Escalation is a serious management problem, and sunk costs are believed to be a key factor in promoting escalation behavior. While many laboratory experiments have been conducted to examine the effect of sunk costs on escalation, there has been no effort to examine these studies as a group in order to determine the effect size associated with the so-called “sunk cost effect.” Using meta-analysis, we analyzed the results of 20 sunk cost experiments and found: (1) a large effect size associated with sunk costs, and (2) stronger effects in experiments involving information technology (IT) projects as opposed to non-IT projects. Implications of the results and future research directions are discussed.
TopIntroduction
The amount of money already spent on a project (level of sunk cost), together with other factors, can bias managers’ judgment, resulting in “escalation of commitment” behavior (Brockner, 1992) in which failing projects are permitted to continue. Project escalation can absorb valuable resources without producing the intended results. While escalation is a general phenomenon occurring with any type of project, software projects may be particularly susceptible to this problem (Keil et al., 2000a).
Prior research has identified psychological as well as other factors that can promote escalation (Staw & Ross, 1987). The sunk cost effect is a psychological factor that can promote escalation and refers to the notion that people have a greater tendency to continue a project once money, time, and effort have been invested (Arkes & Blumer, 1985).
There are several possible explanations for the sunk cost effect. Chief among these is prospect theory (Brockner, 1992; Kahneman & Tversky, 1979), which suggests that people will choose to engage in risk-seeking behavior when faced with a choice between losses. According to prospect theory, people will prefer to make additional investments (even when the payoff is uncertain) rather than terminating a project and “losing” all of the monies already spent.
In the context of software projects, the intangible nature of the product (Abdel-Hamid & Madnick, 1991) can make it difficult to estimate the amount of work completed. This difficulty manifests itself in the “90% complete syndrome”1, which may promote the sunk cost effect by giving a false perception that most of the required money, time, and effort have already been expended.
To investigate the sunk cost effect, researchers have conducted many role-playing experiments in which sunk cost levels are manipulated to determine if they have an effect on decision-making (e.g., Garland, 1990;Garland & Newport, 1991). These published experiments suggest that there is broad agreement that sunk cost increases commitment to projects. However, there are a couple of unanswered questions. First, while prior studies have conducted statistical significance testing, they do not provide much information about the magnitude of the sunk cost effect. Second, although there have been claims that IT projects are more prone to the sunk cost effect, there have been no prior studies to determine if the magnitude of the sunk cost effect is larger in an IT project context than it is in a non-IT project context.
Meta-analysis, a literature review method using a quantitative approach, is very good at assessing a stream of research, discovering the consistencies, and accounting for the variability. Therefore, in this study, we conduct a meta-analysis to determine the mean effect size of sunk cost on project escalation and examine variability of effect sizes across experiments. We also examine whether the effect size of the sunk cost effect on project escalation is different for IT vs. non-IT project contexts.