MethodsDesign

This study aims to investigate whether (i) participating in a live learning session improves learning online compared to recorded videos of either a previous interaction or of a teacher alone and (ii) whether these conditions impact learning differently in autistic adults compared to neurotypicals. To answer these questions, this study adopted a 2 (group) × 3 (learning condition) × 2 (time) repeated-measures design, with between- and within-subjects factors. The between-subjects factor is group (autistic vs neurotypical (NT)), the within-subjects factors are (i) learning condition (live vs recorded of another social learning episode vs recorded of the teacher alone), and (ii) time of recall (immediate vs delay quiz). Specifically, facts about 15 items were presented with two minutes per item. Five items were assigned to each condition: (1) live condition: the participant participated in a live video-call when they learned in interaction with the teacher; (2) recorded-observant condition: the participant was shown a pre-recorded video of the teacher presenting the learning material to a student (confederate); (3) recorded alone condition: the participant was shown a pre-recorded video of the teacher alone presenting the learning material (Fig. 1). The learning score (outcome measure) for each participant was obtained from a multiple-choice quiz (see Materials). Items assigned to each condition and trial order within each condition remained fixed for the whole duration of this experiment. The order of conditions was randomised across participants.

Fig. 1

Schematics of the three experimental conditions

Materials

A selection of 15 items was selected from De Felice et al. [17], three from the exotic food category (Rambutan, Kiwano, Cherimoya), four from the antique category (Strigil, Porte-joupe, Scotch Hands, Chatelaine), four from the animal category (Tarsier, Axolotl, Glaucus, Anhinga) and four from the musical instrument category (Kalimba, Caxixi, Agogo, Hulusi). Each condition presented a mixture of objects from these categories, which remained fixed for all participants for this experiment (but was counterbalanced in experiment 2, see Sect. 3), as follows: (1) live condition [Tarsier, Kalimba, Strigil, Axolotl, Rambutan]; (2) recorded-observant condition [Porte-joupe, Kiwano, Caxixi, Scotch Hands, Glaucus]; (3) recorded-alone condition [Agogo, Cherimoya, Anhinga, Chatelaine, Hulusi]. Learning was tested via the same multiple-choice quiz used by De Felice et al. [17]. Full details of item information and multiple-choice quiz are reported in Additional file 1: Appendix Table 1.

Learning sessions are represented as appearing to participants. From left to right: In the live condition, participants learn about five items as they interact with the teacher in a real-time video-call; Recorded-observant condition: participants learn about five items from a pre-recorded sessions with a confederate acting as a previous participant; Recorded-Alone condition: participants learn about five items from a pre-recorded session of the teacher alone. In each condition, participants learn about five different items. Items were assigned to each condition randomly and remained fixed within each experiment, and counterbalanced between experiments.

Procedure

Participant recruitment This study was approved by the UCL ethics committee. Participants were recruited via the online platform Prolific (www.prolific.co). The platform retains demographic details as well as information on any disabilities/diagnoses of users, as reported by the users at the time of account registration. Such anonymous information can be used to create adverts targeting a specific pool. Two separate adverts were published: one targeting neurotypical participants and one targeting autistic participants. As a further check, users who responded to our adverts were asked to confirm their diagnosis via a questionnaire on Gorilla Experiment Builder (www.gorilla.sc). To ensure that the experimenter was blind to the participant’s diagnosis, recruitment was done by a researcher who was not involved in data collection.

To be eligible, all participants had to (i) be fluent in English (speaking English regularly for > 5 years); (ii) be aged 18–65; (iii) give consent to having their camera and microphone on; and (iv) give consent to being recorded for the whole duration of the experiment. Participants were paid at the hourly rate of £7.50 for a total of £15 over two hours. An additional £3 was offered for those who completed a 10 min quiz a week later.

Participants who responded to our advert were asked to complete four main tasks: (1) background battery (independently online, on Gorilla Experiment Builder), (2) learning session (over a video-call), (3) online learning multiple-choice quiz immediately after the learning session (independently online on Gorilla Experiment Builder), and (4) repeat the quiz a week later.

Background battery Users who responded to the Prolific adverts were redirected to Gorilla Experiment Builder (www.gorilla.sc), where they received instructions on the study and gave consent for participation. They then completed the Background Battery tasks. This comprises of (i) Spot-the-word test, a measure of verbal fluency [3], (ii) matrix reasoning item bank (MaRs-IB), a measure of non-verbal reasoning [11] and (iii) Animated Triangle test, a measure of mentalising [2, 61].

Upon completion of the Background battery task, an independent researcher sent the participant ID to the experimenter (teacher), who arranged a video-call with the participant (via Prolific chat) while remaining blind to their diagnosis.

Video-call The experimenter greeted the participant and checked that audio and video worked adequately. The participant was asked to open the zoom window in full-screen mode and chose the gallery view (i.e. everyone in the call is shown equal size, this ensured that view during the live session was comparable to the view during pre-recorded video watching). Participants were told that the aim of the study was to investigate how people learned online and whether this differed in autistic people. They were asked not to disclose personal information to the teacher, who was blind to their diagnosis. The experimenter then explained that the participant would learn some facts about exotic food, animals, antiques and rare musical instruments over three formats: in live interaction with the teacher (live condition) and through watching pre-recorded videos showing either the teacher with a previous participant (recorded-observant condition) or the teacher alone (recorded-alone condition). They were instructed to memorise as much information as possible, as at the end of the video-call, they would complete a multiple-choice quiz to test their learning. During the live condition, participants were told they were free to ask questions and interact with the teacher. Before starting the learning sessions, the participant’s pre-knowledge was tested. If any item was known, this was excluded from the analysis (but not from the learning session). Learning sessions started with either the live, recorded-alone or recorded-observant condition in a counterbalanced and semi-randomised order. The call lasted approximately 40 min (i.e., 10 min per condition, with 2 min per item and five items in each condition, plus 10 min for instructions).

Learning quiz Immediately after the learning session, participants were redirected to Prolific, where their IDs were included in a ‘white list’ so that a new advert was visible to them only. By replying to that advert, participants were redirected to Gorilla Experiment Builder (www.gorilla.sc), where they reported on the quality of the video call (audio and video) before completing the learning quiz. After the learning quiz, they also completed an ‘enjoyment questionnaire’ and the inclusion-of-other-in-the-self questionnaire (Aron et al., 1992). This part lasted approximately 20 min and was completed by the participant independently (note that the ‘immediateness’ of the quiz was ensured by the experimenter, who terminated the video-call only a few moments after the participant initiated the quiz part on Gorilla Experimenter Builder).

Exactly one week after the learning sessions, participants were invited through Prolific to the final stage of the experiment and directed to Gorilla Experimenter Builder to complete the same learning quiz. Additionally, participants filled in a history questionnaire to check for potential revision of any of the items (e.g. search on Google). This part lasted approximately 10 min.

ResultsSample

53 participants took part in the study (Table 1). Participants were excluded when reporting 3 or less on a 1 (poor) to 5 (excellent) video-call quality scale (N = 3), and being visibly distracted during the video-call (N = 1). Of the remaining 49 participants, 46 (Nautistic = 20, male = 11, female = 4, non-binary = 4; NNT = 26, male = 11, female = 15) completed the whole experiment, including the one-week delay quiz (see 2.1 Design and 2.3 Procedure). We lost demographic data from one participant (autistic group) due to a technical fault.

Table 1 Demographics for the dataset from Experiment 1, 2 and combined

Autistic participants either received a diagnosis by a clinician (N = 18) or were self-diagnosed (N = 2). Autistic and NT groups did not differ on age (meanautistic (sd) = 27.79 (9.22), meanNT (sd) = 29.85 (9.90), t(43) = 0.71, p = 0.48), verbal fluency (Spot the word test, meanautistic (sd) = 47.63 (6.71), meanNT (sd) = 44.73 (6.23), t(43) = − 1.48, p = 0.15) non-verbal reasoning (MaRs-IB, meanautistic (sd) = 64.48 (18.57), meanNT (sd) = 61.64 (16.98), t(43) = − 0.52, p = 0.60) and mentalising test (Animated Triangle, meanautistic (sd) = 9 (2.54), meanNT (sd) = 9.33 (1.92), t(43) = 0.38, p = 0.71). The autistic group scored significantly higher on AQ than the NT group (meanautistic (sd) = 33.37 (6.73), meanNT (sd) = 19.19 (7.29); t(43) = − 6.73, p < 0.0001).

Data pre-processing

Single trials were excluded when: (i) participants reported that they knew the item; (ii) internet connection dropped during the single trial but was good for the rest of the experiment; (iii) the experimenter reported incorrect information about the item; (iv) participants reported revising information about a given item before the delay quiz (excluded from delay performance only). Performance was calculated for each learning condition separately, as an average over the included trials (score = points collected on all trials / total points available on all included trials).

Analysis of variance

An analysis of variance (ANOVA) was run to test the difference in learning performance between 2 (groups: Autistic and NT) × 3 (learning conditions: live, recorded-alone and recorded-observant) × 2 (time of learning quiz: immediate and delay). Means and SD for all conditions are reported in Table 2. Results for main and interaction effects are reported in Table 3 and Fig. 2.

Table 2 Sample size (N), Means and SDs for all conditions for Experiment 1, 2 and combinedTable 3 Results for analysis of variance for Experiment 1, 2 and combined datasetsFig. 2

Results for Experiment 1. Results from the dataset of Experiment 1. *p < 0.05; **p < 0.001;***p < 0.0001. A Boxplots of the three main factors of interest: Time, Learning condition and Group. B Violin plots of learning performance immediately after the learning session (top) and a week later (bottom) for the three learning conditions. Violins are split in half, showing the distribution of NT (blue) and Autistic (red) samples separately. C Line plot for learning performance immediately after the learning session (top) and after one week (bottom), plotted separately for NT (blue) and Autistic group (red). Error bars represent the standard error of the mean

Main effects Findings show a main effect of time: unsurprisingly, people remembered more things straight after the learning session (mean (sd) = 4.2(0.58) than a week later (Mean = 3.8, sd = 0.73; F(1,44) = 56.16, p < 0.001, η2 = 0.56, large effect size; Bakeman, 2005). More interestingly, we found a main effect of learning condition (F(2,43) = 3.86, p = 0.03, η2 = 0.15, medium effect size; Bakeman, 2005). The pairwise comparison revealed a significant learning advantage associated with Live compared to Recorded-alone condition (MLive (sd) = 4.1 (0.09); MRecorded-alone (sd) = 3.9 (0.1); t(44) = 0.19 p = 0.008). No other significant difference between learning conditions was found. No main effect was found for group: in other words, autistic people showed an NT-equivalent performance (meanautistic (sd) = 4.01 (0.09); meanNT (sd) = 3.99 (0.1); t(44) = − 0.01 p = 0.93).

Interaction effects No significant interaction effects were found between the main factors of interest (group, learning conditions and time). However, visualisation of the data (Fig. 1C) revealed a trend specific to the autistic group: while immediate recall showed a similar pattern across conditions between groups, delayed recall dropped specifically for items learned during the Live condition for the autistic group. A 2(group) × 2(time) was therefore run to test the hypothesis that delayed performance was significantly more affected for the autistic group compared to the NT group, specifically for things learned during the Live condition. Results revealed a group*time interaction effect: F(1,44) = 4.88, p = 0.03, η2 = 0.1): for things learned during Live condition, a week later, autistic people recalled significantly less things compared to NT (meanautistic (sd) = 3.75 (0.18); meanNT (sd) = 3.97 (0.16)).

Conclusions from experiment 1 & hypotheses for experiment 2

Experiment 1 found that for both NT and autistic people, learning during the Live session was associated with better recall both immediately after the session and one week later. In addition, it was found that the autistic group exhibits a decline in memory for items learned over Live interaction specifically, to a significantly greater extent than what was observed in the NT group. We acknowledge that our sample size (20 autistic adults and 26 neurotypical adults) for experiment 1 is relatively small to conduct a mixed-effects analysis of variance with sufficient power, limiting the strength of our conclusions about the role of social interaction in the learning of autistic individuals. Therefore, based on these results, a follow-up experiment was pre-registered (https://archive.org/details/osf-registrations-5pga3-v1) to confirm two main hypotheses:

1.

Participants from both groups will learn more from live calls (Live condition) compared to pre-recorded video calls (Recorded-alone and Recorded-observant condition).

2.

There will be an interaction between learning condition, group and time: while neurotypical adults will show a consistent advantage for interactive learning (Live condition) over time, the autistic group will show better immediate learning for material learnt in the Live condition and better long-term learning for materials learned from pre-recorded videos (Recorded-alone and Recorded-observant condition).