The goal of the current experiment was to investigate the effect of increasing working memory load, and thereby reduced attentional capacity, on the attentional blink. To this end, we designed a web-based experiment in which we combined the AB RSVP stream with an orthogonal manipulation of working memory load via a visual n-back task. All experimental materials, data, and analysis code are openly available on GitHub (https://github.com/juliankeil/attentionalblink).
3.1. Sample Size Estimation
We hypothesize a modulation of the AB by working memory load. Thus, the critical effect in the current experiment consists of an interaction between the T1-T2 lag (three levels, 200 ms, 300 ms, 400 ms) and the memory load (three levels, 0-, 1-, 2-back) on the percentage of correct T2 identification, given that T1 was correctly identified (MacLean & Arnell, 2012). In this 3x3 repeated measures design, the required sample to obtain a medium effect of f2 = 0.25 at an alpha error probability of 0.05 and a power of 0.95 is n = 54 (Faul et al. 2007).
3.2. Participants
Of the 134 participants who started the demographic questionnaire (for details on the material, please see section 3.4.), 118 started the actual experiment, and 102 completed all parts. Of these, 5 participants were excluded due to being left-handed (possible confound in the response using the keys “a”, “s”, and “d”, see section 3.4.), and one was excluded to uncorrected hyperopia. Moreover, 5 participants never responded, 7 had too many errors during the RSVP streams (error rate above mean + 2 SD, M = 30.43, SD = 14.78), and 5 had too many errors during the n-back task (hit rate below mean + 2 SD, M = 58.34, SD = 22.96, correct-rejection rate below mean + 2 SD, M = 94.88, SD = 7.00). Finally, 19 participants had to be excluded, as the screen refresh rate was not set to approximately 60 Hz (see section 3.4.). This resulted in a final sample of N = 60, and an estimated power of 0.975 to detect a medium effect of f2 = 0.25, with a critical F value for the 3x3 repeated measures interaction of 2.45. The final sample consisted of 41 females and 19 males (no participant reported “diverse” or “N/A”), aged between 18 and 58 years (M = 25.5, SD = 9.79), of which one reported a doctorate as the highest level of education, 14 reported a university degree, 44 reported a high-school diploma (Abitur), and one a middle-school diploma (Realschule). Participants were recruited from the participant pool of the Institute of Psychology of the Christian-Albrechts-University Kiel and from social networks in return for partial course credit and the chance to win a 10€ voucher. All participants provided written informed consent to participate in the experiment, and the experiment was conducted in accordance with the 2008 declaration of Helsinki. The experimental protocol was evaluated and approved by the ethics committee of the Christian-Albrechts-University Kiel (ZEK-1/22).
3.3. Experimental Procedure
The current experiment was designed as a web-based study to be completed on the participants’ own computers, consisting of three parts. The first part was programmed in LimeSurvey (https://www.limesurvey.org/) and comprised the informed consent into the data collection including the information regarding data protection and contact information to the data security officer. After being informed about the experimental procedures, the participants themselves generated an individual code to anonymize the data. Before continuing to the demographic questionnaire, the participants had to indicate that they conducted the experiment at a computer with dedicated off-screen keyboard, as the experiment was not designed to be completed on a smartphone, tablet, or touch-screen display. The demographic questionnaire comprised questions on age (numerical value), sex (male, female, diverse, N/A), highest level of education (primary school, middle school, high school, university, doctorate, N/A), handedness (right, left, both), uncorrected hyperopia (yes, no), and information regarding recent head injuries, neurological disorders, or epilepsy (yes, no). Before continuing to the second part, the participants were shown two illustrations of the experimental design.
The second part of the study comprised the RSVP AB streams and the interleaved n-back task (please see section 3.4. for details). Participants reached the second part via a link provided at the end of the first part.
After completing the second part of the experiment, participants were automatically forwarded to another, independent, LimeSurvey questionnaire. The use of two separate questionnaires ensured the collection of independent datasets to avoid conclusions regarding the individual participants’ identities. Here, participants could provide their email addresses to obtain partial course credit or take part in the lottery to win a 10€ voucher. All email addresses were deleted after completion of the project and the final dataset does not contain any identifiable information.
3.4. N-Back Attentional Blink Experiment
The second part of the experiment represents the main part comprising the RSVP AB task and the interleaved n-back task (Figure 1A). This part was programmed in PsychoPy (Peirce et al. 2019) and hosted on Pavlovia.org for online participation (https://github.com/juliankeil/attentionalblink/01_experiment/). It comprised one practice block and 27 experimental blocks of 10 RSVP streams interleaved with the n-back task in the inter-trial interval (ITI). The screen background was set to neutral grey (RGB [128,128,128]) and all stimuli were presented in white in “Open Sans” type. Instructions were presented with a letter height of 0.05 relative to the individual screen height (except for the block number, which was presented in black letters, letter height = 0.075). A black fixation cross (letter height = 0.01) was displayed in the ITI with a random duration between 750ms and 1250ms. During the practice trials, feedback (letter height = 0.1) was provided for 1 s after each RSVP stream and n-back stimulus, comprising “Wrong” in red, “Correct” in green, and “You should have pressed a/s/d” in black (in the original experiment, all feedback was in German).
One RSVP stream comprised 6 stimuli, numbers served as distractors and uppercase letters as targets. RSVP stimuli were presented for 4 screen refresh frames and separated by 2 frames. Only datasets with a screen refresh rate of approximately 60 Hz (+/- 2 Hz) were included in the analysis, resulting 66.6 ms presentation time followed by a blank screen for 33.3 ms. Similar numbers and letters were excluded from the stimulus set (e.g., 1, I, J, 5, S, 8, B, Q, O, 0), and vowels were excluded to avoid the possibility to form syllables as a mnemonic strategy (Ferlazzo et al. 2007). Distractors (2,3,4,7,9) were presented randomly, but never twice in a row. Targets (C, D, E, F, H, L, M, P, R, Y) could occur anywhere in the RSVP stream except on the last position, as it has been shown that this leads to improved T2 detection (Giesbrecht and Di Lollo 1998; Vogel and Luck 2002). Thus, T2 could be presented at lags 2,3 and 4 (i.e., 200ms, 300ms, or 400ms after T1). Following the RSVP stream, participants had 1.5 s to respond (Figure 1B).
The participants’ task was to decide whether T1 and T2 were the same (by pressing “s”) or different (by pressing “d”), or whether only one target was presented (by pressing “a”). Previous studies showed a near perfect T1 detection (Chun and Potter 1995; Martens and Wyble 2010), and that the AB is due to the inability to correctly identify T2. Using a paired response allows testing the encoding and consolidation of T1 and T2, as the three-alternative forced choice task can only be correctly answered if T1 and T2 have been correctly identified (T2|T1). In case of identical T1 and T2 (condition “S”), the correct response was pressing “s” (s|S), in case of different T1 and T2 (condition “D”), the correct response was pressing “d” (d|D), and in case only T1 was presented (condition “A”), the correct response was pressing “a” (a|A). Thus, the AB can be seen in cases in which T2 was incorrectly identified (s|D, d|S), or not encoded (a|D, a|S), and it was operationalized as the percentage correct responses to both targets. All conditions (A, S, D) were presented 10 times, resulting in (3 (load) x 3 (condition) x 3 (lag) = 27 possible combinations) 270 trials overall, split into 27 blocks. The order of lags and conditions was pseudorandomized within a block, and the order of n-back load was randomized between participants to avoid order effects of working memory load.
The orthogonal n-back task presented in the ITI was aimed to increase the working memory load during each block. Before the start of each block, participants were informed about the condition, i.e., whether they should report a target stimulus (‘=’, 0-back) or a symbol repetition with respect to the last (1-back), or second-to-last stimulus (2-back) by pressing the space bar, and they had to confirm the condition by pressing the appropriate number to proceed. Within one block, 11 n-back symbols (@, <, %, !, &, ?, #, §) were presented, and each block started and ended with an n-back symbol. The symbol “=” was only used as a target in the 0-back condition to avoid confusion with the other n-back conditions. Each n-back stimulus was presented for 500ms (white, letter height = 0.1) and a target was drawn with 33% likelihood. After the n-back stimulus, participants had 1.5s to respond. The performance in the n-back task was operationalized as the sensitivity d’, i.e., the difference between the z-transformed hit rate and the z-transformed false alarm rate (Green and Swets 1966; Hautus 1995).
3.5. Hypotheses
In general, we hypothesize a modulation of the AB by working memory load. As mentioned above, the critical effect in the current experiment consists of an interaction between the T1-T2 lag (factor Lag) and the memory load (factor nBack) on the dependent variable percentage correct target identification (T2 | T1) (MacLean and Arnell 2012).
As a manipulation check, we first aim at testing the effect of the n-back task (0-, 1-, 2-back) and the different target combination conditions (A, S, D) on the dependent variable (DV) percentage correct responses. Here the null hypothesis (H01) states no difference in the DV depending on the n-back task or target condition. Accordingly, the first alternative hypothesis (H11) assumes a difference in the DV between the three target combinations (A, S, D, main effect for the factor Target). The second alternative hypothesis (H12) assumes a difference in the DV between the three n-back conditions (main effect for the factor nBack). More specifically, we assume that small increases in working memory load should improve the target identification (1-back > 0-back, Olivers and Nieuwenhuis 2006), and large increases in working memory load should impair the target identification (1-back > 2-back, Shapiro et al. 1994; Chun and Potter 1995; Wyble et al. 2009). Further hypotheses originate from significant H11 and H12: If we accept H11, we need to examine the effect of the factor nBack separately for the three target combinations. Specifically, the condition A, in which only T1 is presented, should capture the basic perceptual ability of the participants, as T1 is usually identified correctly, and sufficient attentional resources should be available. Accordingly, we assume no difference in the DV depending on the factor nBack (H13). Importantly, the factor Lag does not exist in the single target condition A. In contrast, for the two-target conditions (factor Target, S and D), we can examine the combined effects of the factors Lag (200ms, 300ms, 400ms) and nBack (0-, 1-, 2-back) in a three-way repeated-measures ANOVA, and we assume (H14) that the working memory load will have different effects at short versus long lags (MacLean and Arnell 2012).
Finally, we hypothesize that the effect of working memory load on the DV is not uniform across participants, but that those who are most affected by the n-back task should also show the strongest attentional blink (H15).
3.6. Statistical Analyses
All statistical analyses were performed in R (R Core Team 2011), and the analysis code is available at https://github.com/juliankeil/attentionalblink/03_analysis/.
The Mauchly test was used to verify the assumption of sphericity and the Greenhouse-Geisser correction was applied when necessary to correct for non-sphericity. For these cases, the corrected degrees of freedom and p-values are reported.
The first two hypotheses (H11 and H12) were tested in a repeated-measures ANOVA with the factors nBack (3 levels: 0-, 1-, 2-back) and Target (3 levels: A, S, D). The third hypothesis (H13) assumes no difference in the DV depending on the factor nBack for the single target condition (A) and was tested with an equivalence test for paired samples with a smallest raw effect size of interest of +/-7%, i.e., approximately half of the difference between paired and unpaired target reports across experiments in (Ferlazzo et al. 2007). The fourth hypothesis (H14) was tested in a repeated-measures ANOVA with the factors Lag (3 levels: 200ms, 300ms, 400ms), nBack (3 levels: 0-, 1-, 2-back) and Target (2 levels: S, D). For the fifth hypothesis, the relationship between perceptual sensitivity (d’) in the n-back task and the percentage correct responses in the RSVP stream across target conditions was examined using a linear mixed effects model. In the computation of perceptual sensitivity (d’), extreme values were corrected (Hautus 1995).
The alpha level was set to 0.05 in all tests and corrected for the number of comparisons in all post-hoc paired t-tests using a Tukey correction.