Simulated Viewing Distance Impairs The Confdence–accuracy Relationship For Long, But Not Moderate Distances: Support For A Model Incorporating The Role Of Feature Ambiguity Part 2
Oct 13, 2023
Experiments 1a and 1b
Experiments 1a-3 were pre-registered at the Open Science Framework. These pre-registrations, Supplementary Materials, and all materials and data used for the analyses here are available at https://osf.io/7wdvy/.
There is a strong relationship between scientific frameworks and memory. A scientific framework is a way of organizing knowledge and theories together to help us better understand and remember new knowledge.
When learning new knowledge, we often can only remember some irrelevant details, but cannot integrate them into meaningful knowledge structures. At this time, the role of the scientific framework becomes apparent. By learning and understanding a scientific framework, we can connect new knowledge with existing knowledge to form an organic whole, thereby having a deeper understanding and memory of the new knowledge.
For example, when studying ecology, we can learn basic concepts and theories about ecological systems and apply them to different ecological scenarios to form a comprehensive and orderly knowledge framework. This framework can help us understand the complexity of ecosystems and better remember the basic concepts and principles of ecology.
Additionally, scientific frameworks can help us better understand and remember the results of scientific experiments. In scientific experiments, we often need to integrate a series of variables and control them to verify our hypotheses. By understanding and remembering the experimental process, we can better understand the meaning of the experimental data and apply it to a wider range of scientific fields.
In short, scientific frameworks not only help us better understand and remember new knowledge but also help us build deeper scientific understanding and ways of thinking. As long as we actively learn and apply scientific frameworks, we can achieve greater results in study and research. It can be seen that we need to improve our memory. Cistanche deserticola can significantly improve memory, because Cistanche deserticola can also regulate the balance of neurotransmitters, such as increasing the levels of acetylcholine and growth factors. These substances are very important for memory and learning. In addition, meat can also improve blood flow and promote oxygen delivery, which can ensure that the brain receives sufficient nutrients and energy, thus improving brain vitality and endurance.
Click know ways to improve brain function
Method
Participants
We determined the sample size based on our previous research (Davis et al., 2019), in which we found that a sample of 50 participants per condition was sufficient to examine differences between conditions in calibration curves. Participants for Experiment 1a and 1b were 102 students attending Skidmore College who participated for partial course credit (MAge=19 years), with 51 participants in each experiment.
Materials and procedure
The procedure was similar to that of Davis et al.’s (2019) Experiment 2. The stimuli were 60 male and female faces of different races (e.g., Caucasian, Hispanic, African American, and Asian) with a neutral close-mouthed expression from the Chicago Face Database (Ma et al., 2015). The distance was simulated (see Fig. 1 for example stimuli) by applying a Gaussian blur of 5 for Experiment 1a and a blur of 10 for Experiment 1b over the photographs using Sketch photograph editing software.
Gaussian blur is a method of photograph smoothing that creates an average color value for each pixel based on the other pixels surrounding it, weighting closer pixels more heavily. The level of blur refers to the number of surrounding pixels considered in the calculation; for a Gaussian blur of 5 (for Experiment 1a), each pixel would contain the Gaussian-weighted average of the five pixels surrounding it in each direction. In Experiment 1b, the blurred faces used a Gaussian blur factor of 10, which averages across the nearest 10 pixels and thus produces a blurrier image.
We estimate that the Gaussian blurs of 5 and 10 correspond to approximately 43 and 172 feet (see Loftus & Harley, 2005; but note that the Gaussian blur filter used here is slightly different from their low-pass filter).
Each participant encoded 30 faces, 15 of which were clear and 15 of which were blurred. We will refer to the clear condition as the near-distance condition and the blurred conditions as the medium- (Experiment 1a) or far- (Experiment 1b) distance conditions. The presentation order of the near and simulated distant faces was randomized separately for each participant, and the assignment of a given face to be studied or not and near or distant were counterbalanced across participants.
Before encoding, participants were told that they would see a series of faces, some blurred and some unblurred, and that their memory for these faces would be tested later. They then encoded each face for 1.5 s. Following the encoding phase, participants completed a series of word puzzles for 2 min and then received instructions for the recognition test. During the test, all 60 faces (30 previously encoded, and 30 new) were presented without blur, much like when a witness views faces in a lineup.
Faces were presented one at a time in a random order, and participants were first asked to indicate whether a given face was seen or not seen, and then to indicate their confidence in their response on a scale from 0 (a guess) to 100 (certain). Both the old/new recognition task and the confidence judgments were self-paced. After completing the test, participants answered a series of demographic questions and were debriefed.
Results
Experiment 1a
We first examine the impact of simulated viewing distance on memory performance overall (see Table 1 for a summary of hits, false alarms, and d-prime scores). D prime scores represent a standardized measure that deducts false alarms2 (incorrectly responding “Seen” to a new face) from hits (correctly responding “Seen” to an old face), with scores of zero indicating no discriminability between old and new faces and higher scores representing better levels of discriminability. Unsurprisingly, discriminability was much poorer for faces that had been encoded at a medium simulated distance (M=0.99) than faces that had been encoded at a near simulated distance (M=1.77), t(50)=6.14, p<0.001, d=0.86.
Of critical interest was the calibration between confidence and accuracy. Given the greater applied and theoretical importance of old relative to new responses, we only analyzed data for faces that participants identified as old (see Weber & Brewer, 2003 for a discussion of why old responses are more important for understanding the confidence–accuracy relationship in eyewitness memory).
We divided old responses into four confidence bins, whose size was based on our prior research (see Davis et al., 2019 for a discussion of this method), and plotted overall accuracy (e.g., the number of times a participant correctly responded to old divided by the total number of old responses) as a function of each of these four levels of confidence.3 As can be seen in Fig. 2, calibration was generally quite good for faces that were encoded under both the.
near- and medium-distance conditions. Four paired-sample t-tests confirmed that there was no significant difference between the two conditions at any of the four levels of confidence, ts<1.69. Thus, whereas memory was impaired overall when faces were encoded at a medium distance, there was no impact on the relationship between confidence and accuracy. This is largely consistent with the extant literature that has examined the impact of estimator variables on the confidence–accuracy relationship (e.g., Davis et al., Palmer et al., 2013).
Experiment 1b
Again, we frst examined memory accuracy between the clear and simulated distant faces using d prime as the dependent measure (see Table 1). As in Experiment 1a, discriminability was lower for faces that were presented at a far simulated distance (M=0.58) than when they were presented clearly to simulate a near distance (M=1.77), t(50)=11.42, p<0.001, d=1.60.
Once more, we were primarily interested in the relationship between confidence and accuracy, depicted in Fig. 3. Here, there were no differences between conditions at the first three levels of confidence, it's<1.41, but accuracy was significantly worse for faces encoded at a far simulated distance relative to faces that were encoded at the near simulated distance at the highest bin of confidence, t(35)=3.39, p=0.002, d=0.57.4
Discussion
Experiments 1a and 1b revealed a seemingly disparate pair of findings similar to Lockamyier et al. (2020) and Nyman et al. (2019). Whereas encoding at a simulated distance did impair memory overall in both experiments, the impact of simulated viewing distance on the confidence–accuracy relationship depended on the level of distance. In Experiment 1a confidence is predictive of accuracy across all four confidence bins. However, at a longer simulated distance, participants were less accurate for faces that had been encoded at a far simulated distance than the near distance.
This finding is particularly problematic for triers of fact in the criminal justice system, as high-confidence identifications are the most likely to serve as probative evidence at trial. Thus, if confidence is not a predictor of accuracy at high levels of confidence for faces viewed at a far distance, recommendations for best practices under these conditions would likely include not allowing such eyewitnesses to testify at all. Therefore, we thought it prudent to replicate this finding using a different sample in a single experiment (see Open Science Collaboration, 2015; for a discussion of the importance of reproducibility in psychological science). Specifically, we compared the difference between the near--, medium-, and far-distance conditions in a single experiment using workers on Amazon’s Mechanical Turk.
Experiment 2
Method
Participants, design, materials, and procedure
One hundred and five MTurk workers participated for monetary compensation (MAge=40 years), with 53 participants in near-distance conditions and 52 participants in the far-distance condition. The materials used here were identical to those used in Experiments 1a and 1b. However, we manipulated the degree of Gaussian blur (medium-distance and far-distance) between subjects in a single experiment. Thus, the design was a 2 (Face Distance: Near vs. Distant)×2 (Level of Distance: Medium vs. Far) mixed design, with Face distance,e manipulated within subjects and Level of Distance manipulated between subjects.
Results and discussion
We first examined memory performance overall with a 2 (Face Distance: Near vs. Distant) 2 (Level of Distance: Medium or Far) mixed ANOVA (see Table 1). There was a main effect of Face Distance, F(1, 103)=50.69, p<0.001, ηp 2=0.33, with lower discriminability for faces that were encoded at a simulated distance. Tere was also a nearly significant main effect on the Level of Distance, F(1, 103)=3.97, p=0.05, ηp 2=0.04, with lower discriminability in the Far Distance condition than the Medium Distance condition. Te interaction was not signifcant, F(1, 103)<0.001, p=0.99, ηp 2<0.001.Thus, memory performance was worse for distant faces than near faces and was worse still for the far distance relative to the medium distance.
As in Experiments 1a and 1b, we were primarily interested in the impact of the varying levels of distance on the relationship between confidence and accuracy (see Fig. 4). As specified in our pre-registration, we conducted four separate 2 (Face Distance: Near vs. Distant)×2 (Level of Distance: Medium or Far) mixed ANOVAs at each level of confidence. At the first three levels of confidence, there were no main effects or interactions, F’s<1.79. At the highest level of confdence, there was a main efect of Face Distance, F(1, 67)=21.39, p<0.001,ηp 2=0.24, as well as a main efect of Level of Distance, F(1, 67)=8.08, p=0.006, ηp 2=0.11. These main effects were qualified by significant interaction, F(1, 67)=4.90, p=0.03, ηp 2=0.07.
Follow-up tests at this highest level ofconfidencee (using a Bonferroni correction for multiple comparisons, critical alpha=0.013) revealed that in the medium-distance condition, the comparison between the near and distant faces approached but did not reach significance t(39)=2.41, p=0.02, d=0.38. In the far-distance condition, the comparison between near and distant faces was significant, t(28)=3.59, p=0.001, d=0.67, with lower accuracy for the faces that were encoded at a far distance (M=0.60) than for faces that were encoded at a near distance (M=0.85). No difference between the medium-distance and far-distance conditions for faces was encoded at a near distance, t(89)=1.20, p=0.23, d=0.25. However, accuracy was lower for faces that were encoded at a far distance (M=0.60) than faces that were encoded at a medium distance (M=0.82), t(70)=2.98, p=0.004, d=0.71.
Similar to Experiments 1a and 1b, it appears that there was little impact of simulated distance on the confidence–accuracy relationship at the lower level of distance. However, at the greater level of distance, participants were significantly more overconfident relative to clear faces and to faces viewed from a medium distance.
The goal of Experiment 3 was to attempt to implement an instructional warning (see Blank & Launay, 2014, for a meta-analysis on similar warning manipulations in the misinformation literature) designed to ameliorate the effect of encoding faces at a far simulated distance on the confidence–accuracy relationship.
For more information:1950477648nn@gmail.com
