Two types of typicality: Rethinking the role of statistical typicality in ordinary causal attributions

Abstract

Recent work on the role of norms in the use of causal language by ordinary people has led to a consensus among several researchers: The consensus position is that causal attributions are sensitive to both statistical norms and prescriptive norms. But what is a statistical norm? We argue that there are at least two types that should be distinguished-agent-level statistical norms and population-level statistical norms. We then suggest an alternative account of ordinary causal attributions about agents (the responsibility view), noting that this view motivates divergent predictions about the effect of information about each of the two types of statistical norms noted. Further, these predictions run counter to those made by the consensus position. With this set-up in place, we present the results of a series of new experimental studies testing our predictions. The results are in line with the responsibility view, while indicating that the consensus position is seriously mistaken. Consider the following scenario based on a thought experiment by Joshua Knobe (2006):Lauren and Jane both work for a company that uses a mainframe that can be accessed from terminals on different floors of its building. The mainframe has recently become unstable, so that if more than one person is logged in at the same time, the system crashes. Therefore, the company has instituted a temporary policy restricting the use of terminals so that two terminals are not used at the same time until the mainframe is repaired. The policy prohibits logging into the mainframe from the terminal on any floor except the ground floor.One day, Lauren logged into the mainframe on the authorized terminal on the ground floor at the exact same time that Jane logged into the mainframe on the unauthorized terminal on the second floor. Lauren and Jane were both unaware that the other was logging in. Sure enough, the system crashed.When this scenario is given to people without training in philosophy, they are significantly more likely to say that Jane caused the system to crash than Lauren (Livengood, Sytsma, and Rose, Not surprisingly, some philosophers have sought to explain such results away, aiming to preserve the assumption that the ordinary concept of causation is purely descriptive in character.For example, Julia Driver (2008a) suggests that ordinary causal attributions might still be best explained in terms of a type of descriptive consideration-statistical norms. Thus, it might be that Jane is identified as the cause of the system crashing, in the above example, not because she violated a prescriptive norm per se, but because she did something that is statistically atypical in doing so-it presumably being unusual to violate company policy. Subsequent empirical work has suggested against this, however. In fact, a consensus position has emerged that focuses on norms more generally: Ordinary causal attributions are thought to be directly sensitive both to behaviors being out of the prescriptive norm (impermissible) and to behaviors being out of the s...

Normality and actual causal strength

“…Particularly in the case of prescriptive norms, there are several explanations for this effect. This effect is predicted by our model, but it is also predicted by a wide variety of other views which invoke everything from conversational pragmatics to motivational bias (Alicke et al, 2011;Driver, 2008;Strevens, 2013;Sytsma et al, 2012), in addition to being predicted by one of the measures discussed above (namely SP ). However, our model predicts a further effect, for both prescriptive and statistical norms.…”

“…Researchers have suggested that the effect might arise as a result of conversational pragmatics (Driver, 2008), motivational bias (Alicke et al, 2011), relativity to frameworks (Strevens, 2013), responsibility attributions (Sytsma, Livengood, & Rose, 2012), or people's understanding of the question . Here, we will be exploring a general approach that has been defended by a number of researchers in recent years, namely, that abnormal inflation reflects a process in which certain counterfactuals are treated as in some way more relevant than others (Blanchard & Schaffer, 2016;Halpern & Hitchcock, 2015;Knobe, 2010;Phillips et al, 2015).…”

Following the FAD: Folk Attributions and Theories of Actual Causation

“…In the previous section, however, we saw that statistical norms did not have a notable impact on folk causal attributions for the Lauren alone case. And this finding is in keeping with the results reported in Sytsma, Livengood, and Rose (2012 (2012) for discussion. That said, one might think that there is good reason to believe that statistical norms play an independent role in some folk causal attributions.…”

“…Hitchcock's theory fails to satisfy the FAD for the original Lauren and Jane case at least in part because the rules of thumb for assigning default and deviant values say nothing about prescriptive norms. Many studies have shown that ordinary causal attributions are sensitive to such norms (e.g., Alicke, 1992;Hitchcock and Knobe, 2009;Alicke et al, 2011;Sytsma, Livengood, and Rose, 2012). Since the Lauren and Jane case includes information about prescriptive norms, it is not completely surprising that Hitchcock's theory fails to satisfy the FAD with respect to it.…”

“…In Sytsma, Livengood, and Rose (2012) we argued that there are two types of typicality that need to be considered when testing the effects of statistical norms on folk causal attributions-what is typical for an agent (agentlevel typicality) and what is typical for members of the relevant population to which that agent belongs (population-level typicality). 37 In Study 17, we tested the role of both types of typicality on a further variation on the Lauren and Jane without permissibility information scenario.…”