Maintaining fixation does not increase demands on working memory relative to free viewing

Participants

Twenty healthy young adults between the ages of 18 and 30 years old (13 females, mean age = 24.5 years, mean education = 17.3 years) were recruited through the Baycrest participant pool. Participants were ensured anonymity, and were informed of any risks of the experiment, as well as their right to withdraw at any time without consequence. The study was approved by the Baycrest Research Ethics Board (Approval Number: 02-01), and all the participants provided written informed consent before any experiments were completed. Exclusion criteria were history of neurological disease, active significant medical illness, serious psychiatric illness, or substance abuse during the last 12 months.

Participants reported their handedness as part of the screening questionnaire. Of our 20 participants, 17 were right-handed, two were left-handed, and one was ambidextrous. The finger tapping pattern was always performed with the right hand (described in more detail below), meaning that some participants performed the task with their dominant hand and others with their non-dominant hand. Right-handed finger tapping shows increased interference with verbal cognition in right-handed individuals (Kinsbourne & Hiscock, 1978). For this reason, handedness was included as a factor in our analysis of n-back performance across the motor conditions; it did not impact the results.

Procedure

Participants completed an auditory n-back task while simultaneously performing three dual task conditions. Participants heard spoken letters presented to them one after the other via earphones. Participants responded to the appropriate target letter by verbalizing the statement “yes” into a microphone. For the 0-back task, participants responded every time they heard the target letter “X”. For the 1-back task, participants responded when they heard a letter repeated one after the other (for example, “B-B”). For the 2-back task, participants responded for every letter that was the same as the letter presented two letters before (for example, “B-F-B”). During each of these tasks, participants were also presented with lure items, for which they were required to withhold a response. For a schematic outlining the n-back task, please see Fig. 1.

Participants performed the auditory n-back task while viewing the screen according to three different dual task conditions. The two oculomotor conditions were (1) free viewing and (2) fixed viewing. In the free viewing condition, participants were instructed to move their eyes freely anywhere on the screen while completing the n-back task. In the fixed viewing condition, participants were instructed to fixate on a central cross at all times while responding to the n-back task. If their eyes strayed outside of a 400x400 pixel window surrounding the fixation cross, the cross would flash red to communicate that they should move their eyes back to the cross as quickly as possible. The fixed viewing condition was conducted in this manner after participants made considerable viewing errors during pilot testing with a smaller fixation window, resulting in increased distraction caused by the error feedback. To be sure that any effects observed during the fixed viewing condition were due to limited eye movements and not to processing of the error signal, we enlarged the viewing window slightly, so that eye movements would still be constrained relative to free viewing. For the finger tapping condition, participants could examine the screen freely, but were instructed to continuously and steadily execute a finger tapping pattern with their right hand on the keyboard in front of them. The pattern was ‘1’-‘3’-‘2’-‘enter’ (i.e., index-ring-middle-pinky) typed repeatedly on the number pad (Moscovitch, 1994). Participants practiced this pattern to a point of proficiency during practice trials, but there was no penalty for out-of-sequence taps during the test trials.

Participants completed three practice trials for each n-back load (0-, 1-, and 2-back) and viewing condition (free and fixed viewing, and finger tapping) to become familiar with the tasks, followed by the dual-task viewing/n-back task in three blocks such that all combinations of n-back (0- 1- and 2-back) and viewing (free, fixed, finger tapping) were equally represented across blocks. The order in which these viewing/n-back combinations were presented was counterbalanced across blocks. Each viewing/n-back trial was equivalent in terms of the number of letters presented—64 letters (including 16 target letters). The jittered interstimulus interval (i.e., the amount of time between the presentation of each letter) was 800-975 ms. This was also the amount of time participants had to make a response. Vocal “yes” responses were recorded using a Blue SnowBall iCE USB Microphone.

Equipment used during data acquisition

The eye tracking equipment used in this experiment was the same as that described in Ryan, Riggs & McQuiggan (2010).

Data analysis

From the eye movement record, we derived measures of the number of fixations made to the computer screen for a given viewing/n-back trial, and the average saccade amplitude (ASA; the mean distance in degrees of visual angle of a participant’s saccade). To measure overall accuracy on the task, we computed d’ scores for each participant. These d’ scores were calculated as the standardized proportion of hits out of targets minus the standardized proportion of false alarms out of lure items (d′ = z (% hits) –z (% false alarms)). We also measured reaction times (RTs) to initiate an audible “yes” vocalization in registering these ‘hit’ responses. For each participant, the median RT on each condition was retained for further analysis.

We inspected the data from our eye movement and behavioural measures for outliers, defined as a data point greater than 2.5 standard deviations above or below the mean on a given condition. Two participants were outliers on multiple measures. One such participant was >2.5 standard deviations above the mean for RT on the free viewing 0-back condition, and >2.5 standard deviations below the mean for d’ on the free viewing and finger tapping 2-back conditions. The other multi-measure outlier was >2.5 standard deviations above the mean for number of fixations on the finger tapping 0-back condition, as well as ASA on each of the fixed viewing 0-, 1-, and 2-back conditions. These two multi-measure outliers were excluded from subsequent analyses. Of the remaining data, there were two participants who were outliers on RT—one on the fixed viewing 0-back and one on the finger tapping 2-back condition. Despite the above outliers, a test of skewness for each measure on each condition demonstrated that none of the distributions were significantly skewed, including the eight distributions with outliers. For this reason, the single measure outliers were retained and not adjusted for further analyses. To be sure that the retained outliers were not affecting the results, each analysis was performed both with and without these two single measure RT outliers. The retained outliers did not influence the pattern of results for any of the measures of interest.

For our data analysis, we ran both traditional frequentist and Bayesian models for each measure. First, we compared performance on the two eye movement measures (number of fixations and ASA) and two n-back measures (RT and d’) across the various n-back memory loads and viewing conditions, by running four 3x3 frequentist repeated measures ANOVAs. For each ANOVA, the three memory loads (0- versus 1- versus 2-back) and three viewing conditions (free viewing versus fixed viewing versus finger tapping) were included as within-subjects factors, with either number of fixations, ASA, RT, or d’ as the dependent variable. To determine whether participants performed the viewing task in adherence with instructions, we ran planned orthogonal contrasts of fixed viewing with each of the conditions allowing free examination of the screen—free viewing and finger tapping. For the n-back measures, we again ran linear planned contrasts for memory load, in this case as a manipulation check to see if participants would show the expected increase in RT and decrease in d’ scores with successive memory load conditions. For the eye movement measures, we ran linear planned contrasts to see if the number of fixations and ASA would decrease with increasing memory load, which could represent an eye movement response to increasing task difficulty (Meghanathan, Leeuwen & Nikolaev, 2015; Schut et al., 2017). In order to determine whether the motor interference effect of the secondary task was specific to finger tapping, we conducted hypothesis-driven planned orthogonal contrasts (free viewing versus finger tapping; fixed viewing versus finger tapping). For all four ANOVAs, significant interactions were probed using post hoc tests of simple effects, and effect size was measured as partial eta squared.

We complemented our frequentist analysis with four Bayesian repeated measures ANOVAs (Nathoo & Masson, 2016), composed of the same factors. These Bayesian models were run to obtain a Bayes factor for each measure—that is, a likelihood ratio providing a continuous measure of the observed data occurring under one hypothesis relative to another (Kass & Raftery, 1995). In terms of notation, BF₁₀ represents the Bayes factor for the alternative over the null hypothesis, whereas BF₀₁ gives the likelihood of the null over the alternative hypothesis. The Bayesian approach is especially useful for investigating null effects, which cannot be legitimately argued for within a frequentist framework (Wagenmakers, 2007). In this paper, we report both the frequentist and Bayesian results, as recommended by Dienes & Mclatchie (2018).

Eye movements

To assess adherence to the viewing task instructions, we compared two eye movement measures—number of fixations and ASA—across the three viewing conditions and n-back loads. There was a main effect of viewing condition (F(2, 34) = 15.03, p < .001, ${\eta }_{\mathrm{p}}^2=.469$, Bayes factor BF₁₀ = 2.174¹⁰) on the number of fixations executed. As expected, participants made more fixations during the two conditions involving free examination of the screen—free viewing (M = 162.96, SE = 13.04) and finger tapping (M = 143.22, SE = 9.09)—relative to the fixed viewing (M = 113.13, SE = 11.94) condition (vs. free viewing: F(1, 17) = 22.23, p < .001, ${\eta }_{\mathrm{p}}^2=.567$, BF₁₀ = 6.374⁶; vs. finger tapping: F(1, 17) = 11.68, p = .003, ${\eta }_{\mathrm{p}}^2=.407$, BF₁₀ = 5, 140.16). In visualizing the fixation data as heat maps, we could see that not only did participants make more fixations during free viewing and finger tapping, they also explored a larger area of the screen as compared to staying more central during fixed viewing (see Fig. 2). There was also a main effect of memory load (F(2, 34) = 9.92, p < .001, ${\eta }_{\mathrm{p}}^2=.368$, BF₁₀ = 7.155) on the number of fixations, with a decreasing linear trend (F(1, 17) = 10.92, p = .004, ${\eta }_{\mathrm{p}}^2=.391$) across the 0-back (M = 144.56, SE = 10.90), 1-back (M = 146.52, SE = 10.36), and 2-back (M = 128.24, SE = 10.27) conditions. These main effects were qualified by a significant interaction of viewing condition with n-back load on fixation count (F(4, 68) = 3.52, p = .011, ${\eta }_{\mathrm{p}}^2=.171$, BF₁₀ = 1.360; see Fig. 3A). Post hoc tests of simple effects revealed that whereas participants did show a linear decrease in fixations with successive memory loads for free viewing (F(1, 17) = 14.21, p = .002, ${\eta }_{\mathrm{p}}^2=.455$), they did not demonstrate a significant linear effect of load for the fixed viewing (F(1, 17) = 4.16, p = .057) or finger tapping (F(1, 17) = 0.19, p = .667) conditions.

As for ASA, we observed the expected main effect of viewing condition on saccade amplitude (F(2, 34) = 42.83, p < .001, ${\eta }_{\mathrm{p}}^2=.716$, BF₁₀ = ∞). In keeping with the task instructions, participants executed larger amplitude saccades during the conditions allowing free examination of the screen—free viewing (M = 3.78, SE = 0.32) and finger tapping (M = 3.60, SE = 0.38) –than during the fixed viewing (M = 1.36, SE = 0.12) condition (vs. free viewing: F(1, 17) = 66.86, p < .001, ${\eta }_{\mathrm{p}}^2=.797$, BF₁₀ = 1.311¹⁴; vs. finger tapping: F(1, 17) = 39.32, p < .001, ${\eta }_{\mathrm{p}}^2=.698$, BF₁₀ = 5.018⁹). There was no main effect of memory load on ASA (F(2, 34) = 0.45, p = .643, BF₁₀ = 0.064), nor was there an interaction of viewing condition with n-back load (F(4, 68) = 0.30, p = .878, BF₁₀ = 0.020; see Fig. 3B).

N-back task

Participants differed in their RTs across the three memory loads. As expected, the main effect of memory load on response times was significant (F(2, 28) = 10.18, p < .001, ${\eta }_{\mathrm{p}}^2=.421$, BF₁₀ = 645.720), with a linear increase (F(1, 14) = 13.04, p = .003, ${\eta }_{\mathrm{p}}^2=.482$) in RT from the 0-back (M = 808.35, SE = 16.43) to 1-back (M = 866.05, SE = 19.30) to 2-back (M = 866.97, SE = 23.46) task across viewing conditions. There was no main effect of viewing condition on RT (F(2, 28) = 1.31, p = .286, BF₁₀ = 0.192), nor was there a significant interaction between viewing condition and memory load (F(4, 56) = 1.20, p = .323, BF₁₀ = 0.169; see Fig. 4A).

Considering task accuracy, the expected main effect of working memory load on d’ scores was observed (F(2, 32) = 33.99, p < .001, ${\eta }_{\mathrm{p}}^2=.680$, BF₁₀ = 1.982⁹). When collapsed across the viewing conditions, accuracy decreased with increasing memory load, with d’ scores becoming progressively lower from the 0-back (M = 3.02, SE = 0.10) to 1-back (M = 2.69, SE = 0.13) to 2-back (M = 2.08, SE = 0.13) condition (F(1, 16) = 60.60, p < .001, ${\eta }_{\mathrm{p}}^2=.791$). There was also a significant main effect of viewing condition (F(2, 32) = 6.89, p = .003, ${\eta }_{\mathrm{p}}^2=.301$, BF₁₀ = 40.970) on task accuracy. Averaged across memory loads, participants had significantly lower d’ scores during finger tapping (M = 2.35, SE = 0.11) than either free (M = 2.83, SE = 0.10) or fixed viewing (M = 2.62, SE = 0.16) (vs. free viewing: F(1, 16) = 15.34, p = .001, ${\eta }_{\mathrm{p}}^2=.489$, BF₁₀ = 83.855; vs. fixed viewing: F(1, 16) = 5.07, p = .039, ${\eta }_{\mathrm{p}}^2=.241$, BF₁₀ = 1.145). While we were unable to test the free-fixed viewing comparison using a planned orthogonal contrast, we could run a post hoc comparison as part of our complementary Bayesian ANOVA to probe this effect. This Bayesian comparison yielded a Bayes factor of BF₁₀ = 0.762, meaning that the observed data were less likely under the alternative than the null hypothesis, and thus suggesting that there was no difference in task accuracy between the free and fixed viewing conditions. There was no significant interaction of viewing condition and memory load on d’ scores (F(4, 64) = 1.01, p = .408, BF₁₀ = 0.630; see Fig. 4B).