• No results found

7 Chapter 2. Predicting Failures of High Tech Innovations-in-Use: Application of

2.2 Theoretical Foundation and Hypotheses Development

2.2.3 Development of Hypothesis 3 (H3) on the Consistency of Prediction

Beyond precision, consistency or repeatability is a critical consideration in any prediction process.

Presence of systematic judgment bias has the potential to adversely influence the consistency of prediction. Firms often exhibit judgment bias in reacting to market signals when making strategic decisions, where prediction is fundamental to decision making. Specifically, firms in the medical device industry – the high tech industry that serves as the empirical setting for this study – appear to exhibit judgment bias in the form of over-reaction or under-reaction to user reported adverse events related to high tech innovations-in-use in the marketplace. Such judgment bias has the potential to negatively impact a medical device firm’s ability to predict the failure of a high tech innovation-in-use, i.e., medical device recall.

The theoretical perspective of system neglect is founded on the notion that the decision makers exhibit systematic under-reaction in environments of high instability and vice-versa. In this

study, we follow Massey and Wu (2005) to investigate if the data stream on adverse events indicates the presence of systematic judgment bias in detecting the market signals of failure of a high tech innovation-in-use that, in turn, may adversely influence the consistency of prediction. Systematic judgment bias in detection would either mean an over-reaction to a weak signal or an under-reaction to a strong signal. Massey and Wu (2005) experimentally studied the causes of over-reaction and under-reaction in detecting regime shifts. Their study showed that decision makers show systematic bias towards over-reaction in a relatively stable environment and a bias toward under-reaction in a relatively unstable environment. Further, Massey and Wu (2005) concluded that decision makers put excessive attention to the signal and less attention to the system that generates the signal probably due to the relative saliency of the signal with respect to the system parameters that generates the signal. Another significant study in the system neglect literature is Kremer et al. (2011) who investigated judgment bias and signal detection issues in the context of supply-chain forecasting. Kremer et al. (2011) found that decision makers systematically over-reacted to changes in a stable environment and under-reacted to changes in an unstable demand environment.

Now, we synthesize the concepts from the literatures on system neglect and signal detection theory to the third study hypothesis related to consistency of prediction of failures of high tech innovations-in-use. According to SDT, the decision makers reaction is captured through the decision criterion parameter, &' (see Figure [2.1]). The decision criterion represents the propensity of a decision maker to over-react or under-react, given a signal-noise distribution. The question that would need to be answered is what would be a reasonable choice of the decision criterion for a given noise and signal-plus-noise distribution. Is there an optimality condition for the parameter &'? Under what conditions would a decision maker adopt a high level for &' versus a low level for &'? A relatively high value for &' would mean under-reaction to signals and vice-versa.

As has been mentioned earlier, there are costs associated with the decision process. If there is a signal in the data that is not detected, then a decision maker incurs an opportunity cost. Similarly, if there are no signals but the decision maker wrongly detects a signal, then the decision maker incurs a sunk cost. This scenario is pictorially depicted in Figure [2.3].

Figure 2.3 Errors of Detection with Associated Cost

A rational decision maker would like to minimize the expected costs associated with the two errors. We assume that there are no additional marginal costs associated with the correct detections. The decision criterion of the rational decision maker can be stated as:

minE

F GHI∗ =(K ) + H ∗ =(02)L … [2.5]

Figure 2.4 Operationalization of the Signal Detection Theory (SDT) Depicting the Effect of Decision Process (Decision Criterion)

The objective function for the optimization problem is:

minE

F NHI∗ O P(Q| , ) QEF

ST + H ∗ O P(Q|T

EF , ) QU … [2.6]

Normalizing the mean of the noise distribution to zero (WLOG), we further simplify the optimal value for the decision criterion as follows. See Appendix A.3 for proof.

&'= 2 +W

… [2.7]

where,

W = log(\) = log ]H

HI^ = log ] ) _ H5

`aa5 ) H5 ^

From the above relationship, we can conclude the following:

(i) Holding all other factors a constant, &' increases as signal strength increases. In other words, as the detectability increases as the signal becomes more precise.

(ii) Again, holding all other factors a constant, the &' increases as W increases. Since the parameter W = 15*( ) _ H5 |0-1 21- +) − 15*(`aa5 ) H5 |K ), as the sunk cost increases in relation to the opportunity cost, &' increases. Similarly, as the opportunity cost increases in relation to the sunk cost, &' decreases. However, pre-facto, neither the sunk cost not the opportunity cost is precisely known to any decision maker.

Decision makers interpret values of this cost from the characteristics of a high tech innovations-in-use or the characteristics of an adverse event. As the severity of adverse events increases, the interpreted future cost estimates increase. Due to inherent risk averseness of decision makers, a disproportionately high cost may be assigned to adverse events with high severity measure. This would lead decision makers to choose a low decision threshold, which would result in over-reaction to market signals. On the other hand, a low severity measure may lead to choose a high decision threshold leading to under-reaction.

(iii) &' increases (case of under-reaction) with system variance when W is positive, i.e., when the perceived sunk cost of a false detection is more than the perceived opportunity cost of no detection. However, &' decreases (case of over-reaction) with system variance when W is negative, i.e., when the perceived future opportunity cost of no detection is more than the perceived sunk cost of false detection. As mentioned, higher severity leads to a perceived higher future opportunity cost which may be disproportionate to the actual cost given inherent risk averseness of decision makers. Thus, we may expect over-reaction bias to increase with increased severity and increased perceived opportunity cost of not detecting a credible signal. Incorporating this in the equation for decision criterion, we can depict the above conclusions pictorially in Figure [2.5].

Figure 2.5 Depicting the Operationalization of the Concept of System Neglect and the Resulting Biases

Building on the depiction in Figure [2.5], we posit the following set of hypotheses:

Hypothesis 3A (H3A): Noise-to-signal ratio (in H2) is associated with systematic bias in the prediction of failure of a high tech innovation-in-use.

High noise-to-signal ratio is associated with under-reaction bias and low noise-to-signal ratio is associated with over-reaction bias.

Hypothesis 3B (H3B): Severity of an innovation failure (in H2) is associated with systematic bias in the prediction of failure of a high tech innovation-in-use. High severity is associated with over-reaction bias and low severity is associated with under-reaction bias.

Note, we use noise-to-signal ratio, the inverse of signal-to-noise ratio, to indicate high variance of the noise-plus-signal distribution. Noise-to-signal ratio makes the interpretation of H3A and H3B direct and simple, and in keeping with the system neglect literature.

Variance (Noise-to-Signal Ratio)