Daryl Bem (Reference Bem and Berkowitz1972) foresaw a slippery slope in resorting to unconscious processes as explanatory variables. The point was not to deny that mental activity could occur outside of awareness, but to warn researchers about abandoning sound epistemological practices when explaining phenomena. We add to this a concern about deriving prescriptions from theories that have not been extensively tested; from studies that have not been widely replicated; and from the usage of terms, such as unconscious, that have not been consistently defined. In this light, we commend Newell & Shanks (N&S) for their critical review of unconscious influences on decision making and agree with their conclusions. However, we identify areas needing clarification following their use of the lens model (Brunswik Reference Brunswik1952; Hammond & Stewart Reference Hammond and Stewart2001) as an interpretive framework.

Greenwald (Reference Greenwald1992, p. 775) concluded that unconscious cognition occurs rapidly and is “severely limited in its analytic capability.” In contrast, unconscious-thought theory (UTT; Dijksterhuis & Nordgren Reference Dijksterhuis and Nordgren2006) assumes a deliberative and temporally extended unconscious that can sift through vast amounts of information to arrive at optimal decisions. UTT experiments, as described by N&S, use a unique multi-attribute evaluation task that presents participants with cue values, sequentially and randomly. We agree with N&S that the evidence supporting the benefits of unconscious thinking is weak. The application of the lens model to research based on UTT, however, is not self-evident. First, N&S identify points within the lens model where lack of awareness could take place; however, lack of awareness and UTT's unconscious processing do not equate. Second, it is worth adding that UTT is not Brunswikian in spirit. Representative design is absent; most UTT studies use a small set of objects with attributes and their values selected by the experimenters. The cue values are presented in random order across cases in UTT studies, whereas in most judgment situations the unit of information acquisition is organized first by case/object. And in terms of accuracy, UTT uses agreement between judgments and the experimentally defined “best option” rather than by correspondence of judgments with agreed-upon environmental criteria. N&S note that within the lens model, a source of lack of awareness may occur at the weighting of cues stage. We add that this relates to UTT's Principle 4, which claims that unconscious weights the relative importance of attributes in an efficient manner; but the evidence supporting this principle is missing (see González-Vallejo et al. Reference González-Vallejo, Lassiter, Bellezza and Lindberg2008).

From a historical perspective, we note that multiple-cue judgment research did not directly attempt to study unconscious processes, even when considering the topic of self-insight. Hammond (Reference Hammond1955) advocated for the use of a quantitative technique to make the judgment process explicit, in the sense of revealing which cues were most influential. The impetus behind this was not rooted in discovering unconscious processes, but simply in the realization that judgments had not been systematically studied and were impacting lives in important domains (e.g., clinical judgments). Because most psychological and physical processes are not easy to verbalize, modern psychological research shifted from relying on verbal reports to using psychometric techniques, and this ensued in judgment research as well. The focus on self-insight evolved from contrasting statistically estimated cue weights with the verbal descriptions of what was important. As N&S show, that agreement is variable, but the goal of the approach was not about understanding unconscious processes but rather about employing statistics to help individuals communicate the basis for their judgments (Hammond & Adelman Reference Hammond and Adelman1976). More generally, mathematical models of cognition are ubiquitous and use many function forms. The view that individuals may be able to verbalize model parameters, thus showing self-insight, is an interesting but not very useful proposition. Indeed, even if we think mathematical models are about the unconscious, a notion like self-awareness would be unnecessary. We do agree with N&S that the validities of measures of self-insight are questionable, but we add the caveat that both subjective assessments and statistical estimates of parameters depend on a model, so neither has priority over the process they are measuring.

From another perspective, lens model research has yielded a rich body of work (Karelaia & Hogarth Reference Karelaia and Hogarth2008). The main results are (a) linear models capture similar and relatively high proportions of variance in environmental outcomes and in human judgments, and (b) judges reach high levels of accuracy when predicting criteria in many domains. Factors that affect accuracy can be safely classified as task/environmental characteristics (see also Stewart et al. Reference Stewart, Roebber and Bosart1997). Therefore, on logical grounds, there is little need to resort to unconscious thinking as an explanatory variable of judgments, or as a mechanism for improving accuracy.

Because of Hammond's central role in lens model research, we feel his views on intuition must be mentioned. Cognitive continuum theory (CCT; Hammond Reference Hammond1986; Reference Hammond1996; Hammond et al. Reference Hammond, Hamm, Grassia, Pearson, Goldstein and Hogarth1997) states that both tasks and cognitive processes are located on an intuitive-to-analytic continuum. According to Hammond, most judgment is “quasi-rational,” involving a combination of intuition and analysis (Hammond Reference Hammond1996), thus contrasting with dual-process conceptions (Epstein Reference Epstein1994; Reference Epstein, Millon and Lerner2003; Kahneman Reference Kahneman2011) and with UTT's first principle of two modes of thought. (We refer the reader to our analysis of this principle in González-Vallejo et al. Reference González-Vallejo, Lassiter, Bellezza and Lindberg2008.) In short, Hammond's notion of quasi-rationality is similar to modern conceptions of cognition. In particular, with the emergence of parallel processing models (e.g., Rumelhart et al. Reference Rumelhart and McClelland1986) and more generally connectionist models (Phaf & Wolters Reference Phaf and Wolters1997), psychologists favor the view that responses reflect a mixture of unconscious and conscious contributions.

We end by revisiting the prescription that complex decisions should be left to unconscious thinking. Many years of research converge on the conclusion that selecting important predictors is best done by experts, but the combination of cues is best left to a statistical tool (Bishop & Trout Reference Bishop and Trout2005; Dawes Reference Dawes1979). Imagine a psychiatrist judging the likelihood that a patient will commit suicide; the prescription that she or he should let the unconscious decide is not only wrong, it is also unethical.