interpretation makes sequentially spaces of possibility visible and at the same time closes other spaces of possibility.

In the following it will be argued that this method enables an examination of the trustworthiness of counterfactual scenarios using the example of a counterfactual simulation scenario of a successful non-violent conflict regulation within a criminal group: ‘They had a meeting at their lawyer’s office to assess the value of his investment, and Achim complied with the request. Thus, trust was restored, and the group continued their criminal activities’ (names are fictitious). Following Dickel and Neumann (2021) it is argued that this is a meaningful story. It is an example of how the linking of the algorithmic rules generates something new from the individual parts of the empirical material. However, it also shows how the individual pieces of the puzzle of the empirical data material are put together to form a collage that tells a story that makes sense. A sequence that can be interpreted in a meaningful way is produced. It should be noted, however, that this is a counterfactual sequence. In fact, a significantly different sequence is found in the empirical data: ‘Achim was ordered to his lawyer’s office. Instead of his lawyer, however, Toby and three thugs were waiting for him. They forced him to his knees and pointed a machine gun at his stomach’. In fact, this was by no means a non-violent form of conflict regulation. However, after Achim (in the real case) was forced to his knees by three thugs and threatened with a machine gun, the way to non-violent conflict regulation was hardly open any more. The sequence generated by the simulation, on the other hand, shows a way how the violence could have been avoided – a way that was not taken in reality. Is this now a programming error in the modeling? On the contrary, it is argued that it demonstrates the trustworthiness of the counterfactual scenario: from a methodological point of view a comparison of the factual with the counterfactual is instructive: Factually, Achim had a machine gun pointed at his stomach. Counterfactually, Achim agreed on a settlement. From a sequence-analytic perspective, this is a logical conclusion to a story, even if it does not apply to the factual course of events. Thus, the sequence analysis shows that the simulation here has decided between two possibilities, a path branching in which certain possibilities open and others close.

The trustworthiness of a counterfactual narrative is shown by whether 1) a meaningful case structure can be generated at all, or whether the narrative reveals itself as an absurd series of sequence passages from which no rules of action can be reconstructed. 2) it can be tested whether the case structure withstands a confrontation with the ‘external context’ and can be interpreted as a plausible structural variation. If both are given, scenarios can be read as explorations of the space of cultural possibilities, or of a cultural horizon (in this case: a specific criminal milieu). Thereby the interpretation of the counterfactual scenario provides a means for assessing the trustworthiness of the simulation.


Barreteau, O., et al. (2003). Our companion modelling approach. Journal of Artificial Societies and Social Simulation 6(2): 1.

Cho, J., Trent, A. (2006). Validity in qualitative research revisited. Qualitative Research 6(3), 319-340.

Creswell, J., Miller, D. (2000). Determining validity in qualitative research. Theory into Practice 39(3), 124-130.

Dickel, S., Neumann. M. (2021). Hermeneutik sozialer Simulationen: Zur Interpretation digital erzeugter Narrative. Sozialer Sinn 22(2): 252-287.

Étienne, M. (2014)(Ed.). Companion Modelling: A Participatory Approach to Support Sustainable Development. Springer, Dordrecht.

Fjelland, R. (2016). When Laypeople are Right and Experts are Wrong: Lessons from Love Canal. International Journal for Philosophy of Chemistry 22(1): 105–125.

Funtowicz, S., Ravetz, J. (1993). Science for the post-normal age. Futures 31(7): 735-755.

Kurt, R.; Herbrik, R. (2014). Sozialwissenschaftliche Hermeneutik und hermeneutische Wissenssoziologie. In: Baur, N.; Blasius, J. (eds.): Handbuch Methoden der empirischen Sozialforschung, pp. 473–491. Springer VS, Wiesbaden.

Lotzmann, U., Neumann, M. (2017). Simulation for interpretation. A methodology for growing virtual cultures. Journal of Artificial Societies and Social Simulation 20(3): 13.

Lincoln, Y.S., Guba, E.G. (1985). Naturalistic Inquiry. Sage, Beverly Hill.

Oevermann, U. (2002). Klinische Soziologie auf der Basis der Methodologie der objektiven Hermeneutik. Manifest der objektiv hermeneutischen Sozialforschung (Download am 01.03.2020).

Pohl-Wostl, C. (2002). Participative and Stakeholder-Based Policy Design, Evaluation and Modeling Processes. Integrated Assessment 3(1): 3-14.

Schneider, W. L. (1995). Objektive Hermeneutik als Forschungsmethode der Systemtheorie. Soziale Systeme 1(1): 135–158.

Stilgoe, J. (2009). Citizen Scientists: Reconnecting Science with Civil Society. Demos, London.

Tesfatsion, L. (2021). “Agent-Based Modeling: The Right Mathematics for Social Science?,” Keynote address, 16th Annual Social Simulation Conference (virtual), sponsored by the European Social Simulation Association (ESSA), September 20-24, 2021.

Wynne, B. (1992). Misunderstood misunderstanding: social identities and public uptake of science. Public Understanding of Science 1(3): 281–304.

Neumann, M. (2023) The Challenge of Validation. Review of Artificial Societies and Social Simulation, 18th Apr 2023.

© The authors under the Creative Commons’ Attribution-NoDerivs (CC BY-ND) Licence (v4.0)The Challenge of Validation

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.