5.2 A perfect mixing model
5.2.1 Model development
5.2.1.3 Adding product recall
To date, the SARF literature has provided little evidence on the ways in which risks can be amplified by the very organizations having responsibilities to handle them. The main exception to this is the work of Freudenburg (2003) who has suggested that the perception of organizational functioning can have a great influence on perceptions of real risks. He used the term ‘recreancy’ to denote the failure of an organization to meet its obligations. In particular, product recall is one of the most common responses from the company involved (for example, Choi and Chung, 2013; De Matos and Rossi, 2007; Souiden and Pons, 2009) and of the most important sources of negative publicity that can significantly raise public concern (for example, Desai and Patel, 2014; Korkofingas and Ang, 2011; Magno, 2012; Souiden and Pons, 2009). However, there has been little work on how product recalls contribute to the process of risk amplification in risk events. How risk is socially amplified and how public risk perception evolves over time in a recall event is still a crucial topic to be addressed.
This stage of model development introduces product recall to look at how individuals perceive the risk of products in question and how they make a decision about actions to take upon hearing news of a product recall. The product recall literature was reviewed to extract decision rules relating to consumer responses to recalls. As demonstrated by Chattoe-Brown (2014), agent-based modelling provides a way of integrating different kinds of research data, and for this study the aim was to draw on various pieces of empirical research in the recall literature. The recall literature survey is not presented in this thesis, because this study does not contribute to the recall literature but is used in contextualising risk amplification down to a specific domain of crisis, in this case product recall. The extraction of decision rules for the agent model was carried out as follows:
1) From each empirical study of product recall, identify the causes (independent variables) and effects (dependent variables) underlying agent (e.g. consumer, organization, and media) behaviour.
2) Translate these aggregate, statistical relationships into condition-action decision rules, mapping causes into condition codes and effects into action codes. The complete result of this exercise is tabulated in detail in Appendix A.
3) For condition codes, use a threshold to represent the magnitude of numerical variables (e.g. if the condition is ‘consumers perceive high risk for the defective product’, then the condition code is ‘Perceived Risk > RH ’), and use “True” or “False” to denote the value of
Boolean variables (e.g. if the condition is ‘the company issues a product recall’, then the condition code is ‘Recall = True’).
consumers are highly involved with recalled products and perceive the CEO’s apology speech as truly sincere, then an apology has more positive effects on consumer attitudes’, then the condition codes will be written as ‘Involvement > IH and Sincerity (Apology) > SH’, and the
action code will be expressed as ‘Attitudet = Attitudet1 + c × Involvement × Sincerity (Apology)’ with some constant c.
5) Classify decision rules in terms of dependent variables, i.e. effects, to facilitate further selection. The decision rules can be classified into four categories generally: organizational reputation, risk perception, brand attitude, and purchase intention.
6) Specify the variables that are significant to consumers’ perceptions of risk in the particular context being modelled. For example, product involvement significantly affects consumers’ judgment of risk associated with a questionable product (Choi and Lin, 2009a; De Matos and Rossi, 2007), but it is not considered in the milk contamination context model. This is because product involvement is measured by many indicators with respect to consumers’ inherent needs, values, and interests (Zaichkowsky, 1985), and incorporating this variable in the model will add excessive complexity to the model.
Based on the agent rules extracted from the recall literature, recall information, recall timing, and recall voluntariness are chosen as critical elements influencing consumer reaction to recall events. This justifies part of the conceptual model presented in Figure 5.2. Thus there are two main aspects to the effect of a product recall on public perceptions: 1) it provides information that the product is defective, which combines with the three sources of information already described to determine public risk beliefs; 2) it increases or decreases the public’s trust in the company – depending on its timing, and whether it is voluntary. This is shown in the conceptual model in Figure 5.2.
The second effect, on recreancy, will be dealt with in the next section. For the first, information effect, the model assumes a single producer agent. The producer issues a recall message during the time between contamination release and contamination termination, but does not withdraw the product, and therefore consumers can continue to experience a contamination event. Note that the contamination level Chigh
t stays constant when a recallis in force, which means that the recall does not affect the proportion of products that are contaminated. This represents a situation in which the producer simply makes an announcement of recall and contaminated products are still on sale and a situation in which producer behaviour has no impact on the likelihood of consumers experiencing contamination. It obviously differs from a more realistic situation in which a recall is associated with product withdrawal that can affect contamination level over time. The delay between recall announcement and recall action is assumed to be zero for the sake of convenience. The recall announcement a t
is either 0 (no announcement) or 1 (announcement). The recall is issuedwith a probability equal to the contamination risk C t
at any given time. This means the recall timing is random but the most likely delay is 1 tick only and the delay is distributed as an exponential distribution. There is likely to be some, small delay between the sudden contamination increase and the recall. After a recall announcement has been made, the recall,
0, 1r t , stays in force until the contamination level falls to its original, very low level,
that is, r t
0 until a t
1, then r t
1 until tTend, and then r t
0. The public’srisk belief decision rule now also incorporates recall information (simply a binary value):
1 1 1 1 4 K i i nj i nj b t b t b t e t r t K
(5.4)These four factors including an agent’s prior belief, neighbours’ perception, direct experience, and product recall information is described in summary as the ‘discovery’ component of the conceptual model (see Figure 5.2).