4.3.1.1 Introduction
In order to further explore whether the moderately reduced trustworthiness facial judgment accuracies of Experiments 1 and 4 under high perceptual could be accounted for by some facet of working memory we turn to the next experiment.
Explicitly, we hypothesized that in contrast to the already observed effects of perceptual load on trustworthiness facial judgments; such judgments under working memory load will not significantly differ between the low and high working memory load conditions.
4.3.1.2 Method
4.3.1.2.1 Participants
11 new participants were included in Experiment 7, (5 males) (Mean age = 22.54, SD 3.24 (10 females) (range 18 – 27) (2 outliers removed). All of the participants in this experiment, had normal or corrected-to-normal vision, as in the prior experiment, any subject with less than 50 % accuracy on the Low Load face or search tasks was classified as an outlier and removed from the subsequent analyses, participants participated for course credit or were paid £3 pounds.
4.3.1.2.2 Stimuli and procedure
The apparatus and stimuli generation were identical to Experiments 4, 5 and 6, however as this was an experiment on working memory there were some modifications.
The flow of the experiment is illustrated below in Figure 4-5. Participants were provided with a written explanation and told that their task was to mentally verbally rehearse each number presented on each trial. As in the experiment in the last Chapter, each trial in the visual search task began with key press that was self-initiated. This was followed by a fixation point presented for 500 ms in the centre of the screen, followed immediately by a presentation of a single number in the low load condition (of 500 ms duration) and a six number string for the high load (of 1500 duration); the target numbers were equally likely to appear in any of the possible combinations of positions. Following this number, a fixation cross would appear on the screen; followed by a face. The face, centred at fixation was
displayed with a 150 ms presentation time. The participants’ first response would be to indicate whether the face shown (within a response window of 2000 ms) was trustworthy or untrustworthy (for Experiment 7 the same procedure was employed but the participants were asked to indicate dominant or not dominant and for Experiment 9, threatening or not threatening).
The presentation of the face was then followed by a mask of 500ms. To maintain comparability with the prior experiments in Chapter 3, the number keypad values 1 and 2 were used for social facial evaluations. Following the response to the face a “probe” number would appear (with a total delay duration between face onset of 3000ms) – whereupon the participant was instructed to indicate, using the keyboard whether the probe number that they were mentally verbally rehearsing appeared at the beginning of the trial: indicating “X” for present or “Z” for absent, (within a response window of 4000 ms). As aligned with the prior experiments, there was a total of 4 blocks, 2 high and 2 low (randomized across participants). Each block consisted of 24 trials, (12 trustworthy and 12 untrustworthy images, presented in random order). Each face was presented once during the experiment and the assignment of positive valence faces to negative or positive blocks was randomized and counterbalanced across participants and across blocks. Before starting the experiment, participants completed one practice block of 12 example trials to familiarise themselves with the keys.
Figure 4-5 Example screen display of Working Memory load manipulation (high load condition illustrated) The flow of the experiment is illustrated above for the “6 numerals” or “high load” condition in which participants endeavoured to remember a six digit number. As in the prior experiments, there were total of 4 blocks, 2 blocks of high load and two of low load; 96 Trials in total – 24 unique identities 4 levels of emotion/valence. In the high load a number sting is displayed e.g. 126353 as shown above, at probe, e.g. the number 2 as above, a correct response would be for the participant to say that this number was indeed present. An instance of a low load would be a single number between 1 and 9
4.3.1.2.3 Design
The design employed here was almost identical to that employed in the prior experiments (Experiments 1-6), only slightly differing on the dependent variables (and restated here for clarity).
Mean percentage search accuracy rates and mean reaction times for the working memory response were the dependent variables, and were entered separately into a 2 x 4 repeated measures ANOVA. With perceptual load (Low, High) and Valence (-3, -2, +2, +3 - SD faces) as within subject independent variables (IVs).
Note, the mean reaction times for the working memory response is a relatively uncontaminated measure, as it does not measure from the time of the memory number stimuli which would then include processing of the face viewing, but measures instead probe response time minus onset of memory probe. This measurement should give some indication of the comparative difficulty between the levels of working memory load.
Next mean percentage trustworthiness accuracy rates were the dependent variable and was entered into the same 2 x 4 repeated measures ANOVA., again with perceptual load (Low, High) and Valence (-3, -2, +2, +3 - SD faces) as within IV’s, excluding trials in which the search response was incorrect.
This same design was employed for the following two experiments, Experiment 8, for dominance faces and Experiment 9 for threatening faces.
4.3.1.3 Results and Discussion
Memory task The mean correct memory-probe identification rates were significantly better for trustworthiness in the low working memory load condition (M = 97 %) than in the high working load condition (M =91 %), F(1, 10) = 18.80, MSE = .005, p = .001, η2 = .65. There was no significant effect of valence F(3, 30) = 1.55, MSE =.005, p = .222, η2 = .13. Additionally there was also no significant interaction between valence and load (F<1), F(3, 30) = .89, MSE =.006, p = .456, η2 = .082.
Mean reaction times in the low working load condition were significantly faster (M = 770) than in the high working load condition (M =1126), F(1, 10) = 43.27, MSE = 64525.32, p <0.001, η2 = .81. There was no significant effect of valence (F<1),F(3, 30) = .87, MSE = 11215.86, p = .468, η2 = .080. Furthermore, there was no significant interaction between valence and load for mean reaction times (F<1), F(3, 30) = .49, MSE =18377.37 p = .694, η2 = .046.
These results endorse the claim that the working load condition manipulation was indeed more demanding in the low load condition as in contrast to the high load condition for trustworthiness judgments (see Table 4-4).
Perceptual load
Low High Differential Effect of load
Search Accuracy (%) 97(3) 91(5) 6*
Reaction Time (ms) 770(268) 1126 (336) 356*
Face Accuracy (%) 87(7) 85(5) 2 NS
SDs are listed in parenthesis. * = significant NS = not significant
Table 4-4 Mean percentage search accuracy rates, mean reaction time (ms) rates and mean face accuracy rates in the working memory trustworthy task as a function of load in Experiment 7
Face classification. Mean percentage face accuracy trustworthiness rates in the low working memory load condition were not significantly better (M = 87 %) than in the high working memory load condition (M =85 %), F(1, 10) = 1.69, MSE = .006, p = .222, η2 = .145. There was however both a significant effect of valence F(3, 30) =5.79, MSE = .026, p <.005, η2 = .367, and a significant interaction F(3, 30) = 4.073, MSE =.004, p = .015, η2 =.289.
Figure 4-6 Trustworthiness mean percentage face accuracy rates in the working memory task (valence ranges from high untrustworthy valence -3SD on the left to high trustworthy +3SD on the right)
This pattern of results somewhat resembles those seen for trustworthiness judgments in Experiment 4, with higher accuracies for untrustworthy faces (+3SD valence). Furthermore, low load accuracies are numerically higher for all valences, as seen in the prior trustworthiness experiments (Experiments 1 and 4), except for the maximum trustworthy faces where this order is reversed. As evidenced by the main effect of valence, the mid-range faces, particularly +2SD valence, are more challenging to discriminate than the extremes of trustworthiness. To investigate the significant interaction, follow up uncorrected pairwise comparisons revealed that none of the contrasts between
low load and high load accuracies were significantly different, except for -2SD valence (low load M = 89.5 % vs. high load M= 84.5 % , t(10) =2.38 p = .04).
The results of Experiment 7 broadly support our prediction that the level of working memory load in a comparable task to that of perceptual load (seen in Experiments 1-6) would not impact facial judgment accuracies, and thus that the already observed effects are more likely to be specific to perceptual load than an influence upon working memory. Although follow up testing revealed a small effect for -2SD valence, it is possible that the smaller sample size employed here contributes to this. Reassuringly the more distinct valence faces -3SD and +3SD valence were not impacted by load.
The last two trustworthiness experiments (Experiments 4 and 7) both have similar results profiles (see Figure 4-2 and Figure 4-6) and of particular note, have reduced accuracies in the mid- range valences (-2SD valence and particularly +2SD valence). These valences may be more ambiguous for participants to evaluate and thus may potentially obfuscate inferences on the role of attention in trustworthiness facial judgments. Future experiments may be better served by using the maximum valence -3SD and +3SD (an approach which we will adopt in Experiment 10).