5.5 Validation and Verification
5.5.2 Validation
Verification of complex systems models is certainly difficult (Pathak et al., 2007a; Pathak et al., 2007b; Davis et al., 2007; Pathak et al., 2010). The main
problem being that any valid model is nondeterministic and emergent, inferring that small changes in some variables may produce disproportionate changes in outcomes. Verification is further complicated by the difficulty of gathering real- world empirical data against which the system can be compared. This is particularly true of systems such as supply networks that are of large-scale and geographical spread bound together by autonomous decisions made under conditions of bounded rationality.
This thesis uses two approaches to validating the computer model: Sound theoretical anchors for the algorithms
Verification of algorithm descriptions using an expert panel. 5.5.2.1 Theoretical Anchors
Table 5-16 summarises the theoretical anchors for each element of the model described previously in this chapter.
Table 5-16: Theoretical anchors for each element of the model
Sub-model / Core Component
Algorithm Theoretical anchor
Core component Supply risk TCE: (Kraljic, 1983;
Williamson, 1996)
Collaboration Relational risk SET: (Hunt and Morgan,
1994)
Information sharing Supply chain management: (Fisher, 1997)
Price Competition Price adaptation Economics: (Marshall , 1930; Williamson, 1996) Purchasing Strategy Single/dual sourcing TCE, purchasing (Kraljic,
1983)
Inventory Management Inventory management Operational research: (Waters, 2003)
Birth and Death New entrants CAS: (Bak, 1999)
Supplier Selection Supplier selection criteria Supplier selection : (Weber et al., 1991; Dempsey, 1978; Dickson, 1966) Relational inertia SET (Hunt and Morgan,
5.5.2.2 Panel Verification
In addition to the theoretical anchors provided above, the model was validated at a number of workshops comprising practitioners and academics from the field. The selection of the panel was designed to balance academics with practitioners. The composition of the panel is given in Table 5-17
Table 5-17: Composition of the validation workshop panel
Panel Member Background Title and affiliation
Colin Dulson Practitioner Supply Chain Director, AAH
Pharmaceutical
Mike Griffiths Academic Department of Informatics,
Shrivenham
Professor Peter Allen Academic Director Complex Systems
Group, Cranfield University
XX Practitioner World Wide Duty Free
Richard Street Practitioner Supply Chain Director,
Mothercare
In addition to the workshops the approach was presented at the Logistics Research Network conference and the European Operations Management Association conference.
The workshops consisted of the author presenting the algorithms described previously and inviting the audience to challenge the algorithms and the underpinning assumptions.
A number of challenges were presented, and these are summarised in Table 5-18 together with the conclusion reached by the author and the conclusion justification.
Table 5-18: Challenges to the model arising from the validation workshops
Challenge Response Justification
EOQ sensitive to holding costs and re-order costs
Agent checks that EOQ covers lead time demand
This amounts to a hybrid approach incorporating elements of both EOQ and order up to levels
Use of a single product does not reflect the complicated nature of real supply chains
The single product abstraction involves less assumptions regarding interdependencies between different products in different organisations which would only obscure the system response and response mechanisms
Single product adopted, but observation noted and reflected in recommendations for further work
Risk assessment is less formalised
This is probably true but it was agreed that it would include the principles of the proposed algorithm
Retain the calculations of risk proposed but note comments and include in
recommendations for further work
Supplier selection not as sophisticated as the model Births of new companies involves an assessment of the market
Undoubtedly true but the operationalisation of this within the model would introduce added complication
It was agreed that market assessments are typically based on bounded rationality and are often naïve
Responses to underutilisation are more varied than price adjustments
This is also true, but
suggestions include activities anticipated in the model such as bundling and strategic location
The single product abstraction does not permit bundling and re-location of resources would require extension of the model run time to beyond
reasonable limits within the broad range of experiments necessary
Product criticality ignored This is also true but the
assumption of using a single product requires the strategic purchasing options to be reduced to whether or not to collaborate, and whether or not to dual source
Whilst the implications of ignoring product criticality are clearly a simplification of Kraljic’s (1983) purchasing strategy they are supported by TCE, SET and indeed to a large extent by Kraljic
Why not allow multiple sourcing to be more than two suppliers
Extending the possibility of multiple sourcing including more than 2 suppliers would require agents to balance the benefit with the resources and costs required to manage the extra suppliers
Two suppliers was felt to be realistic if not all embracing
5.6 Summary
This chapter has described in detail how the underpinning conceptualisation of supply networks as CASs susceptible to normal accidents as described by NAT has been incorporated into the design of an ABM suitable for assimilation in an
experimental design to provide answers to the research questions identified by the literature review.
The model design has been verified at elemental levels through careful stepwise construction, staged testing of components, and integration of components. The algorithms, sub-models and consequently the model itself have been validated by anchoring the design in strong theoretical anchors drawn from SET, TCE, and supply chain management. Further validation was secured through the reflections of an expert panel on the components of the model.
The model was developed using Java programming language in the Anylogic 6 development environment. The model code is provided in Appendix A
The next chapter will describe the design of the experiments used to generate data from which the answers to the research questions can be developed through appropriate analysis congruent with the theoretical underpinnings of CAS and NAT.
6 Experiment Design
Chapter 4 showed how the collection of real-world empirical data using survey methods whilst notionally attractive has practical limitations when trying to understand a complex system’s behaviour in response to events that are unpredictable and often rare. Furthermore, the difficulty in establishing real- world systems boundaries is prohibitively resource expensive.
Chapter 4 also showed how computer models can be built using established supply chain practice as the basis for organisational behaviour. Although the model abstraction is founded on empirically validated organisational behaviours it extends the environment within which these behaviours play out, thereby providing a more realistic representation of the real world than that which would be deliverable from even the highest quality surveys.
Chapter 5 specified a computer model that is anchored in extant theory, but which accepts a new dynamic environment. Furthermore, the model specification, cognisant of the research questions identified in Chapters 2 and 3, provides the basis for submitting the model to a series of experiments designed to answer the research questions. As a consequence, this chapter is primarily concerned with describing a rigorous and robust approach to the definition of an experimental program.
This chapter starts by describing the design principles adopted and then uses these to specify an experiment design before considering the inference this has on the data collected with a view to the subsequent analysis. The chapter concludes by integrating design principles, design and analysis into a validation of the approach taken.