Number processing outside awareness? Systematically testing sensitivities of direct and indirect measures of consciousness

Iris A. Zerweck, Chung-Shan Kao, Sascha Meyen, Catarina Amado, Martin von Eltz, Maren Klimm, Volker H. FranzAttention 2021 · DOI

TL;DR

This paper challenges the common interpretation of "unconscious processing" in masked priming studies. By applying a more rigorous sensitivity analysis, the authors found no evidence that indirect measures of awareness (like reaction times) are more sensitive than direct measures, suggesting that claims of unconscious processing may be overstated.

What Problem Does This Paper Address?

For decades, researchers in cognitive psychology have explored whether stimuli can be processed outside of conscious awareness. A common paradigm, masked priming, involves presenting a briefly visible stimulus (the "prime") that is then followed by a target stimulus. Researchers often observe two patterns: participants perform poorly on a direct task asking them to report on the prime itself (suggesting lack of awareness), but show significant "priming effects" in an indirect task (e.g., faster reaction times when the prime and target are related). The "standard reasoning" in this field interprets this pattern as evidence that the prime was processed unconsciously, and that this unconscious processing is even more effective than conscious processing. However, the authors argue this reasoning is flawed because it relies on comparing significance tests from different tasks rather than directly comparing their sensitivities. This can lead to spurious claims of unconscious processing, especially when direct tasks are "underpowered" (not sensitive enough to detect subtle awareness). This paper aims to re-evaluate the evidence for unconscious processing of numbers, a common stimulus in these studies, by employing a more robust methodology.

Methods

The core of this paper's contribution lies in its methodological critique and application of a "sensitivity analysis." Instead of relying on the "standard reasoning" (comparing significance of effects in direct vs. indirect tasks), the authors advocate for comparing the sensitivities of both tasks using the same metric.

Here's a breakdown of their approach:

  • Direct Task: Participants are asked to make a judgment about the masked prime stimulus (e.g., is it larger or smaller than 5?). Performance is measured by accuracy and, crucially, by the sensitivity index $d'$ from Signal Detection Theory.
  • Indirect Task: Participants perform a task on a "target" stimulus that follows the prime (e.g., is the target larger or smaller than 5?). Priming effects are typically measured by differences in reaction times (RTs) between congruent and incongruent prime-target trials.
  • Sensitivity Analysis: To compare the direct and indirect tasks on an equal footing, the authors transform the indirect task's RT data into a measure of sensitivity. They use a "median-split" technique: RTs shorter than a participant's median are classified as congruent, and longer RTs as incongruent. This allows them to calculate a classification accuracy and, importantly, a $d'$ value for the indirect task, analogous to the direct task.
  • Parametric Variations: To ensure they weren't missing an "indirect task advantage" (ITA) under specific conditions, the researchers systematically varied stimulus visibility in Experiments 2 and 3. They manipulated prime contrast and prime duration, respectively, to observe how sensitivities in both tasks changed across a wide range of perceptibility.

Datasets

The study did not introduce new datasets. Instead, it used stimuli and paradigms from established, highly influential studies on number priming, specifically referencing:

  • Dehaene et al. (1998)
  • Kouider & Dehaene (2009)
  • Naccache & Dehaene (2001a)

The authors also re-analyzed data from these original studies to compare their findings with their own results.

Key Insights and Ideas

  • The "Standard Reasoning" is Flawed: The paper's central argument is that the traditional method of inferring unconscious processing from a pattern of poor direct task performance and significant indirect task effects is unreliable. It can lead to false positives for unconscious processing.
  • Sensitivity is Key, Not Just Significance: A significant priming effect in the indirect task doesn't automatically mean high sensitivity. It could simply reflect a strong effect even with poor underlying discrimination ability. Comparing sensitivities ($d'$) directly is crucial.
  • No Evidence for Unconscious Number Processing: Across three experiments and re-analyses of prior work, the authors consistently found no evidence of an "indirect task advantage" (ITA) where the indirect task was more sensitive than the direct task.
  • Visibility Matters for Both Tasks: When stimulus visibility (contrast or duration) was low, sensitivities in both direct and indirect tasks were similarly low. As visibility increased, sensitivities improved in both tasks, often more so in the direct task. This suggests that similar processing mechanisms are at play, rather than distinct conscious and unconscious pathways.
  • Parametric Variation is Powerful: Systematically varying stimulus parameters (like contrast or duration) is essential for a comprehensive understanding of how direct and indirect measures behave across different levels of perceptibility. It helps avoid missing critical conditions where an ITA might (or might not) appear.

Key Results

  • Experiment 1 (Replication): Replicated findings of number priming. Direct task performance was close to chance ($d' \approx 0.23$), and indirect tasks showed significant priming effects. However, the sensitivity analysis revealed no significant difference in sensitivity between the direct and indirect tasks ($d'_{indirect} - d'_{direct} = 0.04$).
  • Experiment 2 (Prime Contrast): Varying prime contrast showed that as contrast increased, sensitivity in both tasks improved. Crucially, the direct task's sensitivity often exceeded the indirect task's, particularly at higher contrasts, indicating no ITA. At low contrasts, sensitivities were similarly low.
  • Experiment 3 (Prime Duration): Varying prime duration also showed improvements in sensitivity for both tasks with longer durations. Again, no ITA was found. In some conditions, the direct task's sensitivity was significantly higher than the indirect task's.
  • Re-analysis of Original Studies: When applying their sensitivity analysis to data from key original studies, the authors found that most did not support an ITA, contradicting their original conclusions.

Limitations and Future Work

  • Limitations of the "Standard Reasoning": The paper's primary limitation is the critique of the standard reasoning itself, highlighting how it can lead to misinterpretations. The authors acknowledge that their own direct tasks might not be perfectly "underpowered" but argue that their sensitivity analysis is robust regardless.
  • Stimulus Complexity: The authors speculate that ITAs might be more likely with simpler stimuli (e.g., Gabor patches) where processing might be more automatic and less subject to top-down control, compared to the more complex semantic processing of numbers.
  • Other Measures: While the paper focuses on behavioral (RT) measures, the authors briefly mention that neurophysiological measures like EEG (e.g., Lateralized Readiness Potential - LRP) have also been used to infer ITAs. They suggest that similar sensitivity analyses should be applied to these measures to avoid similar misinterpretations.
  • Future Research: The authors propose that future research should adopt their sensitivity analysis approach and utilize parametric variations of stimulus parameters to more rigorously test for unconscious processing. They also suggest exploring neurophysiological measures with this refined analytical framework.

Summary updated

Abstract

In priming research, it is often argued that humans can discriminate stimuli outside consciousness. For example, the semantic meaning of numbers can be processed even when the numbers are so strongly masked that participants are not aware of them. These claims are typically based on a certain pattern of results: Direct measures indicate no conscious awareness of the masked stimuli, while indirect measures show clear priming effects of the same stimuli on reaction times or neurophysiological measures. From this pattern, preserved (unconscious) processing in the indirect task is concluded. However, this widely used standard reasoning is problematic and leads to spurious claims of unconscious processing. Such problems can be avoided by comparing sensitivities of direct and indirect measures. Many studies are affected by these problems, such that a reassessment of the literature is needed. Here, we investigated whether numbers can be processed unconsciously. In three experiments, we replicated and extended well-established effects of number priming over a wide range of stimulus visibilities. We then compared the standard reasoning to a sensitivity analysis, where direct and indirect effects are compared using the same metric. Results show that the sensitivities of indirect measures did not exceed those of direct measures, thereby indicating no evidence for preserved unconscious processing when awareness of the stimuli is low. Instead, it seems that at low visibility there is residual processing that affects direct and indirect measures to a similar degree. This suggests that similar processing modes cause those effects in direct and indirect measures.
BibTeX
@inproceedings{zerweck2021number,
  author = {Iris A. Zerweck and Chung-Shan Kao and Sascha Meyen and Catarina Amado and Martin von Eltz and Maren Klimm and Volker H. Franz},
  title = {{Number processing outside awareness? Systematically testing sensitivities of direct and indirect measures of consciousness}},
  booktitle = {Attention},
  year = {2021},
  doi = {10.3758/s13414-021-02312-2},
}
Figure