University of Southern California Dissertations and Theses

Learning self-agency during a contingency paradigm in infants at commnity and elevated risk of autism

(USC Thesis Other)

Learning self-agency during a contingency paradigm in infants at commnity and elevated risk of autism

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Copyright 2023 Marcelo Ramon Rosales

LEARNING SELF-AGENCY DURING A CONTINGENCY PARADIGM IN INFANTS AT
COMMUNITY AND ELEVATED RISK OF AUTISM

by

Marcelo Ramon Rosales

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOKINESIOLOGY)

August 2023
LEARNING CONTINGENCY PARADIGM IN INFANTS

ii
DEDICATION
To everyone who has ever taught me something along my academic journey,
this dissertation is the culmination of all those lessons.

LEARNING CONTINGENCY PARADIGM IN INFANTS

iii
ACKNOWLEDGEMENTS

When I first entered college, as an undergraduate I choose Kinesiology as my major
because I wanted to be a physical therapist to help people who had difficulties with motor
impairments. However, along the way I was introduced to the concept of research and being able
to help others through asking, developing, and answering questions. I am thankful for Dr. Rosa
Angulo-Barroso for introducing me to research and for pushing me into the right direction.
I would first like to acknowledge the unsung heroes of my research. These are the
individuals whose names do not make their way into the publications and presentations but are
integral to my work. I am speaking of course of the undergraduate research assistants and
research participants who make this work possible. I have enjoyed working with all these
individuals and I wish them the best on their bright futures!
Secondly, I would like to acknowledge my advisor Dr. Beth Smith. Thank you for
creating an environment where every research question I had was possible. Regardless of our
expertise in smart-phone surveys, eye tracking, and assessment of caregiver-child interaction, we
were able to learn these concepts and devices and produce excellent work along the way. Thank
you for developing me into the researcher I am today and for allowing me to explore several
concepts and devices with encouragement and support.
In addition to my advisor, I would also like to thank the developmental discussion group
and the Interaction Lab that has had a long list of cast members throughout my time here at USC.
I have been able to learn so much from everyone and will miss our discussions.
LEARNING CONTINGENCY PARADIGM IN INFANTS

iv
I would also like to thank everyone on my dissertation committee. Although the
pandemic made it difficult to connect and discuss research, I have learned a tremendous amount
from each of you and I was always able to learn something new form each of our meetings.
I would like to especially thank Dr. Bradley and Winstein for their support and time. It
has been a pleasure to sit in both of your offices and discuss my research and learn from each of
you. Additionally, both of you devoted several hours to helping my academic writing and taught
me to enjoy writing. I wish you both the best on the next chapter in your careers!
I would like to acknowledge all of the support from the Biokinesiology staff and faculty;
and research grant support of the National Science Foundation (CBET 1706964 PI: Smith Co-PI
Matarić) and the CHLA Best Starts to Life Research Support Grant (PI: Rosales) for partially
funding this dissertation work.
Lastly, I would like to thank my family and my lab mates for always having a listening
ear. Thank you for discussing anything with me and for supporting me throughout these years!

LEARNING CONTINGENCY PARADIGM IN INFANTS

v
TABLE OF CONTENTS
Dedication…………………………………………………………………………………...…….ii
Acknowledgements………………………………………………………………………………iii
LIST OF TABLES…...…………………………………………………………………………..vii
LIST OF FIGURES…………..……………………………………………………..…….….…viii
ABSTRACT……………………………………………………………………………….……..ix
CHAPTER I: OVERVIEW …………………………………………………………….………....1
CHAPTER II: LEARNERS DISPLAY ANTICIPATORY GAZE DURING A MOTOR
CONTINGENCY PARADIGM.….……………………….…..…………...…………………..…8
Abstract……..……………………………………………………………………………..8
Introduction……………………………..………………………………..………………10
Methods ………………………………………………………………….………………12
Participants……………………………………………………….………………12
Procedures………………………………………………………….…………….14
Data preparation …………………………………………………………………16
Data reduction …………………………………………………………………...19
Statistics …………………………………………………………………………20
Results …………………………………………………………………...………………22
Discussion …………………………………………………………….…………………30
Conclusion……………………………………………………………….………………35
CHAPTER III: BEHAVIORAL DIFFERENCES BETWEEN INFANTS AT COMMUNITY
AND ELEVATED RISK FOR AUTSIM DURING A CONTINGENCY PARADIGM…...…..37
Abstract…………………………………………………………………….……….……37
LEARNING CONTINGENCY PARADIGM IN INFANTS

vi
Introduction………………………………………………………………………………39
Methods ……………………………………………………………….…………………40
Participants……….………………………………………………………………40
Procedures……….……………………………………………….………………42
Data preparation …………………………………………………………………44
Data reduction ……………………………………………………...……………46
Statistics …………………………………………………………………………47
Results ……………………………………………………………………..…………….50
Discussion ………………………………………………………………….……………58
Conclusion………………………………………………………………….……………63
CHAPTER IV: SUMMARY AND FUTURE STEPS…………………………………………..65
REFERENCES………………………………………………………………………..…………71
APPENDICES………………………………………………………………………..………….77

LEARNING CONTINGENCY PARADIGM IN INFANTS

vii
LIST OF TABLES
Table 2.1. Participant characteristics for all infants (Mean +/- SD)……………………………..13
Table 2.2. Average behavioral state in each condition for all infants……………………………17
Table 2.3. Learning threshold, peak number of robot activations, total number of activations
during each contingency condition for each infant………………………………………………23
Table 2.4. Variables for contingency learning paradigm compared between classically
defined learners and non-learners ……………………………………………………………….28

Table 3.1. Participant characteristics for all infants (Median (Range))………………………….42
Table 3.2. Median percent for each behavioral state and condition for all infants and groups….45
Table 3.3. Contingency table depicting the number of classically defined learners and non-
learners in each group……………………………………………………….…………………...50
Table 3.4. Amount of reinforcements, predictive gazes, duration of looking, and intertrial
duration……………………………………………………………………….………………….52
Table 3.5. FYI v3.1 factors from the study sample groups (CR and ER) and an
aged-matched sample from Baranek et al. (2022) mean (standard deviation)……….………….54
Table 3.6. Average (standard deviation) behavioral measures from the contingency
data for the whole sample and three individuals at ER for autism………………………………57
Table 3.7. Individual data from the Bayley-4 from three infants at ER…………………………58

LEARNING CONTINGENCY PARADIGM IN INFANTS

viii
LIST OF FIGURES
Figure 2.1. Image of a single frame of the eye gaze data………………………………………..16
Figure 2.2. Regions of interest for overall looks at the robot (a) and predictive
and reactive looks (b)…………………………………..………………………………………...18
Figure 2.3. Predictive and reactive gaze coding for a single robot activation…………………...19
Figure 2.4. Examples of distribution for the timing of gazes on the robot from
classically defined learners…….………………………………………………………………...24
Figure 2.5. Proportion of gazes for learners……………………………………………………..25
Figure 2.6. Potential number of activations at baseline for classically defined
learners and non-learners………………………………………………………………………...27

Appendix 1. The timing of gazes on the reinforcement for all infants…………………………..77
Appendix 2. Individual plot for each infant from each group and learning classification………78
Appendix 3. Individual bar plot for each infant at community risk for autism depicting the
amount of time spent looking (y-axis) during each minute block (x-axis) of the contingency….87
Appendix 4. Individual bar plot for each infant at elevated risk for autism depicting the
amount of time spent looking (y-axis) during each minute block (x-axis) of the contingency....89

LEARNING CONTINGENCY PARADIGM IN INFANTS

ix
ABSTRACT
In the infant literature, the contingency learning paradigm has been used to examine
memory (Rovee-Collier, Sullivan, Enright, Lucas, & Fagen, 1980; Rovee & Gekoski, 1979;
Rovee & Rovee, 1969), motor learning (Angulo-Kinzler & Horn, 2001; Sargent, Kubo, &
Fetters, 2018), and the regulation of emotion (Alessandri, Sullivan, Imaizumi, & Lewis, 1993;
Alessandri, Sullivan, & Lewis, 1990). However, one area in which the contingency learning
literature can be expanded upon is an infant’s learning of self-agency (Kelso, 2016; Kelso &
Fuchs, 2016), a concept that is important for the learning and development of motor skills
(Kenward, 2010a). In the contingency learning paradigm, self-agency is defined as an infant’s
understanding of being in control of the stimulus used to reinforce their behavior (Kelso, 2016;
Kelso & Fuchs, 2016). The classical definition of learning the paradigm has been measured using
movement rate and authors considered this variable sufficient evidence of self-agency (Angulo-
Barroso et al., 2017; Dunst, Strock, Hutto, & Snyder, 2006; Kelso & Fuchs, 2016). However, the
ability to perform an action more frequently does not necessarily represent an expectation of the
outcomes for a planned movement; it merely describes that a behavior is occurring more
frequently. Therefore, we suggest that the contingency learning paradigm should incorporate an
additional measure to provide information about an infant’s understanding of an expected
outcome once a behavior is produced.
Predictive gaze is one measure that can examine an understanding of self-agency in the
contingency paradigm (Falck-Ytter, Gredebäck, & Von Hofsten, 2006; Gredebäck & Falck-
Ytter, 2015; Johnson, Amso, & Slemmer, 2003; Kenward, 2010; Rosander & von Hofsten, 2011;
Wang et al., 2012). Predictive gaze is defined as a visual behavior where the performer
anticipates an event through visual monitoring and is interpreted as demonstrating understanding
LEARNING CONTINGENCY PARADIGM IN INFANTS

x
of the phenomenon (Braukmann et al., 2018; Gredebäck & Falck-Ytter, 2015; Kanakogi &
Itakura, 2011). Predictive gaze, in addition to its ability to examine an understanding of self-
agency, can be used to describe motor learning in infants with typical development, as well
infants at risk of impairments in their motor development. In this dissertation, predictive gaze
was examined in a contingency learning paradigm to determine if infants demonstrated
predictive gaze while learning the paradigm and to describe behavioral differences between
infants at community and elevated risk for autism while they learn a contingency paradigm.
Children at an elevated likelihood of being diagnosed with autism have been described in
the literature as having delayed achievement of motor milestones (West, 2018), displaying
atypical motor control (Ekberg et al., 2016; Sacrey, Zwaigenbaum, Bryson, Brian, & Smith,
2018), and exhibiting atypical visual behavior (Johnson et al., 2015; Jones & Klin, 2013).
Furthermore, the majority of children with a diagnosis of autism are described as having motor
impairments in addition to their impairments in social communication (Bhat, 2020). While these
descriptions of motor impairments in autism are important for development of treatment plans
for these children, there is a need to better understand why these impairments exist. This
dissertation investigates if a potential reason for motor impairment in infants at elevated risk for
autism is related to how they learn motor skills. A better understanding of infant motor learning,
in general development and special populations, can be used to better design early interventions
geared at improving future and current motor impairments.
The purpose of Chapter II was to determine if learners of a contingency paradigm
demonstrate visual anticipation on a reinforcement that is linked with their movements. Fifteen
infants with typical development between 6 to 9 months old participated in the study. We used
an infant-sized humanoid socially assistive robot to provide the contingent reinforcement that
LEARNING CONTINGENCY PARADIGM IN INFANTS

xi
infants activated with their right leg movements. The robot generated a laughing sound, clapped
its hands, and flashed lights when the infant moved their right leg. Classically defined learners of
the paradigm were defined as having a ratio of 1.5 for right leg movements during the
contingency phase compared to the baseline phase. Head-mounted eye tracking was used to
determine the timing of gazes on the robot for each activation of the robot (i.e. determine if
infant anticipated the robot’s movements). Results showed that most learners (all except one
infant) had a positively skewed distribution (median: 1.09; range: 0.56-2.42) for the timing of
gazes on the robot’s activations. Additionally, the median timing of gazes on the robot preceded
the robot’s activations (median: -0.31 s; range: -0.40- 0.18 s) and classically defined learners
visually anticipated the robot’s activations more often than random chance (W= 21; p= 0.028).
These results indicate that infants classified as classically defined learners of the paradigm
demonstrated visual anticipation while engaging in the paradigm.
Additionally, we found that when comparing classically defined “learners” and “non-
learners” there was only one significant difference. Classically defined “learners”, compared to
“non-learners” had a low number of potential activations during the baseline period (Median
(range)- classically defined Learners: 11 (5-16), classically defined Non-Learners: 33 (14-42),
W= 98; p= 0.01). Given that there is nothing to learn during the baseline period and that infants
in the two groups performed the contingency portion of the paradigm similarly, we suggest that
the classical definition of learning does not adequately capture learning in this paradigm and that
further follow up assessments (i.e. a retention test or transfer task) are needed.
In Chapter III, the purpose of this section was to determine and describe the behavioral
differences between infants at community and elevated risk of autism during a contingency
learning paradigm. Eleven infants at elevated risk for autism between 6 to 9 months old, in
LEARNING CONTINGENCY PARADIGM IN INFANTS

xii
addition to the fifteen prior infants at community risk for autism (participants in Chapter II),
participated in the same contingency paradigm that was discussed in the prior section. The total
number of classically defined learners, number of robot activations, looking duration, and
intertrial duration was compared between the two groups. Results showed no group differences
in terms of the proportion of infants classified as classically defined learners and non-learners,
and there were no visual motor differences. Overall, these results indicated that infants at
elevated risk for autism learn contingency paradigms in a similar way as infants not at elevated
risk. However, we did find three infants at elevated risk for autism who did not show the visual
motor patterns that were described in Chapter II. Our assessment of these individuals using a
screening questionnaire for autistic traits and a developmental assessment showed lower
performance on the assessment and some concerns of early signs of autism in one individual.
However, these scores seem to be within the average ranges; and follow up is needed once the
diagnosis status of these children is known, to determine if infants who are later diagnosed with
autism have difficulties learning motor skills. Additionally, our analysis for evidence of autistic
traits showed no grouped differences in this small sample so it is possible that none of these
participants will be diagnosed with autism in the future.
Our small sample size might have contributed to the null findings in Chapter III as well,
it should be noted that contingency learning is only one type of motor learning paradigm and
other motor learning paradigms may revel subtle differences in motor learning. It has been
described in the literature that the severity of motor impairments in autistic children is correlated
with the severity of their autism and that there is variability in the severity of their motor
impairments (Bhat, 2021). Therefore, we suggest that an examination of other motor learning
paradigms such as error-based learning may provide us with more information about motor
LEARNING CONTINGENCY PARADIGM IN INFANTS

xiii
learning in autism. Finally, we also suggest that children at risk of autism may use motor
learning adaptations while acquiring motor skills and that an examination of adaptive motor
learning strategies could shed light on why motor impairments exist in children with autism.
In summary, findings suggest that infants display anticipatory gaze in a contingency
paradigm. This shows evidence of infants learning self-agency and understanding the connection
between their actions and the outcomes for their actions. In terms of next steps, we provide the
groundwork for a novel study design that can be used to examine how infants integrate their
visual motor system to learn a motor paradigm and how visual gaze can be used to support a
determination of learning in a contingency paradigm. While the use of a retention or transfer task
is needed to confirm our findings about motor learning in infants at community risk for autism,
this paradigm should be implemented in other pediatric populations who may have motor
learning difficulties. Additionally, we suggest that other motor learning paradigms or an analysis
of alternate motor learning behaviors can inform us about potential motor learning differences
across the autism spectrum.

LEARNING CONTINGENCY PARADIGM IN INFANTS

1
CHAPTER I: OVERVIEW

Statement of the problem
The contingency learning paradigm has long used a behaviors’ rate of occurrence to
categorize if an infant learned the paradigm or not (Angulo-Barroso et al., 2017; Angulo-Kinzler,
Ulrich, & Thelen, 2002; Fitter et al., 2019). Although the measure of behavioral rate is the
classical definition of learning, other behavioral measures of learning should be utilized (Kelso,
2016; Kenward, 2010a; Krakauer, Hadjiosif, Xu, Wong, & Haith, 2019). Modern perspectives of
infant learning suggest that infants actively learn motor skills and display behaviors such as
visual anticipation and a lack of pupil dilation once a concept is learned (Gredebäck, Johnson, &
Von Hofsten, 2010; Kenward, 2010b). These variables of visual behavior have rarely been
explored in the contingency learning literature (Kenward, 2010a; Zaadnoordijk, Otworowska,
Kwisthout, & Hunnius, 2018), possibly due to the convenience of measuring the occurrence of
the reinforced behavior and reliance on using the classically defined variables. However, an
examination of visual behavior during a contingency learning paradigm will shed light on how
infants utilize visual motor behavior to learn the consequences of their motor behaviors (i.e. self-
agency), acquire an understanding of the motor paradigm, and can be used to determine and
describe motor learning characteristics in populations who need early motor interventions.
Additionally, other measures of cognitive learning can be used to enhance the contingency
learning literature by expanding our knowledge of infant motor learning.
One population that has been characterized in the motor literature as having difficulties
with the development and execution of motor skills are children on the autism spectrum (Bhat,
2020; West, 2018). There is a growing literature that suggests motor impairments exist in
LEARNING CONTINGENCY PARADIGM IN INFANTS

2
children who go on to be diagnosed with autism spectrum disorder (ASD) (Bhat, 2020; Kaur,
Srinivasan, & Bhat, 2018) and motor delays in infants at elevated risk of ASD. However, we do
not know the reasons for these motor impairments and delays. Given the existing literature
describing atypical sensory motor processing in children who are later diagnosed with autism
(Sacrey et al., 2018; Thorup, Nyström, Gredebäck, Bölte, & Falck-Ytter, 2018); one rationale for
motor impairments are differences in ways these children learn motor skills. However, it is yet to
be determined if children who go on to be diagnosed with autism have impairments in motor
learning or motor execution (i.e. motor control), or both. Therefore, an examination of the ways
infants at elevated risk for autism use their visual motor system to learn a motor contingency
paradigm will help characterize potential learning processes as well as inform future early motor
interventions for this population.
The two areas that this dissertation aims to address are 1) to determine if infants use
visual motor learning while engaging in a contingency learning paradigm, and 2) to describe and
assess potential visual motor differences in infant motor learning between infants at elevated and
community level risk for autism. These two areas will expand our knowledge of infant motor
learning and help inform future early motor interventions for infants with or at elevated risk for
autism.

Assessing predictive gaze for a better understanding of self-agency during infancy
Self-agency is a construct that is defined as a feeling of control over an external event
(Chambon, Sidarus, & Haggard, 2014; Moore, 2016; Swiney & Sousa, 2014) or an
understanding of producing an outcome in the environment (Kelso, 2016; Kelso & Fuchs, 2016).
Although there is a sizable literature studying self-agency (Chambon et al., 2014; Kelso, 2016;
LEARNING CONTINGENCY PARADIGM IN INFANTS

3
Moore, 2016), research is still lacking in our understanding of when self-agency arises in an
individual and when humans first experience self-agency (Kelso, 2016). The works of Kelso
suggest that we can better understand self-agency in infants through the use of contingency
learning paradigms (Kelso, 2016; Kelso & Fuchs, 2016). He states that during a contingency
learning paradigm the infant is in control of their own actions and must learn the contingency
between their behavior and the reinforcement given for the correct behavior. Additionally, Kelso
hypothesizes that once an infant realizes that they are in control of the reinforcement, they will
rapidly produce the behavior in order to maximize the amount of reinforcement (Kelso & Fuchs,
2016).
Although Kelso’s perspective logically makes sense, this understanding of self-agency
can be examined further during infancy. In the adult literature examining self-agency,
researchers use a motor task that has a number of possible movement strategies and give binary
feedback for correct and incorrect movements (i.e. a signal for being right and no signal for
being wrong) (Chambon et al., 2014; Moore, 2016). While the participants are performing the
task, they are asked to describe how they moved to perform the task correctly when they believe
they have figured out the paradigm. Then researchers examine their behavior to see how their
vocalization of understanding the paradigm match with their behavioral data (Chambon et al.,
2014; Moore, 2016). In Brooks et al. (1995), participants saw the greatest improvements in
motor performance right before they vocalized the correct movement pattern, and then refined
the correct motor pattern through practice. Although an infant cannot explicitly answer if they
know how to perform a task, we can examine their behavior and use the methodology that is
implemented in the adult literature to probe their understanding beyond considering frequency of
behavior alone.
LEARNING CONTINGENCY PARADIGM IN INFANTS

4
One behavior, in particular, is anticipatory visual behavior, or predictive gaze
(Gredebäck, Johnson, & Von Hofsten, 2010; Johnson et al., 2003). In the adult literature
examining self-agency (Moore, 2016) and discovery learning (Sailer, Flanagan, & Johansson,
2005), predictive gaze has been shown to be present in individuals who have an explicit
understanding of a paradigm (Moore, 2016) or in individuals who can perform the motor
behavior accurately and consistently (Moore, 2016; Sailer et al., 2005). Additionally, research
suggests that an infant’s ability to perform a motor skill is related to their ability to visually
anticipate (interpreted as understanding) the actions of another person performing the skill in
question (Falck-Ytter et al., 2006; Gredebäck & Falck-Ytter, 2015). Therefore, it can be stated
that there is evidence of a relationship between understanding a particular motor skill (as
demonstrated by predictive gaze) and motor learning and performance. Through examining the
visual behavior and motor performance of an individual, we can begin to understand their
cognitive ability to learn and perform a motor paradigm.
With these ideas in mind, self-agency will be better understood in infancy if we examine
visual behavior during a contingency learning paradigm. The results will describe the visual and
motor behavior while infants engage with the paradigm; and support the claim that infants
display visual anticipation while learning motor paradigms. In doing so, we will lay down the
groundwork for examining self-agency and motor learning during infancy. Additionally, we will
be able to compare the learning of a sample that is at risk of motor delays and later motor
impairments (i.e., infants at elevated risk for ASD) to a sample at community risk for autism.

LEARNING CONTINGENCY PARADIGM IN INFANTS

5
Motor learning difficulties in infants at elevated risk of ASD
Infants at elevated risk of ASD have been described in the literature as having delayed
social, cognitive, and motor development (Braukmann et al., 2018; Johnson et al., 2015; West,
2018). Focusing in on motor development, most of the research that examines motor delays in
infants at elevated risk of ASD only describe their motor milestones as being achieved later than
usual or their lower motor performance on assessments (Sacrey et al., 2018). Literature focusing
on motor control has suggested that infants at elevated risk for autism have difficulties timing
their motor behaviors to accomplish a motor action, such as reaching for an object (Ekberg et al.,
2016; Sacrey et al., 2018) or balancing while seated (Kyvelidou et al., 2021). Although these
assessments of motor control and motor performance are valuable for getting a general sense of
motor delays and impairments, they do not provide information on the behavioral rationale for
why these impairments exist. It is hypothesized that older children with ASD have difficulties
with motor learning (Moraes et al., 2017) and that these difficulties begin in infancy (Johnson et
al., 2015). However, further research is needed to better understand how infants at elevated risk
for autism learn motor skills and if impaired motor learning is a potential rationale for their later
motor impairments.
To our knowledge, no studies exist that have examined the possible differences in
contingency learning among infants at community and elevated risk of ASD. Assessing infants at
elevated risk of ASD during a contingency paradigm is valuable because it will aid in our
understanding of motor learning difficulties in this population by assessing how they attempt to
understand that their actions are controlling the reinforcement. Hence, describing how infants at
elevated risk of ASD learn (or do not learn) that their motor behavior causes an outcome.
Additionally, we will be able to describe the possible differences in how infants at community
LEARNING CONTINGENCY PARADIGM IN INFANTS

6
and elevated risk of ASD attempt to learn the paradigm. Through this, we will be able to
formulate new hypotheses about specific impairments in motor learning that infants at elevated
risk of ASD may possess.

Aims and Hypothesis
Specific Aim 1: To determine if classically defined learners of a contingency paradigm
demonstrate visual anticipation on a reinforcement that is linked with their movements.
Hypothesis 1a: Classically defined learners will display a positive skewed distribution
for the time of their gazes on the robot activations (i.e. gazes that occur prior to the robot
activation).
Hypothesis 1b: Classically defined learners of the paradigm will display more predictive
gazes on the robot activations compared to random chance.

Specific Aim 2: Determine and describe the behavioral differences between infants at
community and elevated risk level of ASD during a contingency learning paradigm.
Hypothesis 2a: Infants at elevated risk compared to community risk for autism will be
less likely to learn the paradigm.
Hypothesis 2b: Infants at elevated risk compared to community risk for autism will
exhibit lower amounts of predictive gaze, robot activations, and looking duration on the robot.
Hypothesis 2c: Infants at elevated risk compared to community risk for autism will
exhibit longer intertrial durations.

LEARNING CONTINGENCY PARADIGM IN INFANTS

7
Summary
In the long term, we want to understand how infants use their visuomotor behaviors to
learn motor paradigms. Knowledge of infant motor learning is used to help facilitate the design
of motor learning interventions in pediatric populations who may have difficulties learning motor
skills. Specifically in this project we identified behavioral differences between infants at elevated
and community risk for autism while they learned a motor paradigm. We used head mounted eye
tracking and wearable sensors during a contingency learning paradigm to assess how infants
used their vision and movements in the paradigm. Findings from this project provided the
knowledge and tools to investigate motor learning and potential intervention strategies in infants
at elevated risk for autism.

LEARNING CONTINGENCY PARADIGM IN INFANTS

8
CHAPTER II: LEARNERS DISPLAY ANTICIPATORY GAZE DURING A MOTOR
CONTINGENCY PARADIGM

Abstract
Background: An examination of visual behavior during a motor learning paradigm can enhance
our understanding of how infants learn motor skills and understand the consequences of their
actions (i.e. self-agency).
Aim: To determine if learners of a contingency paradigm demonstrate visual anticipation on a
reinforcement that is linked with their movements.
Methods: Fifteen full-term infants (6-9 months of age) at community risk for autism (i.e. no first
or second degree relatives diagnosed with ASD) participated in a contingency learning paradigm.
A robot provided reinforcement by clapping when the infant produced a right leg movement. The
timing and type of visual gaze (i.e. predictive, reactive, or not looking) on the robot activations
was identified using a frame-by-frame analysis. The distribution for the timing of the gazes on
the robot’s activations was tested for skewness and was described using the median time of
occurrence. The proportion of predictive gaze was tested using a Wilcoxon signed ranked test to
determine if learners visually anticipated the robot’s activations more than random chance.
Lastly, an exploratory analysis describing the trends in visual-motor behavior was used to inform
future questions and practices in contingency learning studies.
Results: For most classically defined learners (all but one infant), results showed that the
distribution for the timing of the gazes on the robot had a positive skew range (0.56-2.42) and the
median timing of gazes preceded the robot’s activations by 0.31 seconds (range: -0.40- 0.18 s).
LEARNING CONTINGENCY PARADIGM IN INFANTS

9
Additionally, classically defined learners of the paradigm visually anticipated the robot’s
activations more than random chance (W= 21; p= 0.028).
Conclusion: This study suggests that infants classified as learning the paradigm use visual
anticipation while performing the paradigm. However, our exploratory analyses sheds light on
the issues of using the classical definition of learning a contingency paradigm and provided
suggestions and further questions for infant contingency learning studies and the processes
infants use to learn motor skills.

LEARNING CONTINGENCY PARADIGM IN INFANTS

10
Introduction
In the adult motor learning, adults utilize different types of motor learning (i.e. implicit
and explicit learning) and behaviors (e.g. verbal statements of learned concepts and differences
in movement and visual patterns) in order to acquire and express that an individual has learned
different aspect of a motor paradigm (Brooks et al., 1995; Johansson, Westling, Backstrom, &
Flanagan, 2001; Krakauer et al., 2019; Sailer et al., 2005). For example, in Brooks et al. (1995),
learners of the motor reversal paradigm had to first learn the general movement pattern via an
explicit learning strategy and then refine the skill and learn the motor tactics through a more
implicit motor learning strategy. The explicit learning was identified through a participant
verbally stating the correct movement pattern and the implicit learning strategy was identified
through improvement in motor performance. Thus, through examining multiple means of
learning skills, we can gain a more complete understanding of motor skill acquisition.
In the infant contingency learning literature, the main means of defining learning has
been through the examination of behavioral rate (Angulo-Barroso et al., 2017). Infants are
usually placed in a paradigm, where a certain behavior is reinforced over a set period (Angulo-
Kinzler, 2001; Rovee-Collier et al., 1980; Sargent, Schweighofer, Kubo, & Fetters, 2014).
Learning is then classically defined by increasing the number of reinforced behaviors during a
contingency period (i.e. a period where an infant’s behavior is contingent for activating the
reinforcement) compared to an unreinforced period recorded prior to the contingency period.
Although motor practice and exploration are important for motor learning in the long-term
(Ulrich, 2010), solely focusing on the frequency of behavior limits the scope of our
understanding of infant motor learning.
LEARNING CONTINGENCY PARADIGM IN INFANTS

11
One specific area of interest that has been discussed in the contingency learning literature
is the understanding of self-agency (Kelso, 2016; Kelso & Fuchs, 2016; Zaadnoordijk et al.,
2018). Self-agency is a construct that is defined as a feeling of control over an external event
(Chambon et al., 2014; Moore, 2016; Swiney & Sousa, 2014) or an understanding of producing
an outcome in the environment (Kelso, 2016; Kelso & Fuchs, 2016). In the context of the
contingency learning literature, this would be an understanding that the reinforced behavior is
causing the reinforcement to occur. Scholars have suggested that an examination of visual
behavior during these paradigms can help aid in understanding if infants have an understanding
of self-agency (Zaadnoordijk et al., 2018; Zaadnoordijk, Otworowska, Kwisthout, Hunnius, &
Van Rooij, 2017) and if they use visual motor behavior to learn concepts about their
environment.
The purpose of this chapter was to examine if infants who are classified as learners of a
contingency learning paradigm demonstrate predictive gaze while learning the contingency
paradigm. Predictive gaze is defined as looking at the area of interest prior to an event occurring.
This has been used in several looking paradigms that examine object permanence and action
perception (Gredebäck & Falck-Ytter, 2015; Gredebäck et al., 2010) and is typically used to
support that an infant knows a concept that has been learned prior to the experiment. However, it
is not clear if infants display these anticipatory looks during the learning of novel motor skills
and how this visual motor behavior evolves during a motor learning paradigm.
In this study, we used a socially assistive robot to reinforce movements of the right leg.
The aim of the study was to determine if learners of a contingency paradigm demonstrate visual
anticipation on a reinforcement that is linked with their movements. We hypothesized that
infants who are classified as classically defined learners of the paradigm would display a right
LEARNING CONTINGENCY PARADIGM IN INFANTS

12
skewed distribution for the timing of their gazes on the reinforcement with the median value for
their gazes occurring prior to the robot’s activations. Secondly, we hypothesized that classically
defined learners would exhibit predictive gaze (time their gazes to occur prior to the activation of
the robot (i.e. the reinforcement)) more than random chance.
To better understand the variability of infant motor learning, we examined the behavioral
differences between infants classified as classically defined learners and non-learners; as well as
describing the common visual motor patterns that were displayed amongst the whole group. We
had a reasonable suspicion that classically defined learners and non-learners were more similar
than not, given that the classical definition of learning the paradigm based on movement rate
alone has been debated (Kelso, 2016; Kelso & Fuchs, 2016; Kenward, 2010a; Zaadnoordijk et
al., 2020, 2017). Further explanation of these analysis and results are discussed later; and we
provide future recommendations for contingency learning studies as well further questions about
the process of infant motor learning.

Methods
Participants
Seventeen infants at community risk for autism were enrolled in the study. They were
recruited by fliers, online postings, and word of mouth in the greater Los Angeles area between
October 2021 and April 2023. The inclusion criteria were that the infant was born full term (>37
weeks gestational age) and between the age of 6 and 9 months old. Infants with any known
visual, hearing, or orthopedic impairments were excluded from the study. Additionally, infants
with a first or second degree relative diagnosed with autism were excluded from this group of
infants but were included in our second study using the same methods (See Chapter III). Lastly,
LEARNING CONTINGENCY PARADIGM IN INFANTS

13
infants were excluded from the data set if they scored lower than the fifth percentile on the
Bayley Scales of Infant and Toddler Development (fourth edition) for the average of all three
domains tested: Cognitive, Language, and Motor (Bayley & Alyward, 2019) or cried
continuously for 1 minute of the contingency paradigm.
Two infants were excluded from our data set which give us the final sample of 15 infants.
One infant was unable to participate in the contingency paradigm due to a technical error with
our robot and their data were excluded. Additionally, one infant was excluded for crying for
longer than 1 minute during the paradigm. Participant characteristics are in Table 2.1.

Table 2.1. Participant characteristics for all infants (Mean +/- SD).
Age (days) 212 (20)
Sex 7 males, 8 females
Weight (kg) 7.58 (0.59)
Bayley-4 Cognitive, Percentile 72 (11)
Bayley-4 Language, Percentile 65 (16)
Bayley-4 Motor, Percentile 68 (20)
Ethnicity 6 Hispanic, 9 Not Hispanic/Latino
Race 3 Asian, 3 Black/African American, 2
White, 6 Other/Multi-Racial, 1 Declined
to answer
*Bayley-4: Bayley Scales of Infant and Toddler Development, fourth edition.

LEARNING CONTINGENCY PARADIGM IN INFANTS

14
Procedures
The research was approved by the Institutional Review Board of the University of
Southern California (HS-14-00911). A parent or legal guardian signed an informed consent form
before their infant’s participation in the study. Data were collected at Children’s
Hospital Los Angeles. At each data collection, we collected the infant’s anthropometric data
(thigh length and circumference, shank length and circumference, foot length and width, and
weight), assessed their motor, cognitive, and language development using the Bayley Scales of
Infant and Toddler Development (version 4), and had a parent fill out the First Year Inventory
(FYI version 3.1) (Baranek, Watson, Crais, Turner-Brown, & Reznik, 2013). The Bayley-4 is a
standardized observational assessment that assess the motor, cognitive, and language
development of children between the ages of 1-42 months that yields standardized scores, age
equivalents and percentiles based on normative data. The FYI v3.1 is a 69-item parent report
questionnaire about infant behaviors that may indicate an elevated likelihood for later
neurodevelopmental conditions such as autism (Baranek et al., 2022). The questionnaire
generates risk scores on seven computed factors; these factors were developed from a large
community sample of infants ranging from 6 to 17 months of age who were followed to age 3 to
assess diagnostic and developmental outcomes. The factors are: 1-communication, imitation, and
play; 2- social attention and affective engagement; 3- sensory hyperresponsiveness; 4- sensory
hyporesponsiveness; 5- self-regulation in daily routines; 6- sensory interest, repetitions, and
seeking behaviors; and 7- motor coordination and milestones.
Infants were supported by a chair and secured at the trunk with a cloth band in front of an
infant-sized humanoid socially assistive NAO robot (Aldebaran United Robotics Group).
Participants engaged in a twelve-minute contingency paradigm where movements of their right
LEARNING CONTINGENCY PARADIGM IN INFANTS

15
leg were reinforced by a robot (see Figure 2.1) that would produce a clapping motion and a
laughing sound. The structure of the paradigm was a 2-minute baseline, a 10-second
demonstration, an 8-minute contingency condition, and a 2-minute extinction phase (see Figure
2.2). Infant right leg movements were not reinforced during the baseline, demonstration, and
extinction periods.
During the contingency paradigm, infants wore a head-mounted eye tracker (Positive
Science) on their heads and four wearable Opal sensors (APDM Inc) that were placed on the
arms and legs (one for each limb). Eye gaze data were recorded at 30 frames per second and
wearable sensor data were collected at 128 Hz. Prior to the start of the contingency paradigm, a
5-point calibration was performed for the eye gaze data where infants would look at a spinning
globe toy that would light up in 5 different points in front of the robot. The 5-point calibration
was performed at 1.5 meters between the seated infant and the robot.
The infant-robot interaction system employed in this study was comprised of the NAO
robot and wearable sensors. This is the same system that was designed and used in Fitter et al.
(2019), except infants only received a single type of reinforcement (i.e. the robot clapping and
laughing). Briefly, the robot would activate during the contingency phase if the infant produced
an acceleration of greater than 3 m/s
2
as measured by the right leg wearable sensor. For further
detail please refer to Fitter et al. (2019).
Lastly, two cameras were placed externally in the contingency paradigm space. One
camera was behind the robot and recording the infant in their chair; the second camera was
placed on the side of the infant and the robot and was recording both the infant and the robot
from a side view. Prior to the start of the contingency and after the 5-point calibration, the
spinning globe toy (refer to 5-point calibration section above) was turned on and off repeatedly 3
LEARNING CONTINGENCY PARADIGM IN INFANTS

16
times and shown simultaneously to the two external cameras and to the eye tracker. This was
performed to synchronize all video data post collection.

Data Preparation
After the data collection, data from the eye tacker were imported into the software
Yarbus. Yarbus is an eye tracking software that was developed by Positive Science that
synchronizes the two cameras on the eye tracker (the scene camera located on the infant’s
forehead and the pupil camera located over the eye). Once the cameras are synchronized, a single
researcher selects the 5 points that the infant looked at during the 5-point calibration. Given that
the eye tracker has an error of a 4-degree radius, we import a graphic overall of a target that
represents the 4-degree radius that contains the approximate area that the infant is looking at
during the paradigm. The overall result is a single video clip of the whole contingency paradigm
that is depicted in Figure 2.1.

Figure 2.1. Image of a single frame of the eye gaze data.

*The red circle is the 4-degree radius
LEARNING CONTINGENCY PARADIGM IN INFANTS

17
Once the eye gaze video was created, it was synchronized with the side view camera’s
video using ELAN software (ELAN 5.8, Language Archive). The common starting point that
was used to synchronize the two videos was the point where we flashed the globe toy 3 times.
This calibration has an error of +/- 1 frame. After the synchronized video file was created, a
custom Python software was used to identify each time the robot activated according to the video
data’s timeline. Data from the sensors and the robot indicated when the robot activated. The
timing for each robot activation was confirmed in our frame-by-frame analysis discussed next.
Trained video coders coded the synchronized video files using a frame-by-frame analysis.
Coders were trained on testing data sets and had to reach a reliability above 80% on their video
files before analyzing data. Coders coded for behavioral state and visual behavior and one third
of the videos were assessed for reliability. Percent of agreement for behavioral state was 95%,
types of visual gaze was 87.5%, and time spent looking at the robot was 95.25%. Behavioral
state was coded as sleeping, drowsy, alert, fussy, or crying (Lester & Tronick, 2004) throughout
the whole contingency paradigm (Table 2.2).

Table 2.2. Average behavioral state in each condition for all infants. Values are percent of time
spent in each state (Mean (SD)).
Baseline Contingency Extinction
State Alert Fussy Crying Alert Fussy Crying Alert Fussy Crying
Percent 92.1
(16.8)
7.3
(15.9)
0.7
(2.6)
95.6
(8.3)
3.5
(6.7)
0.9
(2.4)
94.5
(17.8)
2.7
(7.6)
2.7
(10.6)

LEARNING CONTINGENCY PARADIGM IN INFANTS

18
Visual behaviors that were coded for included: the start and end of each gaze on the robot
(regions of interest defined in Figure 2.2a) and each instance of predictive and reactive gaze on a
robot’s activation (region of interest defined in Figure 2.2b). Predictive gazes (Gredebäck &
Falck-Ytter, 2015) were defined as a visual fixation (3 or more frames of no eye movement)
(Franchak, Kretch, Soska, & Adolph, 2011) on the robot prior to its activation and no earlier than
12 frames prior to the robot’s activation (Wass et al., 2015). In instances when no predictive gaze
occurred, a reactive gaze (Gredebäck & Falck-Ytter, 2015) was defined as a visual fixation on
the robot during its activation (60 frames). Last, if no gaze occurred on the robot during an
activation, behavior coders marked the occurrence as a non-robot look. Figure 2.3 shows the
depiction of a single robot activation.

Figure 2.2. Regions of interest for overall looks at the robot (a) and predictive and reactive looks
(b).

a.
b.
LEARNING CONTINGENCY PARADIGM IN INFANTS

19
Figure 2.3. Predictive and reactive gaze coding for a single robot activation. Note: Predictive
gaze would be a time window of 0.4 seconds, and a reactive gaze would be in a time window of
2 seconds.

Data Reduction
Determining learning based on leg movement and the classical definition of learning
A learning threshold was determined based on the prior literature where an infant is
classified as a classically defined learner if they produce the behavior that is reinforced 1.5 times
more during the contingency phase compared to the baseline. Potential activations during the
baseline needed to be greater than an acceleration of 3.0 m/s
2
(i.e. the robot activation threshold)
and had to occur more than 2 seconds after the last potential activation (i.e. the duration of a
single robot activation). Using MATLAB, the number of times an infant would have been
reinforced was calculated using the wearable sensor data from the right leg and then multiplied
by 1.5 to calculate the learning threshold. If an infant activated the robot above the number
specified as the learning threshold in a two-minute moving window during the contingency
phase, they were classified as a classically defined learner of the paradigm.

Leg
movement
Predictive (12 frames) Reactive (60 frames)
Robot
on
Robot
off
LEARNING CONTINGENCY PARADIGM IN INFANTS

20
Timing of gazes on each robot activation
To calculate the timing of the gazes on each robot activation, the start of each predictive
or reactive gaze was subtracted from the start of the robot activation. This means that negative
values were predictive gazes and positive values were reactive gazes (see Figure 2.3).

Proportion of gazes
To calculate the proportion of each type of gaze on the robot, the frequency of each gaze
was counted for each infant (i.e. predictive, reactive, and none gaze count) and then was divided
by the total number robot activations.

Statistical Analysis
Non-Parametric tests and descriptive statistics were used given the small sample size and
because most variables were not normally distributed. Effect sizes were also calculated for each
comparison using Pearson’s r with the following explanations for r: small effect was |r| > 0.1,
moderate effect |r| > 0.3, and large effect |r| > 0.5 (Cohen, 1988). All computations for frequency
and proportion of each type of gaze, classification of learning, and timing of gazes on the
reinforcement were computed using custom MATLAB programs and exported to be analyzed in
SPSS (v.27).
Our first aim was to determine if classically defined learners of a contingency paradigm
demonstrate visual anticipation on a reinforcement that is linked with their movements. We
hypothesized that infants classified as classically defined learners would exhibit a positive skew
in the distributions for their timing of gazes on the robots’ activations. We perform our analysis
of skewness on the classically defined learners to answer our first hypothesis, and then on the
LEARNING CONTINGENCY PARADIGM IN INFANTS

21
classically defined non-learners for the purpose of our exploratory analysis. Qualitatively, the
distribution for the timing of gazes on the robot’s activations were plotted using a frequency
graph and these plots were visually inspected for skewness (Bulmer, 1967). Quantitatively, we
calculated skewness and the median timing of gaze on the robot’s activation for each infant.
Skewness values greater than 0.5 were considered positively skewed (Bulmer, 1967) and
negative median values denote that the center value for the timing of gazes preceded the robots
activations. Using both of our qualitative and quantitative analyses we determined if the timing
of gazes on the robot for each infant were skewed.
Our second aim was to determine if infants classified as classically defined learners
displayed predictive gaze more than random chance. The proportion of predictive and reactive
gazes and non-looks were tested to determine if predictive gazes were occurring more or less
than random chance (i.e. different from 33.33%) using a two-sided one-sample Wilcoxon
Signed-Ranked test for the classically defined learners. No statistical adjustment was used for
these analyses because our hypothesis was only concerned with predictive gaze.
Last, the following variables were compared between classically defined learners and
non-learners using a Wilcoxon Rank Sum test with Bonferroni adjustment: number of potential
activations at baseline and extinction phases, total number of activations during the contingency
phase, proportion of gazes for reactive and predictive gazes, proportion of non-looks, total time
looking at the robot during the contingency phase, median time for gazes on the reinforcement,
skewness for the timing of gazes on the reinforcement, and average intertrial duration.
Additionally, we used Spearman correlations to determine if being classically defined as a
learner or non-learner was associated with the scales from the Bayley-4 (cognitive, language, and
motor) using percentiles scores. This analysis was used to determine differences between the
LEARNING CONTINGENCY PARADIGM IN INFANTS

22
learners and non-learners and supported our exploratory analysis which will be described in the
results section. Bonferroni adjustment was used given our multiple comparisons.

Results
Number of classically defined learners amongst participants
Using the learning threshold of 1.5 times more movements during the contingency phase
compared to the baseline, our data set contained 6 infants who were defined as classically
defined learners and 9 infants who were defined as classically defined non-learners (Table 2.3).
It should be noted that eye gaze data for two classically defined non-learners was not collected
due to the infant not tolerating the eye tracker (see asterisked values in Table 2.3). These infants
were able to tolerate the contingency paradigm without the eye tracker and their data are used for
non-eye gaze related variables.

LEARNING CONTINGENCY PARADIGM IN INFANTS

23
Table 2.3. Learning threshold, peak number of robot activations, total number of activations
during each contingency condition for each infant.
Infant Baseline
(number of
activations)
Threshold
(number of
activations)
Peak contingency
block (activations/
2 minutes)
Extinction
(number of
activations)
Total
number of
activations
NL1 40 60 32 31 90
NL2 37 55.5 40 32 124
L1

10 15 36 18 100
L2

12 18 28 29 63
NL3*

14 21 13 8 31
L3

16 24 29 20 66
NL4* 42 63 48 47 160
L4

5 7.5 31 22 87
NL5 18 27 16 18 43
L5

9 13.5 35 9 77
NL6 34 51 34 23 110
NL7 26 39 14 34 33
NL8 33 49.5 34 21 72
L6

12 18 19 29 41
NL9 33 49.5 41 32 97
*Infants without eye tracker data. Infant classified as classically defined learners are denoted as
L and classically defined non-learners as NL.

Distribution for timing of gazes on the reinforcement for learners
Figure 2.4 shows examples of the distribution for the timing of gazes on the robot for
individual participants and Appendix 1 shows the data for all infants. In support of our Aim 1
hypothesis, visual inspections showed that the classically defined learners displayed evidence of
having a positive skew in their timing of gazes on the robot’s activations. Additionally,
classically defined learners had a range of skewness from 0.56-2.42 (median= 1.09). The median
for the median (min-max) timing of the gazes preceded the robot’s activations by 0.31 seconds (-
0.4- 0.18 s). Descriptively this means that classically defined learners had a distribution for the
LEARNING CONTINGENCY PARADIGM IN INFANTS

24
timing of their gazes that was slightly positive to positively skewed toward having gazes that
were predictive or close to the time that the robot activated.

Figure 2.4. Examples of distribution for the timing of gazes on the robot from classically defined
learners.

Distribution for timing of gazes on the reinforcement for non-learners
Results showed that classically defined non-learners had a range of skewness from 0.1-
2.07 (median= 1.61). The median for the median (min-max) timing of the gazes on the robot’s
activation was -0.37 s (-0.4-0.75 s) for the classically defined non-learners. According to a
Wilcoxon Ranked Sum test no statistically significant differences were found between classically
defined learners and non-leaners regarding the skewness of timing of gazes on the reinforcement
(W= 53, p=0.628, small effect- Pearson’s r=0.16) and the median timing of gazes on the
reinforcement (W=45.5, p=0.628, small effect- Pearson’s r=0.14).
The plotted distribution for the timing of gaze on the reinforcement for all infants can be
seen in Appendix 1. Only one infant had a distribution that was not positively skewed, and four
infants had a positive value for the median timing of gazes on the robot (i.e. their median timing
-0.4 0 0.4 0.8 1.2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
LEARNING CONTINGENCY PARADIGM IN INFANTS

25
of gaze on the robots activations occurred after the robot had activated). It should be noted that 2
of the 4 infants with a positive median for the timing of gazes on the reinforcement were within 2
frames (i.e. 0.066 s) of the reinforcement (median timing was 0.01 s and 0.065 s). The rest of the
infants (9 of the 13 infants) had distributions that were positively skewed with a negative median
timing of gazes on the robot.

Proportion of predictive gazes amongst learners
In support of our second hypothesis, a two-tailed one-sample Wilcoxon Signed Ranked
test showed that classically defined learners visually anticipated the robot’s activations
significantly higher than random chance (i.e. 0.333333) (W= 21, p=0.028) with a large effect
(Pearson’s r=0.895). Classically defined learners did not exhibit reactive gazes (W=12, p=0.753,
small effect- Pearson’s r=0.128) or non-robot looks (W=3, p=0.116, large effect- Pearson’s
r=0.639) significantly different than random chance.

Figure 2.5 Proportion of gazes for classically defined learners.

0
0.2
0.4
0.6
0.8
1
Non-Robot Looks Reactive Predictive
Proportion
Proportion of Gazes
LEARNING CONTINGENCY PARADIGM IN INFANTS

26
Comparing predictive gazes amongst learners and non-learners
Compared to classically defined learners, classically defined non-learners did not exhibit
significant differences in the proportions of predictive gazes (W=50, p=1, no effect- Pearson’s
r=0.04), reactive gazes (W=50, p=1, no effect- Pearson’s r=0.04), or non-robot looks (W=52,
p=0.731, small effect- Pearson’s r=0.12).

Comparison of classically defined learners and non-learners for all variables
For our exploratory analysis, our prior results showed that classically defined learners
and non-learners did not have any statistical differences for the skewness and median timing of
gazes on the reinforcement. Additionally, there were no statistical differences for the proportion
of gazes of each type of gaze between the groups. Upon further exploration of all our variables,
the only statistical differences between classically defined learners and non-learners were the
total number of movements that would have activated the robot during the baseline phase
(W=98, p=0.0012, large effect- Pearson’s r= 0.79) (see Figure 2.6). Table 2.4 shows statistical
data for all other variables.
Additionally, no statically significant associations were found between learning
classification and cognitive (r=0.36, p=0.19), language (r=-0.24, p=0.39), and motor percentiles
(r=0.27, p=0.33) using Spearman’s correlations.

LEARNING CONTINGENCY PARADIGM IN INFANTS

27
Figure 2.6. Potential number of activations at baseline for classically defined learners and non-
learners.

0
5
10
15
20
25
30
35
40
45
Learners Non-Learners
Count
Activations at Baseline
LEARNING CONTINGENCY PARADIGM IN INFANTS

28
Table 2.4. Variables for contingency learning paradigm compared between classically defined
learners and non-learners. Values in classically defined learners and non-learners columns are
median (range). Values in the Statistics column are the Wilcoxon’s W values and p-values. Eye
gaze variables for classically defined non-learners have an n=7 and are denoted with an asterisk
*. Significant results are denoted with two asterisks **.
Classically defined
Learners (n=6)
Classically defined
Non-learners (n=9)
Statistics (p-value)
Total potential
baseline activations
11 (5-16) 33 (14-42) W=98 (p=0.0012)**
Total number of
activations
72 (43-100) 90 (31-160) W= 77 (p=0.607)
Total potential
extinction activations
21 (9-29) 31 (8-47) W= 83.5 (p=0.181)
Proportion of
predictive gazes
0.46 (0.34-0.71) *0.51 (0.33-0.72) W= 50 (p=1.0)
Proportion of reactive
gazes
0.37 (0.08-0.54) *0.25 (0.17-0.63) W= 50 (p=1.0)
Proportion of non-
robot looks
0.09 (0.02-0.58) *0.18 (0-0.29) W= 52 (p=0.731)
Skewness for timing
of gaze on the
reinforcement
1.09 (0.56-2.42) *1.69 (0.11-2.4) W= 53 (p=0.628)
Median gaze time on
the reinforcement
(seconds)
-0.31 (-0.4-0.18) *-0.34 (-0.40- 0.75) W=45.5 (p=0.628)
Looking during
contingency phase
(seconds)
267.77 (195.9-
277.21)
*247.83 (152.88-
297.00)
W=44 (p=0.534)
Intertrial duration
(seconds)
6.53 (4.76-10.56) 5.31 (2.98-11.89)

W= 69 (p=0.776)

Exploratory analysis for behavioral patterns of gaze
As stated in the introduction of this chapter, we had a reasonable suspicion that infants
classified as classically defined “learners” and “non-learners” are more similar than not. Given
this suspicion and the results presented prior, we explore the visual motor data to see if any
common behavioral trends were observed amongst the whole group. This analysis was
LEARNING CONTINGENCY PARADIGM IN INFANTS

29
performed to describe the patterns of visual motor behavior observed in the infants while
learning a contingency paradigm and to help guide future contingency learning studies and
questions about infant motor learning.
We plotted the timing and type of visual gazes on the robot activations (see Appendix 2);
as well as the total time spent looking in minute by minute blocks during the contingency phase
of the paradigm (see Appendix 3). We found that all infants exhibited a time point where they
displayed predictive gaze for the majority of reinforcements in one block for more than 5
activations (referred to as the predictive block from here on out; see Appendix 2). Eleven of the
thirteen infants showed a pattern where they increased the amount of robot activations after the
predictive block was identified (see Appendix 2). The other two infants did not exhibit this
increase in robot activations after the predictive block since their predictive block was identified
in the last block of time.
The next pattern that we observed was the trend in the other types of gazes that occurred
after the predictive block occurred. Of the 11 infants who had data after the predictive block, all
of them displayed a greater occurrence of reactive and non-robot looks after the predictive block.
For example, in L4
CR
(see Appendix 3), the predictive block occurs in minute 3. Then, the infant
exhibits a greater number of reactive and non-looks as the paradigm progresses.
Lastly, to explore the relationship between the time point where predictive gaze occurred
for the majority of the time and the amount of time spent looking at the robot, we plotted looking
duration in minute-by-minute blocks (Appendix 3). Seven of the 13 infants decreased the amount
of time spent looking as the contingency paradigm progressed (see Appendix 3). Four infants
maintained the amount of time spent looking at the robot after the predictive block. The last 2
infants as stated prior did not have an additional block to compare after the predictive block.
LEARNING CONTINGENCY PARADIGM IN INFANTS

30
Discussion
The aim of the study was to determine if infants who are classified as classically defined
learners of a contingency paradigm demonstrated visual anticipation while engaged with the
paradigm. According to the way that visual anticipation is defined, learners visual anticipated
greater than random chance (>33.33% of the time). Additionally, the descriptive data supports
that classically defined learners are timing their gaze to occur either close to or prior to the
robot’s activations. Therefore, the data from the learners suggest that learners are demonstrating
visual anticipation while engaged in the paradigm. Importantly, we do not conclude that
classically defined "learners” displayed visual anticipation and demonstrated learning, because
classically defined “non-learners” are exhibiting similar visual motor behaviors while engaged in
the paradigm. Therefore, we conclude that the classic definition of learning a contingency
paradigm is insufficient at classifying learning.
An important key finding of these data was that infants categorized as “learners” and
“non-learners” showed little to no differences for most of the variables analyzed. In fact, the only
variable that was significantly different between the two groups was the total number of potential
robot activations at baseline. Recent literature has begun to classify learners using a retention test
and drifted towards using the classical definition of learning (i.e. learning threshold = 1.5 times
the baseline behaviors) as a way to classify infants who perform or do not perform the task
within a single session (Sargent et al., 2018; Sargent, Reimann, Kubo, & Fetters, 2015; Sargent
et al., 2014). While the use of a retention or a transfer test has been long standing as a measure of
motor learning (Krakauer et al., 2019), the repurposing of 1.5 learning threshold as a classifier of
motor performance has several limitations.
LEARNING CONTINGENCY PARADIGM IN INFANTS

31
As pointed out in the results of this study, the use of the 1.5 learning threshold most
likely classifies infants incorrectly because there is nothing to learn in the baseline phase (where
performance was different) and “learners” and “non-learners” were behaving similarly in the
contingency portion of the paradigm. Using a retention or transfer paradigm would be more
preferred to classify learners and non-learners because a researcher would be able to compare a
performance period where learning is occurring (in this paradigm this would be the contingency
period) to a testing period where the infant is actively showing what they have learned or
transferred to a new situation. For example, in this study the majority of infants were activating
the robot most frequently towards the end of the contingency period. If an infant were to be
retested in the same contingency paradigm the next day and moved their right leg at a similar
frequency at the start of the paradigm, this would support that the infant retained an
understanding that their behavior activated the robot. Additionally, if we were to reinforce an
arm movement instead of a right leg in the proposed retesting, then we can see if the infant
transfers what they learned the other day from their leg movement to learning that their arm
movement activated the robot.
In our analysis of the whole sample and our descriptions of the individual data, we found
that the infants displayed visual anticipation more than 33.33% of the time while engaged in the
paradigm. Additionally, it appears that all the infants showed a period of time where they are
visually anticipating the outcomes of their behavior (referred to as the first predictive block).
These individual data potentially suggest that all of the infants display a time point where they
have learned the connection between their behavior and outcome of that behavior. Then the
infants continue to increase the production of the reinforced behavior, while displaying variable
types of gazes and decreasing the amount of time spent looking at the robot. This finding uses
LEARNING CONTINGENCY PARADIGM IN INFANTS

32
the assumption that evidence of predictive gaze occurring for the majority of the time suggests
that infants have learned a construct or concept (Gredebäck et al., 2010). This is also supported
by the fact that all of the infants display variable types of gazes after the predictive block is
identified and the majority decrease the amount of time spent looking at the robot.
Our findings regarding visual motor patterns can be related to the two behavioral
processes found in adult discovery learning. Discovery learning is the process of acquiring a
movement pattern necessary to achieve the goal for a novel motor task. The acquisition process
involves learning the motor strategy of “what” serial pattern of movements to make and the
tactics of “how” much to scale one’s movements in order to be successful (Brooks et al., 1995).
Participants are usually tasked with the challenge of manipulating a novel tool in order to move a
cursor to a target (Brooks et al., 1995) or pick up an object (Golenia, Schoemaker, Mouton, &
Bongers, 2014). As such, the participant must first search for possible solutions (i.e. the motor
strategy or “what” movement pattern) and then once found, exploit and refine the correct motor
strategy (i.e. learn the motor tactics or “how” to scale the movements) to become skilled at
accomplishing the goal (Brooks et al., 1995; Golenia et al., 2014).
We suggest that our results show two phases of contingency learning through visual
motor behavior, similar to those described in Sailer et al. (2005). The first phase is an
exploratory phase where infants are using visual attention (i.e. behaviors considered to involve
higher amounts cognitive effort) to discover what their behaviors can produce in the world. All
of our infants exhibited this though visually anticipating more than 5 robot activations during a
single 1-minute block.
Second, there is a skill refinement phase where infants are increasing their behaviors.
Eleven of our infants demonstrated this stage of learning.
LEARNING CONTINGENCY PARADIGM IN INFANTS

33
While further assessment of these proposed learning phases of infant contingency
learning are warranted; we suggest that they are plausible given our data and past literature in
adult motor learning. Additionally, these proposed learning phases can be used to explain
performance results in retention and transfer tests for future contingency learning literature.

Limitations
The four limitations of this study pertain to sample size, our definitions of visual
anticipation, our use of the Wilcoxon Signed Ranked test, and the fact that we do not have a
retention or a transfer paradigm in the study. Given the small sample of classically defined
learners and non-learners, the finding that the two groups are not significantly different, aside
from the baseline movements, should be replicated in a further study or analyzed using a
systematic review of past contingency learning studies. However, the controversy of using the
1.5 learning threshold is prevalent in the literature since it is proposed to not reflect an
understanding of self-agency (Kelso, 2016; Kelso & Fuchs, 2016).
Our definition of visual anticipation has one design flaw. In our study, if the infant
continues to look at the robot after its activation and then activates the robot again; the infant will
be scored as having a visual anticipation. One may argue that this is not visual anticipation since
the infant is simply staring at a prior location that an event happened to see if it happens again.
This is a valid argument, however in the context of infant cognitive assessments, this is how
visual anticipation is scored. In the Bayley Scales of Infant and Toddler Development (Bayley &
Alyward, 2019), an assessor identifies if a child has anticipatory gaze by engaging in a game of
peek-a-boo with the child. The assessor makes themselves hidden to the child and reappears
twice to the same location. Then the assessor changes the placement of their head to see if the
LEARNING CONTINGENCY PARADIGM IN INFANTS

34
child assumes that they will move to the prior location. If the child looks or maintains their gaze
in the prior region of interest, then they are said to have visual anticipation skills. This example
is like our paradigm where the infant is looking at a similar region of interest to see if the robot
will activate again.
It could be argued that non-robot looks are time points where the infant is not engaged in
the learning paradigm and therefore should be removed in order to assesses if the infant is
visually anticipating the robot more than random chance (i.e. during the time that the infant is
engaged in the paradigm are they visual anticipating the consequences of their actions).
Therefore, a t-test that assesses if infants use predictive more than reactive gazes would be better
for determining if infants demonstrate anticipatory gaze and understand the contention between
their behavior and the reinforcement. This argument is valid only if the skill is already
considered to be a learned skill because during a learned action, you would expect a performer to
be able to visually anticipate the consequences of their actions most of the time if the skill
required visual anticipation (Sailer et al., 2005). However, during the acquisition of new skill,
you would expect the performer to be exploring multiple behaviors to better understand what
behaviors give the desired outcome and what additional behaviors are allowed (Piek, 2002;
Thelen, Kelso, & Fogel, 1987; Ulrich, 2010). Variability in movement and visual behavior
allows for the performer, during the learning process, to explore opportunities that can be used in
future situations. For example, non-robot looks can be seen as a chance for infants to explore if
socializing with caregiver or another willing participant is allowed in this paradigm.
Therefore, we find that the Wilcoxon Signed Ranked test chosen for our data set
describes and assesses the occurrence of the three types of gazes seen in this paradigm without
eliminating any data that could be important to the learning process. However, it should be
LEARNING CONTINGENCY PARADIGM IN INFANTS

35
mentioned that there is no motor literature in the adult and pediatric realm that describes how
accurate humans are at visually anticipating a movement that is being learned for the first time as
it was here. This information may be pertinent for understanding how humans use visual
behavior to learn novel movements and visual cueing can be used to aid early intervention.
The last limitation is that this study did not use an assessment of learning like a transfer
task or retention test to assess and classify classically defined learners and non-learners. Further
assessment is needed to see how infants demonstrate visual anticipation in a retention or transfer
test to see if they are better at predicting the outcomes of their actions once they have practiced
it.

Conclusion
In summary, this study presents a novel way to examine infant motor learning in the
context of visual motor behavior. We present evidence that while infants engage in a
contingency learning paradigm; they are demonstrating visual anticipation, potentially, as
evidence of learning the connection between their movement and the outcome for that
movement. Additionally, we hypothesis the existence of two learning stages that describe how
infants utilized their visual motor behaviors to learn a contingency learning paradigm. Although
this finding needs further exploration, it potentially indicates that pediatric clinicians should
focus on establishing a behavioral connection during the exploratory phases of learning (i.e.
helping the infant understand general movement pattern through visual cueing, modeling, and
eliciting the behavior of interest) and then focus on repetition and variability in the refinement
phases of learning the movement (i.e. encouraging the infant to move and practice the skill).
Last, the infant should be encouraged to engage in other activities that can aid the child’s
LEARNING CONTINGENCY PARADIGM IN INFANTS

36
development while mastering the skill in the mastery phase of learning. While these clinical
implications are in line with motor learning literature, it should be noted that further evaluation
using a later assessment of learning (i.e. a retention or transfer task) is warranted in order to
better understand infant visual motor learning and effective intervention strategies.

LEARNING CONTINGENCY PARADIGM IN INFANTS

37
CHAPTER III: BEHAVIORAL DIFFERENCES BETWEEN INFANTS AT
COMMUNITY AND ELEVATED RISK FOR AUTISM DURING A CONTINGENCY
PARADIGM

Abstract
Background: There is a growing literature that describes infants at elevated risk for autism and
those who are later diagnosed with autism as having delayed motor skill development.
Additionally, this population has been described as having atypical motor control and visual
behaviors. Although these descriptions exist, we do not know the rationale for early motor
problems.
Aims: The aim of this study was to determine and describe the behavioral differences between
infants at community and elevated risk level of ASD during a contingency learning paradigm.
Methods: Twenty-six full-term infants (6-9 months of age) at community risk (n=15) (i.e. no
first or second degree relatives diagnosed with ASD) and at elevated risk for autism (n=11)
participated in a contingency learning paradigm. A robot provided reinforcement when the infant
produced a right leg movement. Behavioral differences in looking behaviors, robot activations,
and overall learning were examined between the two groups.
Results: Overall, there were no statically significant differences between infants at elevated risk
and community risk for autism. Our visual and motor behavioral data suggest no discernable
group differences in terms of overall looking duration, anticipatory gazes, number of robot
activations, and average intertrial duration. However, we did notice that three of the infants at
elevated risk displayed visual motor patterns that were descriptively different from the whole
group.
LEARNING CONTINGENCY PARADIGM IN INFANTS

38
Conclusion: Our data suggest that infants at community and elevated risk for autism are using
similar behaviors to learn a contingency learning task. However, our descriptive analysis of the
behavioral learning processes warrants follow-up by analyzing the data when diagnosis status is
known.

LEARNING CONTINGENCY PARADIGM IN INFANTS

39
Introduction
Infants at elevated risk of autism (ER) have been described in the literature as having
delayed motor development (West, 2018), and atypical motor control (Sacrey et al., 2018) and
visual behavior (Elsabbagh et al., 2013; Jones & Klin, 2013). Focusing on motor development, it
is hypothesized that older children on the autism spectrum have difficulties with motor learning
(Moraes et al., 2017) and that these difficulties begin in infancy (West, 2018). However, most of
the research that examines motor delays in infants at ER only describes their motor milestone
achievement or overall scores on a motor assessment (Sacrey et al., 2018; West, 2018). Although
these tools are valuable for getting a general sense of motor delays, they do not provide
information on the behavioral rationale for why these differences exist.
The literature that focuses on motor control in infancy has shown that infants at ER have
impairments in balancing while seated (Kyvelidou et al., 2021) and reaching (Sacrey et al., 2018)
once these skills are developed. Additionally, in a complex motor reaching task, Ekberg et al.
(2016) found that infants at ER initiated the timing of their reaches towards a rolling ball later
compared to those not at increased risk. Infants at ER attempted to begin the reach later and
hence missed contacting the ball, while infants not at increased risk were able to grab the ball by
timing their reaches appropriately. All together, these papers suggest that infants at ER have
motor impairments in controlling movements that require prospective control and anticipating
events in the environment.
Overall, the literature suggests that infants at ER have difficulties acquiring motor
milestones at the average age and controlling them once they are learned. However, it is unclear
if infants at ER have difficulties learning motor skills or if they use different behavioral
mechanisms to learn motor skills. An assessment of motor learning in this population will aid in
LEARNING CONTINGENCY PARADIGM IN INFANTS

40
the development of effective early interventions that target specific areas of motor skill
acquisition. Additionally, an evaluation of motor skills can lead to future hypotheses that aim to
explain the rationales for difficulties in motor milestone achievement.
In the current study, we used a contingency learning paradigm to compare infant motor
learning between those at ER and community risk (CR) for autism. We used head mounted eye-
tracking, wearable sensors, and video data to evaluate the behavioral differences between these
two groups. Our hypotheses were that, compared to infants at CR 1) infants at ER will be less
likely to learn the paradigm, 2) infants at ER will exhibit a lower proportion of predictive gaze
on the reinforcement, number of reinforcements produced, and looking duration on the
reinforcement agent, and 3) infants at ER will exhibit longer intertrial durations between
reinforcements. We also performed an additional analysis to see if the infants at ER exhibited a
similar pattern of visual motor behavior that was seen in Chapter II.

Methods
Participants
Twenty-nine infants at elevated (n=12) and community risk (n=17) for autism were
enrolled in the study. They were recruited by fliers, online postings, and word of mouth in the
greater Los Angeles area between 2021 and 2023. The data for the infants in the community risk
group were presented in Chapter II of this dissertation. The inclusion criteria were that the infant
was born full term (>37 weeks gestational age) and between the age of 6 and 9 months old.
Infants with any known visual, hearing, or orthopedic impairments were excluded from the
study. Infants with a first (i.e. sibling or parent) or second degree (i.e. aunt, uncle, grandparent,
or step sibling) relative diagnosed with autism was determined to be in the elevated risk for
LEARNING CONTINGENCY PARADIGM IN INFANTS

41
autism group and those without a relative diagnosed with autism were in the community risk
group. Lastly, infants were excluded from the data set if they scored lower than the fifth
percentile on the Bayley Scales of Infant and Toddler Development (fourth edition) for the
average of all three domains tested: Cognitive, Language, and Motor (Bayley & Alyward, 2019)
or cried continuously for 1 minute of the contingency paradigm.
Three infants were excluded from our recruited sample and our final sample had 26
infants, 15 at CR and 11 at ER. One infant at CR was unable to participate in the contingency
paradigm due to a technical error with our robot and their data were excluded. Two other infants,
1 at ER and 1 at CR, were excluded from the data set due to crying continuously for a full minute
of the contingency paradigm. Lastly, 4 participants (2 at ER and 2 at CR) did not tolerate the eye
tracker but were able to participate in the contingency learning paradigm without eye gaze data.
Therefore, our eye gaze data sample comprises of 9 infants at ER and 13 at CR. Participant
characteristics are in Table 3.1.

LEARNING CONTINGENCY PARADIGM IN INFANTS

42
Table 3.1. Participant characteristics for all infants (Median (Range)).
Variable Community Risk (n=15) Elevated Risk (n=11) W, p
Age (days) 210 (183-267) 203 (191-233) 218, 0.443
Sex 7 male, 8 female 7 male, 4 female
Weight (kg) 7.5 (6.6-8.7) 8.51 (7.25-11.1) 147, 0.003
Bayley-4 Cognitive,
Percentiles
75 (50-91) 75 (50-84) 206.5, 0.838
Bayley-4 Language,
Percentiles
63 (37-87) 63 (37-82) 208, 0.799
Bayley-4 Motor,
Percentiles
73 (15-87) 66 (10-86) 208, 0.799
Ethnicity 6 Hispanic/Latino, 9 Not
Hispanic/Latino
8 Hispanic/Latino, 4 Not
Hispanic/Latino

Race 3 Asian, 3 Black/African
American, 2 White, 6
Other/Multi-Racial, 1
Declined to answer
1 Asian, 0 Black/African
American, 5 White, 4
Other/Multi-Racial, 2
Declined to answer

*Bayley-4: Bayley Scales of Infant and Toddler Development, fourth edition.

Procedures
The research was approved by the Institutional Review Board of the University of
Southern California (HS-14-00911). A parent or legal guardian signed an informed consent form
before their infant’s participation in the study. Data were collected at Children’s Hospital Los
Angeles. At each data collection, we collected the infant’s anthropometric data (thigh length and
circumference, shank length and circumference, foot length and width, and weight), assessed
their motor, cognitive, and language development using the Bayley Scales of Infant and Toddler
Development (version 4) (Bayley-4), and had their parents fill out the First Year Inventory (FYI
LEARNING CONTINGENCY PARADIGM IN INFANTS

43
version 3.1) (Baranek, G., Watson, L. R., Crais, E., Turner-Brown, L., & Reznik, 2013). The
Bayley-4 and FYI v3.1 were described prior in chapter II.
The same contingency paradigm that was explained in Chapter II was used in this study.
Briefly, infants were supported by a chair in front of an infant-sized humanoid socially assistive
NAO robot (Aldebaran United Robotics Group). Participants engaged in a twelve-minute
contingency paradigm where movements of their right leg were reinforced by a robot (seen in
Figure 2.1) that would produce a clapping motion and a laughing sound. The structure of the
paradigm was a 2-minute baseline, a 10-second demonstration, an 8-minute contingency
condition, and a 2-minute extinction phase (see Figure 2.2). Infant right leg movements were not
reinforced during the baseline, demonstration, and extinction periods. The infants right leg
movements were reinforced when they produced an acceleration above 3.0 m/s
2
, which is the
same as Fitter et al. (2019).
During the contingency paradigm, infants wore a head mounted eye tracker (Positive
Science) on their heads and four wearable Opal sensors (APDM Inc) that were placed on the
arms and legs (one for each limb). Eye gaze data were recorded at 30 frames per second and
wearable sensor data were collected at 128 Hz. A 5-point calibration was performed for the eye
gaze data and was process post data collection. The 5-point calibration was performed at a
distance between the seated infant and the robot (1.5 meters).
Video data included two cameras that were placed externally in the contingency
paradigm space. One camera was behind the robot and recoding the infant in their chair; the
second camera was placed on the side of the infant and the robot and was recording both the
infant at the robot from a side view. Prior to the start of the contingency and after the 5-point
calibration, the spinning globe toy was turned on and off repeatedly 3 times and simultaneously
LEARNING CONTINGENCY PARADIGM IN INFANTS

44
shown to the two external cameras and to the eye tracker. This was performed to synchronize all
video data in our post data collection processing.

Data Preparation
After the data collection, data from the eye tacker were imported into the software
Yarbus and were calibrated and processed. Given that the eye tracker has an error of a 4-degree
radius, a graphic overlay of a 4-degree radius target was imported to the video file and
represented the approximate area that the infant was looking at. The overall result is a single
video clip of the whole contingency paradigm that displays a target of where the infant is
looking. A detailed description was explained in Chapter II of this dissertation.
Once the eye gaze video was created, it was synchronized with the side view cameras
video using ELAN software (ELAN 5.8, Language Archive). The common starting point that
was used to synchronize the two videos was the point where we flashed the globe toy 3 times.
Additionally, a custom Python software was used to identify each time the robot activated
according to the video data’s timeline. The timing for each robot activation was confirmed in our
frame-by-frame analysis discussed next.
Trained video coders coded the synchronized video files using a frame-by-frame analysis.
Coders were trained on testing data sets and had to reach a reliability above 80% on their video
files before reducing the data presented in this study. Coders coded for behavioral state and
visual behavior and one third of their videos were assessed for reliability. Percent of agreement
for behavioral state was 94.3% types of visual gaze was 87.2%, and time spent looking at the
robot was 95.7%. Behavioral state was coded as sleeping, drowsy, alert, fussy, or crying (Lester
& Tronick, 2004) throughout the whole contingency paradigm.
LEARNING CONTINGENCY PARADIGM IN INFANTS

45
Table 3.2. Median percent (range) for each behavioral state and condition for all infants and
groups (CR= Community Risk, ER= Elevated Risk).
Baseline Contingency Extinction
State CR ER CR ER CR ER
Alert 100
(49.8-100)
100
(88.6-100)
100
(74.0-100)
100
(94.6-100)
*100
(31.5-100)
*95.5
(15.4-100)
Fussy 0
(0-50.2)
0
(0-11.4)
0
(0-19.7)
0
(0-6.2)
*0
(0-27.3)
*4.5
(0-57.3)
Crying 0
(0-10.18)
0
(0-0)
0
(0-8.57)
0
(0-0)
0
(0-41.2)
0
(0-52.9)
*Significantly different according to a pairwise comparisons using Wilcoxon-Ranked Sum test
P<0.5
We used the same visual behavior coding scheme that was described in Chapter II of this
dissertation. Briefly, the visual behaviors that were coded for included: the timing and duration
of each gaze on the robot and each instance of predictive and reactive gaze on a robot’s
activation. Predictive gazes were defined as a visual fixation (3 more frames of no eye
movement) (Franchak et al., 2011; Gredebäck & Falck-Ytter, 2015) on the robot prior to its
activation and no earlier than 12 frames prior to the robot’s activation (Wass et al., 2015). In
instances that no predictive gaze occurred, a reactive gaze (Gredebäck & Falck-Ytter, 2015) was
defined as a visual fixation on the robot during its activation (60 frames); and a non-robot look
was defined as no look occurring on the robot. For a more detailed description of the visual
behavior coding scheme, please refer to Chapter II of this dissertation.

LEARNING CONTINGENCY PARADIGM IN INFANTS

46
Data Reduction
Determining the classical definition learning based on leg movement
A learning threshold was determined based on the prior literature where an infant is
classified as a classically defined learner if they produce the behavior that is reinforced 1.5 times
more during the contingency phase compared to the baseline. Potential activations during the
baseline had the same activation threshold of 3.0 m/s
s
and had to occur more than 2 seconds after
the last potential activation (i.e. the duration of a single robot activation). Using custom
MATLAB software, the number of times an infant would have been reinforced during the
baseline was calculated using the wearable sensor data from the right leg sensor and then
multiplied by 1.5 to calculate the learning threshold. If an infant activated the robot more times
than the learning threshold in a two-minute moving window block during the contingency phase,
they were classified as a classically defined learner.

Timing of gazes on each robot activation
To calculate the timing of the gazes on each robot activation, the start of each predictive
or reactive gaze was subtracted from the start of the robot activation. This means that negative
values were predictive and positive values were reactive.

Proportion of Gazes
To calculate the proportion of each type of gaze on the robot, the frequency of each gaze
was counted for each infant (i.e. predictive, reactive, and none gaze count) and then was divided
by the total number of robot activations.

LEARNING CONTINGENCY PARADIGM IN INFANTS

47
Looking duration
The total amount of time looking at the reinforcement agent was calculated by adding
together the durations of each time the infant looked at the robot during the whole contingency
paradigm. We will present the amount of looking during the baseline, contingency, and
extinction phases of the paradigm.

Intertrial duration for robot activations
Intertrial duration was calculated for each robot activation by calculating duration of time
from activation to the next activation. For example, if activation 1 happened at 20 seconds of the
contingency and then activation 2 happened at 40 seconds of the contingency, then the intertrial
duration would be 20 seconds. The average for each infant was calculated and average intertrial
duration for each group is reported.

Statistical Analysis
Non-Parametric tests and descriptive statistics were used given the small sample size and
because most variables were not normally distributed. All computations for frequency and
proportion of each type of gaze, classification of learning, and timing of gazes on the
reinforcement were computed using custom MATLAB programs and exported to be analyzed in
SPSS (v.27). The following descriptive statistics were compared using a two-sided Wilcoxon
Rank Sum test to test for differences between infants at ER and CR: behavioral state, age,
weight, and Bayley percentiles for the cognitive, language, and motor scales.
Our first hypothesis was that infants at ER would be less likely to be classified as learners
of the contingency paradigm compared to infants at CR. We used a Chi-squared test to compare
LEARNING CONTINGENCY PARADIGM IN INFANTS

48
the proportion of learners in each group. Cohen’s W was used to describe the effect size and used
to calculate the needed sample size to determine a statistical difference. We used the following
values for the Cohen’s W to describe the effect size: less than 0.1 is a low effect, greater than 0.3
is a moderate effect, greater than 0.5 is a large effect (Cohen, 1988). Additionally, we used
Spearman correlations to determine if being classically defined as a learner or non-learner was
associated with the scales from the Bayley-4 (cognitive, language, and motor scales) using
percentiles scores.
Our second aim was to determine if infants at ER would exhibit a lower proportion of
predictive gazes on the reinforcement, number of reinforcements produced, and looking duration
on the reinforcement agent. Additionally, our third hypothesis was that infants at ER would
exhibit longer intertrial durations between reinforcements. We used a two-sided Wilcoxon Rank
Sum test with Bonferroni adjustment to compare the infants at ER and CR. Effect sizes were also
calculated for each comparison using Pearson’s r with the following explanations for r: small
effect was |r| > 0.1, moderate effect |r| > 0.3, and large effect |r| > 0.5 (Cohen, 1988).
To better examine evidence of autistic traits in our sample, we examined 5 items from the
Bayley-4 chosen from the list of items that show early signs of autistic traits, as well as all 69
items from the FYI v3.1. The items selected from the Bayley-4 were developmentally
appropriate for the age ranges assessed in the study (Bayley & Alyward, 2019) and were as
follows: cognitive items 16 (interacts with mirror) and 20 (pats table when assessor pats the
table), receptive communication item 12 (responds to name), and expressive communication
items 2 (displays social smiles) and 4 (displays social vocalization). Absence of these items
(score equal to 0), indicate the possibility of an autistic trait. We used a chi-squared test to test
for an association between each Bayley item and group membership (i.e. ER vs. CR).
LEARNING CONTINGENCY PARADIGM IN INFANTS

49
For the data from the FYI v3.1, we computed the factor scores (Seven factors: 1-
communication, imitation, and play; 2- social attention and affective engagement; 3- sensory
hyperresponsiveness; 4- sensory hyporesponsiveness; 5- self-regulation in daily routines; 6-
sensory interest, repetitions, and seeking behaviors; and 7- motor coordination and milestones)
for each infant using the methods described by Baranek et al (2022). We combined our data with
an aged-matched sample (all aged 6 to 7.99 months) from Baranek et al (2022) and applied their
model to combined data; while the model was structured to match Baranek et al. (2022), all
model parameters were freely estimated. Factor scores were estimated for each infant in our data
set with values being centered at zero and positive values indicated greater risk of autistic traits.
Higher scores on the factors indicate more autistic features. Then using a two-sided Wilcoxon
Rank Sum test, differences between each group (i.e. ER and CR) were assessed using the
computed factor score. We also descriptively compared each group’s (i.e. ER and CR) factor
scores with the averages from the aged-matched sample in Baranek et al (2022). The age
matched sample (n=53) were infants (mean age in months (standard deviation) = 7.28 (0.43)
without evidence of a later diagnosis of ASD or other developmental disability by age 3 years.
Lastly, to explore the individual data further, we plotted the number of gazes, types of
gazes, and duration spent looking at the robot in minute blocks for the contingency phase. This
was done as an exploratory analysis to see how the infants at ER used their visual motor
behavior to perform the paradigm. Specifically, we wanted to see if infant at ER exhibited the
same visual motor behaviors shown in Chapter II of this dissertation. Infants who did not show
the patterns of visual motor behavior discussed in Chapter II were described further using their
scores and individual items on the Bayley-4 assessment, their FYI v3.1 factor scores, and
performance variables on the contingency learning paradigm.
LEARNING CONTINGENCY PARADIGM IN INFANTS

50
Results
Learning amongst group
For our first hypothesis, the proportion of infants who were classified as classically
defined learners and non-learners amongst the infants at ER (proportion of learners= 0.18) and
CR (proportion of learners= 0.4) were not significantly different according to a chi-squared test
(c
2
=1.42, p=0.234). The effect size for the chi-squared test had low to moderate effect (Cohen’s
w= 0.233). Results for the classification of learning amongst our group can be seen in Table 3.3.
Additionally using Spearman’s correlations, no statically significant associations were
found between learning classification and cognitive (r= 0.27, p= 0.18), language (r= -0.01,
p=0.98), and motor percentiles (r= 0.29, p= 0.16) from the Bayley-4; and between learning
classification and group membership (i.e. ER vs. CR) (r=0.23, p=0.25).

Table 3.3. Contingency table depicting the number of classically defined learners and non-
learners in each group
CR = community risk, ER = elevated risk.

CR ER Total
Learners 6 2 8
Non-learners 9 9 18
Total 15 11 26

Amount of predictive gaze, reinforcements, looking duration and intertrial duration
For our second and third hypotheses, we found no significant differences in the visual
motor behavior between the infant at ER and CR. The number of activations during the
LEARNING CONTINGENCY PARADIGM IN INFANTS

51
contingency paradigm (p=0.683, no effect- Pearson’s r=0.08) and the potential number of
activations during the baseline (p=0.84, no effect- Pearson’s r=0.045) and extinction phase of the
paradigm (p=0.84, small effect- Pearson’s r=0.2) were similar. Additionally, infants in both
groups exhibited a similar proportion of visual anticipations (p=0.324, small effect- Pearson’s
r=0.21), reactions (p=0.471, small effect- Pearson’s r=0.16), and non-robot looks (p=0.421, small
effect- Pearson’s r=0.17) during the paradigm; as well as similar duration of looking at the robot
during all phases of the paradigm (i.e. the baseline, contingency, and extinction phase) (Table
3.4). Lastly, infants at ER and CR exhibited a similar median timing of gaze on the robot’s
activations (p=0.262, small effect- Pearson’s r=0.2) as well as similar intertrial durations
(p=0.683, no effect- Pearson’s r=0.08). The medians and ranges for all variables can be seen in
Table 3.4.

LEARNING CONTINGENCY PARADIGM IN INFANTS

52
Table 3.4. Amount of reinforcements, predictive gazes, duration of looking, and intertrial
duration. Median (range) are presented in the table.
Variable CR (n=15) ER (n=11) W, (p)
Potential Activations during
Baseline
18 (5-42) 27 (8-31) 153, (0.84)
Peak Contingency Block 32 (13-48) 34 (13-46) 155, (0.721)
Potential Activations during
Extinction
23 (8-47) 30 (13-44) 168, (0.305)
Reinforcements* 77 (31-160) 81 (32-144) 157, (0.683)
Proportion of Predictive Gazes* 0.47 (0.33-0.72) 0.42 (0.21-0.71) 88.5, (0.324)
Looking-Baseline (s) 17.67 (2.39-
68.10)
20.29 (6.22-27.44) 104, (1.00)
Looking-Contingency (s) 255.93 (152.88-
297)
267.34 (154.91-
307.25)
115.5, (0.431)
Looking-Extinction (s) 7.89 (0.00-43.27) 10.86 (0.00-26.66) 70.5, (0.431)
Median Gaze Time -0.31 (-0.4- 0.75) -0.24 (-0.4- 0.54) 120.5, (0.262)
Average Intertrial Duration (s) 5.96 (2.98-4.87) 5.7 (3.31-14.02) 140, (0.683)
* Count data. Note: Eye gaze data was not collected on 2 infants from each group due to
compliance with the eye tracker. The sample size for eye gaze variables is 9 for ER and 13 for
CR.

LEARNING CONTINGENCY PARADIGM IN INFANTS

53
Evidence of autistic traits amongst the community and elevated risk for autism groups
No association was found between the selected Bayley items (items that suggest autistic
traits) and the two groups (i.e. CR and ER). However, cognitive item 20 (mimicking pats of a
table) had a trend towards significance where infants at elevated risk appear to score lower than
infants at community risk on mimicking pats of a table (c
2
=5.190; p=0.07).
The caregivers for 9 infants at CR and 7 infants at ER filled out the FYI v3.1. No
significant differences were found between infants at ER and CR for all the factors of the FYI
v3.1 (1-communication, imitation and play (p=0.21); 2- social attention and affective
engagement (p=1.0); 3- sensory hyperresponsiveness (p=0.84); 4- sensory hyporesponsiveness
(p=0.92); 5- self-regulation in daily routines (p=0.76); 6- sensory interest, repetitions, and
seeking behaviors (p=0.47); and 7- motor coordination and milestones (p=0.41)) (see Table 3.5).
Additionally, the average factor scores from each group in our data set (i.e. CR and ER) had
similar values to those from the aged-matched sample from Baranek et al. 2022.

LEARNING CONTINGENCY PARADIGM IN INFANTS

54
Table 3.5. FYI v3.1 factors from the study sample groups (CR and ER) and an aged-matched
sample from Baranek et al. (2022) mean (standard deviation). P-values are for the comparison
between the CR and ER groups from the study.
FYI Factor CR sample
(n=9)
ER sample
(n=7)
Age-matched
sample from
Baranek et al. 2022
(n=53)
P-value
Communication, imitation,
and play
-0.04 (0.74) 0.41 (0.71) -0.08 (1.00) p=0.21
Social attention and affective
engagement
0.07 (0.72) 0.11 (0.52) -0.06 (1.00) p=1.0
Sensory hyperresponsiveness -0.67 (1.1) -0.94 (0.42) 0.23 (0.91) p=0.84
Sensory hyporesponsiveness 0.14 (0.49) 0.10 (0.45) -0.06 (1.02) p=0.92
Regulation in daily routines -0.21 (0.75) -0.09 (0.77) -0.04 (0.89) p=0.76
Sensory interest, repetitions,
and seeking behaviors
-0.52 (0.83) -0.66 (0.43) 0.16 (0.94) p=0.47
Motor coordination and
milestones
-0.02 (0.74) 0.26 (0.86) -0.06 (0.98) p=0.41

Visual motor patterns of learning in infants at elevated risk
In Chapter II of this dissertation, we performed a visual analysis that looked at the
patterns of visual motor behavior infants exhibited while engaged in the paradigm. We perform
the same analysis here for the infants at ER to see if they exhibited the same patterns.
LEARNING CONTINGENCY PARADIGM IN INFANTS

55
Overall, most of the infants at ER exhibited similar patterns in their visual motor
behavior (6 of the 9 infants) (see Appendix 2). These 6 infants displayed a time point where they
visually anticipated the majority of the activations in a minute block (defined as the predictive
block in Chapter II) and showed greater variability in the types of gazes that occurred after this
time point. Additionally, 5 of the infants decreased the amount of time spent looking at the robot
after the predictive block.
However, there were 3 infants at ER who did not exhibit the predictive block that we
described in Chapter II. Infants NL2, NL5, and L2 were identified as the infants who did not
exhibit the predictive block (see Appendix 2). Infants NL2 and NL5 increased their activations
of the robot without visually anticipating the robot activations and Infant L2’s data does not have
a recognizable pattern in their behavioral data.
Individual data from three infants’ data were explored further using the average data of
all the infants. Infants NL2, NL5, and L2 displayed lower amounts of looking at the robot during
the contingency phase and a greater proportion of non-robot and reactive looks towards their
activation of the robot (Table 3.5). Additionally, L2 activated the robot a lower number of times
compared to all of the infants in the sample.
Regarding these 3 infants, one of these infants had a low language percentile on the
Bayley-4 (L2). Additionally, two of the infants (NL2 and 5) did not display cognitive item 20
(pats table when assessor pats the table) and all three (NL2 and 5, and L2) did not display or
showed some ability to perform (i.e. a score of a 1 on the Bayley item) receptive item 12
(responds to name) (see Table 3.7 for further details). However, all three of these infants
received a score of a 2 on the other 3 items that were selected to show early evidence of autistic
traits (i.e. they did not show autistic traits because they fully passed these items).
LEARNING CONTINGENCY PARADIGM IN INFANTS

56
Lastly, two of the infants who were identified from the contingency data had caregivers
filled out the FYI v3.1. Infant NL2 had factor scores that were lower or similar to the Baranek et
al. 2022’s age-matched sample that were not diagnosed with autism or another developmental
disability by age 3 years, indicating that NL2 had a lower risk of autistic traits according to
parent report. Infant NL5, had higher factor scores for communication, imitation and play (factor
score= 1.18) and motor coordination and milestones (factor score= 1.28) when compared to the
aged-matched sample, indicating that NL5 had a greater risk of autistic traits according to parent
report for these two factors.

LEARNING CONTINGENCY PARADIGM IN INFANTS

57
Table 3.6. Average (standard deviation) behavioral measures from the contingency data for the
whole sample and three individuals at ER for autism. These three infants (NL2, NL5, and L2)
were identified to have different behavioral patterns during the paradigm.
Variable Infant Average (SD) NL 2 NL 5 L 2
Infant Robot
Activations
84 (34) 56 81 32
Proportion of Non-
Robot Looks
0.18 (0.15) 0.39 0.41 0.13
Proportion of
Reactive Gazes
0.35 (0.16) 0.39 0.37 0.56
Proportion of
Predictive Gazes
0.47 (0.17) 0.21 0.22 0.31
Median Timing of
Gaze on the
initiated Robot
activation (s)
-0.11 (0.35) 0.49 0.54 0.32
Looking Baseline
(s)
21.97 (17.10) 17.77 12.69 27.44
Looking
Contingency (s)
244.24 (46.40) 190.61 154.91 199.07
Looking Extinction
(s)
10.96 (10.4) 0 18.77 26.66
Proportion of Time
Alert Baseline
94.4 (13.2) 100 100 100
Proportion of Time
Alert Contingency
97.1 (6.6) 98.69

100 100
Proportion of Time
Alert Extinction
86.8 (25.1) 62.4 100 100

LEARNING CONTINGENCY PARADIGM IN INFANTS

58
Table 3.7. Individual data from the Bayley-4 from three Infants at ER. For the Bayley-4 items, a
score of a 0= item not seen, 1= somewhat displayed, 2= the infant performed the items. These
Bayley-4 items are described in the text in further detail.

Infant NL 2 NL 5 L 2
Reason for increased
risk of ASD
Infant sibling Second degree
relative
Infant sibling
Age (Months) 6.84 6.28 7
Bayley-4 Cognitive
Percentile
75 84 63
Bayley-4 Language
Percentile
50 63 37
Bayley-4 Motor
Percentile
66 82 58
Cognitive Item 20 0 0 2
Receptive Language
item 12
0 1 0

Discussion
None of the hypotheses stated in this study were supported. The proportion of classically
defined learners and non-learners were similar amongst the infants at CR and ER (hypothesis 1);
and both groups displayed similar visual motor behaviors while engaged in the paradigm
(hypotheses 2 and 3). Therefore, these results suggest that infants at ER and CR display similar
behaviors while learning a contingency learning paradigm between the ages of 6 and 9 months of
age. However, while group comparisons were not significantly different, exploratory
descriptions of visual motor behaviors during the paradigm and individual assessment data
support questions for further study.
As stated, prior, literary evidence suggest that infants who go on to be diagnosed with
autism exhibit lower performance in visual behavioral (Elsabbagh et al., 2013; Jones & Klin,
2013) and show evidence of motor delays (Bhat, 2020; Johnson et al., 2015; Ozonoff et al.,
2008; West, 2018). These literary descriptions led us to the hypothesis that infants at elevated
LEARNING CONTINGENCY PARADIGM IN INFANTS

59
risk for autism would show behavioral differences when trying to learn a contingency paradigm.
However, the lack of group differences could be due to two reasons.
One possible explanation is that it is possible that our sample contains very few or no
infants who will go on to be diagnosed with autism. When comparing the data from our
assessment of autistic traits, the sample had no statistical differences at the group level for the
Bayley-4 and the FYI v3.1. Additionally, the parent-report data form the FYI v3.1 suggests that
on the whole, the two groups of infants from the study were not exhibiting high levels of autistic
traits compared to the matched sample from Baranek et al. (2022). It should be noted that having
a sibling with autism increases the likelihood of being diagnosed with autism, but this increases
likelihood is 18% (Gibbs, 2021; Ozonoff et al., 2011). Thus, out of our sample of 11 infants at
ER, we anticipate that about 2 infants may go on to receive a diagnosis of autism and it is
possible that none of our infants will go on to receive a diagnosis of autism.
A more interesting point to discuss about our results is that the motor learning paradigm
that we used was potentially within the capabilities of all infants regardless of their future
diagnosis. Infants at ER who go on to be diagnosed with autism compared to those who are not
diagnosed have been shown to display higher amount of physical activity, as expressed through
leg movement, during free play and structured activates (Reetzke et al., 2022). Therefore, a
motor learning paradigm that requires infants to move their legs to activate a reinforcement is
potentially an easy task for them given that the population tends to engage in more physical
activity during infancy.
Additionally, the literature stating that motor delays exist due to statistically significant
differences is much different from claiming delays are occurring outside the ranges of typical
development. In many of the early development studies, results that support the claim of motor
LEARNING CONTINGENCY PARADIGM IN INFANTS

60
delays in autism are claimed based off group comparisons and statistically significant group
differences (Esposito, Venuti, Maestro, & Muratori, 2009; Flanagan, Landa, Bhat, & Bauman,
2012; Landa & Garrett-Mayer, 2006; Ozonoff et al., 2008; West, 2018). However, it appears that
prior to 24 months of age motor skills are still occurring within the typical range of development
even for infants at elevated risk for autism (Landa & Garrett-Mayer, 2006). Therefore, the motor
paradigm that was chosen for this study is potentially too simple to detect differences and a more
difficult motor learning task is needed for future studies about motor learning in infants at
elevated risk for autism.
Behavioral descriptions of motor control impairments in infants who go on to be
diagnosed with autism have been shown in paradigms that use quantifiable measures that are
difficult to assess by visual observation. For example, in Ekberg et al. (2016) researchers found
that infants who were later diagnosed with autism, compared to those who are not, had
difficulties timing their grasp of a ball that was moving down a ramp. Infants who were later
diagnosed with autism missed grabbing the ball by milliseconds and infants who were not
diagnosed were able to make contact with the ball. Therefore, an anlalysis of motor learning
behavior during an error based task may bear more distinctions between infant at CR and ER and
explain the potential difficulties in motor learning for children who develop autism.
Additionally, another endeavor that could highlight motor learning differences between
infants at elevated and community risk for autism would be to examine how infants use adaptive
behaviors to learn motor skills. Autistic children experience sensory information in variable
ways which results in the avoidance or seeking selective sensory information and experiences
(Gowen & Hamilton, 2013). Autistic traits regarding sensory processing potentially leads to
different motor adaptations and developmental profiles. A better understating of motor learning
LEARNING CONTINGENCY PARADIGM IN INFANTS

61
adaptations could aid in the development of better motor interventions of autistic individuals
with motor impairments.
In this study, we explored our indvidual data to see if infants at elvated risk for autism
displayed different patterns of visual motor behavior during the paradigm and then looked into
their assessment scores indepth. This analysis found that 6 of the infants displayed similar
patterns of visual motor behavior and 3 infants did not. Two of these infants (NL2 and NL5)
were potentially engaged in the paradigm using a non-looking adaptation (i.e. they displayed less
looking, greater proportion of non-robot looks, and a greater proportion of reactive looks) and
could have sought the sound of robot instead of looking at the robot. The other infant (L2) could
have been displaying an avoidence pattern since this infant did not activate the robot as much as
the other infants in the study, or potentially the infant just did not move very much and thus did
not activate the robot often. It should be noted that these three infants were in the alert beahvioral
state for the vast majority of the paradigm and showed few occurrences of fussiness and crying
while the robot was being activated by their movements.
While we did find that these three infants did score lower on one of the items (i.e.
congnitve item 20: imitation of pats on a table) linked to autisic traits in the Bayley-4, it should
be noted that not displaying one item on the Bayley-4 -does not mean these children have autism.
All of these children scored within the average ranges for the Bayley-4. Additionally, even
though the FYI v.3.1 data were only available for two of these three infants, one of these infants
did not show elevated risk of autistic traits on any of the seven factors, while the other had higher
scores for only two of the seven factors, as compared to the matched sample from Baranek et al.
2022. This may indicate the possiblity of increased risk of autism for one infant, but further
assessment would be warranted. Follow up is needed at age 3 years to determine whether any
LEARNING CONTINGENCY PARADIGM IN INFANTS

62
child receives a diagnosis of autism, to see if our contingency data findings for these 3 infants are
linked to motor learning in autism.

Limitations
We have covered the limitations regarding our ability to not know the final diagnosis of
autism in our sample and that our motor learning paradigm is potentially too easy for our
participants. However, it is possible that our study did lack power to find statistically significant
group differences due to our sample size. In Table 3.3, it appears that infants at elevated risk are
trending towards being less likely to be classified as classically defined learners of the motor
paradigm. We calculated that increasing our sample to 144 participants (72 participants per
group) would be needed to observe group differences in the classification of learners and non-
learners using the current paradigm.
It should also be noted that this small sample may not have any participants who will be
later diagnosed with autism or a developmental disability. The data from the FYI v3.1 and
Bayley-4 for the CR and ER groups, on the whole, suggest that these infants are not
demonstrating autistic traits and that they are functioning within the typical ranges for cognitive,
motor, and language development. Individual outliers require further study; and replication with
a larger sample and similar methods is needed to study motor learning in infants with prospective
follow-up to determine who will be diagnosed with autism by age 3.
We focused primarily on the differences in motor learning during the contingency portion
of the paradigm. Our statistical analysis of the behavioral state data showed that infants at ER
spent more time being fussy during the extinction phase of the paradigm. These results possibly
indicate that infants at elevated risk may have issues with self-regulation once something is
LEARNING CONTINGENCY PARADIGM IN INFANTS

63
learned and taken away. However, this concept is outside the scope of this paper and could
potentially be explored further.
Lastly, we did not examine evidence for the concept known as “sticky attention” once the
contingency paradigm was presumably learned by the infants. Given our results from Chapter II,
we argue that the classical definition of learning based only on change in movement rate between
the baseline and contingency periods should be abandoned, and other measures of motor learning
(e.g. visual anticipation, pupil dilation, a transfer task, or a retention test) should be incorporated.
Given that a retention or transfer paradigm has been shown to be difficult for autistic individuals
with average IQ scores (Gowen & Hamilton, 2013), using this assessment of learning might not
be appropriate in infants at elevated risk for autism. However, our analysis of motor learning
adaptations could highlight how infants who go on to be diagnosed with autism use behavioral
adaptations to learn motor skills. These concepts should be re-evaluated in future studies when
learning is best determined, to see if infants who have learned the contingency paradigm exhibit
sticky attention.

Conclusion
In summary, while our hypotheses were not supported in this study, it does offer
guidance for future studies that examine motor learning in infants at ER and CR for autism. In
terms of using a contingency learning paradigm, an analysis of motor learning adaptations could
aid in better understating of how infants at elevated risk for autism learn motor skills. Also, the
use of screening measures to assess for evidence of autistic traits should be used in follow up
studies. Our results suggest that neither the CR nor the ER group showed elevated of autistic
traits on behavioral measures (Bayley-4 items and FYI v3.1), which may explain why group
LEARNING CONTINGENCY PARADIGM IN INFANTS

64
differences were harder to detect on the experimental task of interest. Purposive sampling of
infants with elevated scores on screening measures could be useful in future studies. In fact,
given the variety and neurodiversity of symptoms in autism, the use of purposive sampling
would allow for a selection of individual that would benefit from findings that are geared
towards informing early interventions for the selected group.
At the individual level three infants used alternate visual motor behaviors while engaging
with the paradigm. Conceptually, this poses questions about the alternative motor learning
behaviors that children who go on to be diagnosed with autism might use. However, this
statement should be taken with caution. While these three infants had poorer outcomes on their
Bayley-4 scores, and the FYI v3.1 may have indicated that one of these infants maybe at risk for
autism; further follow up and a more adequately powered study is needed to explore these topics
and areas of research.
Last, a different motor learning paradigm could be used to examine the subtle motor
learning differences between infants at ER and CR. Based on the literature about the motor
behavior in infants at ER, it may be more appropriate to assume that motor learning differences
are more subtle, and an error learning assessment would be more fruitful in assessing learning
differences. An error learning paradigm could reveal these subtle motor learning differences by
examining the timing of motor behaviors and how infants learn from past errors to perform a
motor paradigm more proficiently. These future studies can aid in our knowledge of infant motor
learning and the development of motor interventions for infants at ER.

LEARNING CONTINGENCY PARADIGM IN INFANTS

65
CHAPTER IV: SUMMARY AND FUTURE STEPS
This dissertation investigated how infants at community and elevated risk for autism use
their visual motor behavior to engage in a contingency learning paradigm. Using the data from
the infants at community risk we asked if infants display visual anticipation while learning a
contingency learning paradigm and described the changes in visual motor behavior across the
paradigm. Then, we examined if there were any behavioral differences between infants at
community and elevated risk for autism when engaging in a contingency learning paradigm.
In the first part of the dissertation, we conclude that infants do display visual anticipation
while engaging in a contingency learning paradigm and that they use it more than random
chance. Given that the results from our comparisons between classically-defined “learners” and
“non-learners” were not significantly different and that most of the infants engaged in the
paradigm similarly, we state that the classical definition of learning a contingency paradigm is
insufficient for classifying learning. All of our participants at community risk for autism
displayed a time point where they were able to visual anticipate the majority of their activation of
the robot. Therefor it is suggested that infants show evidence of learning the paradigm (via
evidence of visual anticipation), regardless of the classification of learning based on the classical
definition. However, a retention or transfer task is needed to examine learning further in this
paradigm in order to determine if visual anticipation is confirmed evidence of an infant learning
the contingency (Krakauer et al., 2019).
For infants at elevated risk for autism, we found no significant differences in their visual
motor behavior compared to infants at community risk. Additionally, infants at elevated and
community risk had a similar proportion of infants classified as learners according to the
classical definition. While our hypotheses were not supported, our results guide important
LEARNING CONTINGENCY PARADIGM IN INFANTS

66
questions for future studies. Given our lack of statistically significant results, we suggest that to
study motor learning and impairments in autism, a paradigm different from ours is potentially
needed. Specifically, we suggest that the use of an error-based learning paradigm could highlight
the subtle motor learning differences that potentially exist between infants at elevated and
community risk. Additionally, we state that the examination of motor learning adaptation can be
fruitful in better understanding how infants at elevated risk and who are ultimately diagnosed
with autism learning motor skill. Children with autism are described as having variability in the
ways that they process sensory information which can result in use of motor learning adaptations
(Gowen & Hamilton, 2013).
To explore motor adaptations further, we also found that there were three cases where
infants at ER did not display a time point where they visual anticipated most of their activations
of the robot. This possibly indicated the potential for motor learning adaptations and further
follow up and study is needed.

Significance and Clinical Indications
This dissertation developed and implemented a novel approach to study visual motor
behavior in infants during a motor learning paradigm. In terms of literary significance, we
showed that all infants displayed visual anticipation at a single time point during the contingency
paradigm. We also found that infants classified as classically defined learners and non-learners
are more similar in terms of their visual motor behavior during the contingency portion of the
paradigm. Thus, we claim that future contingency learning paradigms need to use a retention or
transfer task to determine if infants learned the paradigm and if visual anticipation can be used as
a marker of learning.
LEARNING CONTINGENCY PARADIGM IN INFANTS

67
The prior practice of using the classical threshold of “1.5 times more reinforced behaviors
during the contingency phase compared to the baseline phase” is flawed and fails to establish if
the infant understands and retains the understanding that their behavior caused an outcome.
Therefore, it should be abandoned for a retention or transfer task or these assessments should be
used to confirm that visual anticipation marks learning of the paradigm. While this makes the
future work of contingency learning much harder, given the need to test on two separate days
(i.e. retention test) or for longer periods of time (transfer task), it opens the opportunity for a
better understanding of what is being learned in these motor paradigms designed for infants.
In terms of clinical significance, we speculate as to what learning processes the infants
are displaying while engaging in the paradigm. If infants learn motor skills in a similar manner
as children and adults; this means that clinicians should focus on establishing behavioral
connections between movements and the outcomes for those movements during the early stages
of learning a motor skill, and then on motor practice in latter stages of learning. However, these
clinical suggestions need further study before implementation to determine if they aid in optimal
motor learning.
Lastly, we found that infants at elevated risk compared to community risk did not exhibit
visual motor or learning differences. While these finding may seem to be insignificant, we
highlight further areas to study to better understand motor learning in infants at elevated risk for
autism. First, we suggest that other motor learning paradigms should be utilized to better
understand motor learning in this population. Potentially, a paradigm that assess error-based
motor learning may be more suitable for determining motor learning issues in infants at elevated
risk for autism. Secondly, we showed that we were able to collect visual motor behavior during a
motor learning paradigm in infants at elevated risk for autism and that these methods can be used
LEARNING CONTINGENCY PARADIGM IN INFANTS

68
in future motor learning paradigm with this population. Specifically, we suggest that differences
in patterns of visual motor behaviors may show motor learning adaptions in infants at elevated
risk for autism. These methods can be used in future studies that are adequately sampled to better
understand potential motor learning differences in both infants at elevated risk of autism and
those who go on to be diagnosed with autism.

Limitation
As described in the previous chapters, the sample size for these data was small and that
limits the interpretations about comparing classically defined “non-learners” and “learners” and
infants at ER and CR. Additionally, the fact that the infants were not followed until a diagnosis
of autism limits our ability to draw a link between our data and motor impairments in autism.
However, these data provide the basis for further questions that can be used to estimate the
appropriate sample size in future studies.
Some of our other limitations should not be seen as a limitation, but more as future
questions for understanding motor learning in neurotypical and neurodiverse populations. For
example, we explain how not using a retention or transfer task limits are ability to classify if
learning occurred. While the current study design limits what we can conclude, it does provide
us with the basis for further questions about infant motor learning and allows us the design future
studies that looks at these motor learning processes more in-depth. These future studies may also
be paired with neural imaging, such as electroencephalography (EEG) or functional near-infrared
spectroscopy (fNIRS), to study the cognitive processes of motor learning more directly in
neurotypical and neurodiverse populations.
LEARNING CONTINGENCY PARADIGM IN INFANTS

69
Another example is that our second study found no significant differences between the
two groups. This possibly indicated that this paradigm may be suitable for examining severe
visual motor learning impairments and that other motor learning paradigms, such as error-based
learning paradigm, are more applicable for infants at elevated risk for autism. However, our
second study does show that we can collect visual motor learning data in a neurodiverse
population and that we can possibly explore motor learning differences in other pediatric
populations.

Future Steps
Throughout this dissertation, we have stated a couple of future steps for this novel
analysis for visual motor learning in infants at community and elevated risk for autism. Given
our evidence that learning occurred by display of visual anticipation for the majority of
activation in a minute block, we should confirm these finding by looking at other markers of
learning. For example, we can look into using a transfer or a retention paradigm as in classical
motor learning literature. Additionally, we can use pupil dilation as a measure of arousal and
surprise, to see if infants do not display pupil dilation after the predictive gaze block. Last, these
data can be paired with a neural imaging method such as EEG or fNIRS to more directly
measure the early cognitive processes involved with infant motor learning.
Although we did not find significant visual motor learning differences between infants at
community and elevated risk for autism, this methodology can be used to study visual motor
learning in other neurodivergent populations at risk for more severe motor impairments. Prior
work from our lab has shown that preterm infants at risk for motor impairments can participate in
a contingency paradigm using a socially assistive robot (Deng et al., 2023). In the future we can
LEARNING CONTINGENCY PARADIGM IN INFANTS

70
use our methods for interpreting eye gaze data to understand how infants who are born very
preterm learn motor skills using their visual motor system.
Lastly, we question the state of the literature describing motor impairments in infants at
elevated risk of autism and those who go on to be diagnosed with autism. We speculate for our
findings and the literature, that motor learning impairments could be more subtle in autism. Error
based learning tasks may be more fruitful given the literature studying balance, gait, and
reaching in autism. Additionally, we still do not know the mechanisms of motor impairments in
autism regardless of age. Therefore, further studies should look to examining motor learning in
autistic children and adults to determine mechanisms of impairments that exist in this population.

LEARNING CONTINGENCY PARADIGM IN INFANTS

71
REFERENCES
Alessandri, S. M., Sullivan, M. W., Imaizumi, S., & Lewis, M. (1993). Learning and Emotional
Responsivity in Cocaine-Exposed Infants. Developmental Psychology, 29(6), 989–997.
https://doi.org/10.1037/0012-1649.29.6.989
Alessandri, S. M., Sullivan, M. W., & Lewis, M. (1990). Violation of Expectancy and
Frustration in Early Infancy. Developmental Psychology, 26(5), 738–744.
https://doi.org/10.1037/0012-1649.26.5.738
Angulo-Barroso, R. M., Peciña, S., Lin, X., Li, M., Sturza, J., Shao, J., & Lozoff, B. (2017).
Implicit learning and emotional responses in nine-month-old infants. Cognition and
Emotion, 31(5), 1031–1040. https://doi.org/10.1080/02699931.2016.1179624
Angulo-Kinzler, R. M. (2001). Exploration and selection of intralimb coordination patterns in 3-
month-old infants. Journal of Motor Behavior, 33(4), 363–376.
https://doi.org/10.1080/00222890109601920
Angulo-Kinzler, R. M., & Horn, C. L. (2001). Selection and memory of a lower limb motor-
perceptual task in 3-month-old infants. Infant Behavior and Development, 24(3), 239–257.
https://doi.org/10.1016/S0163-6383(01)00083-2
Angulo-Kinzler, R. M., Ulrich, B., & Thelen, E. (2002). Three-month-old infants can select
specific leg motor solutions. Motor Control, 6(1), 52–68. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/11842270
Baranek, G., Watson, L. R., Crais, E., Turner-Brown, L., & Reznik, S. (2013). First year
inventory (FYI) 3.1. Chapel Hill, NC: University of North Carolina at Chapel Hill.
Baranek, G. T., Sideris, J., Chen, Y. J., Crais, E. R., Turner-Brown, L., & Watson, L. R. (2022).
Early measurement of autism risk constructs in the general population: A new factor
structure of the First Years Inventory (FYIv3.1) for ages 6–16 months. Autism Research,
15(5), 915–928. https://doi.org/10.1002/aur.2691
Bayley, N., & Alyward, G. P. (2019). Bayley Scales of Infant and Toddler Development
TM
,
Fourth Edition (Bayley
TM
-4) (4th ed.). Toronto, Canada: Pearson.
Bhat, Anjana N. (2021). Motor Impairment Increases in Children With Autism Spectrum
Disorder as a Function of Social Communication, Cognitive and Functional Impairment,
Repetitive Behavior Severity, and Comorbid Diagnoses: A SPARK Study
Report. Autism Research, 14(1), 202–219. https://doi.org/10.1002/aur.2453
Bhat, Anjana Narayan. (2020). Is Motor Impairment in Autism Spectrum Disorder Distinct From
Developmental Coordination Disorder? A Report From the SPARK Study. Physical
Therapy, 100(4), 633–644. https://doi.org/10.1093/ptj/pzz190
Braukmann, R., Ward, E., Hessels, R. S., Bekkering, H., Buitelaar, J. K., & Hunnius, S. (2018).
Action prediction in 10-month-old infants at high and low familial risk for Autism
LEARNING CONTINGENCY PARADIGM IN INFANTS

72
Spectrum Disorder. Research in Autism Spectrum Disorders, 49(February), 34–46.
https://doi.org/10.1016/j.rasd.2018.02.004
Brooks, V., Hilperath, F., Brooks, M., Ross, H. G., & Freund, H. J. (1995). Learning “what” and
“how” in a human motor task. Learning & Memory (Cold Spring Harbor, N.Y.), 2(5), 225–
242. https://doi.org/10.1101/lm.2.5.225
Bulmer, M. G. (1967). Principles of Statistics (First). New York: Dover Publications.
Chambon, V., Sidarus, N., & Haggard, P. (2014). From action intentions to action effects: How
does the sense of agency come about? Frontiers in Human Neuroscience, 8(MAY), 1–9.
https://doi.org/10.3389/fnhum.2014.00320
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, N.J:
L. Erlbaum Associates.
Deng, W., Sargent, B., Havens, K., Vanderbilt, D., Rosales, M., Pulido, J. C., … Smith, B. A.
(2023). Correlation between performance and quantity/variability of leg exploration in a
contingency learning task during infancy. Infant Behavior and Development, 70, 101788.
https://doi.org/https://doi.org/10.1016/j.infbeh.2022.101788
Dunst, C., Strock, A., Hutto, M., & Snyder, D. (2006). Relative Effectiveness of Episodic and
Conjugate Reinforcement on Child Operant Learning. Bridges, 4(1), 1–15.
Ekberg, T. L., Falck-Ytter, T., Bölte, S., Gredebäck, G., Campbell, L., Cauvet, E., … Zander, E.
(2016). Reduced prospective motor control in 10-month-olds at risk for autism spectrum
disorder. Clinical Psychological Science, 4(1), 129–135.
https://doi.org/10.1177/2167702615576697
Elsabbagh, M., Fernandes, J., Jane Webb, S., Dawson, G., Charman, T., & Johnson, M. H.
(2013). Disengagement of visual attention in infancy is associated with emerging autism in
toddlerhood. Biological Psychiatry, 74(3), 189–194.
https://doi.org/10.1016/j.biopsych.2012.11.030
Esposito, G., Venuti, P., Maestro, S., & Muratori, F. (2009). An exploration of symmetry in early
autism spectrum disorders: Analysis of lying. Brain and Development, 31(2), 131–138.
https://doi.org/10.1016/j.braindev.2008.04.005
Falck-Ytter, T., Gredebäck, G., & Von Hofsten, C. (2006). Infants predict other people’s action
goals. Nature Neuroscience, 9(7), 878–879. https://doi.org/10.1038/nn1729
Fitter, N. T., Funke, R., Pulido, J. C., Eisenman, L. E., Deng, W., Rosales, M. R., … Mataric, M.
J. (2019). Socially Assistive Infant-Robot Interaction: Using Robots to Encourage Infant
Leg-Motion Training. IEEE Robotics and Automation Magazine, 26(2), 12–23.
https://doi.org/10.1109/MRA.2019.2905644
Flanagan, J. E., Landa, R., Bhat, A., & Bauman, M. (2012). Head lag in infants at risk for
autism: A preliminary study. American Journal of Occupational Therapy, 66(5), 577–585.
LEARNING CONTINGENCY PARADIGM IN INFANTS

73
https://doi.org/10.5014/ajot.2012.004192
Franchak, J. M., Kretch, K. S., Soska, K. C., & Adolph, K. E. (2011). Head-mounted eye-
tracking: A new method to describe infant looking. Child Development, 82(6), 1738–1750.
https://doi.org/10.1111/j.1467-8624.2011.01670.x.Head-mounted
Gibbs, W. W. (2021). Autism by the numbers: Explaining its apparent rise. Spectrum.
https://doi.org/10.53053/wsyh3045
Golenia, L., Schoemaker, M. M., Mouton, L. J., & Bongers, R. M. (2014). Individual differences
in learning a novel discrete motor task. PLoS ONE, 9(11).
https://doi.org/10.1371/journal.pone.0112806
Gowen, E., & Hamilton, A. (2013). Motor abilities in autism: A review using a computational
context. Journal of Autism and Developmental Disorders, 43(2), 323–344.
https://doi.org/10.1007/s10803-012-1574-0
Gredebäck, G., & Falck-Ytter, T. (2015). Eye Movements During Action Observation.
Perspectives on Psychological Science, 10(5), 591–598.
https://doi.org/10.1177/1745691615589103
Gredebäck, G., Johnson, S., & Von Hofsten, C. (2010). Eye tracking in infancy research.
Developmental Neuropsychology, 35(1), 1–19. https://doi.org/10.1080/87565640903325758
Johansson, R. S., Westling, G., Backstrom, A., & Flanagan, J. R. (2001). Eye – Hand
Coordination in Object Manipulation. Journal of Neuroscience, 21(17), 6917–6932.
Johnson, M. H., Gliga, T., Jones, E., & Charman, T. (2015). Annual Research Review: Infant
development, autism, and ADHD – early pathways to emerging disorders.
Journal of Child Psychology and Psychiatry, 56(3), 228–247.
https://doi.org/10.1111/jcpp.12328
Johnson, S. P., Amso, D., & Slemmer, J. A. (2003). Development of object concepts in infancy.
Journal of Vision, 3(9). https://doi.org/10.1167/3.9.690
Jones, W., & Klin, A. (2013). Attention to eyes is present but in decline in 2-6-month-old infants
later diagnosed with autism. Nature, 504(7480), 427–431.
https://doi.org/10.1038/nature12715
Kanakogi, Y., & Itakura, S. (2011). Developmental correspondence between action prediction
and motor ability in early infancy. Nature Communications, 2(1), 341–346.
https://doi.org/10.1038/ncomms1342
Kaur, M., M. Srinivasan, S., & N. Bhat, A. (2018). Comparing motor performance, praxis,
coordination, and interpersonal synchrony between children with and without Autism
Spectrum Disorder (ASD). Research in Developmental Disabilities, 72(October 2017), 79–
95. https://doi.org/10.1016/j.ridd.2017.10.025
LEARNING CONTINGENCY PARADIGM IN INFANTS

74
Kelso, J. A. S. (2016). On the Self-Organizing Origins of Agency. Trends in Cognitive Sciences,
20(7), 490–499. https://doi.org/10.1016/j.tics.2016.04.004
Kelso, J. A. S., & Fuchs, A. (2016). The coordination dynamics of mobile conjugate
reinforcement. Biological Cybernetics, 110(1), 41–53. https://doi.org/10.1007/s00422-015-
0676-0
Kenward, B. (2010a). 10-Month-Olds Visually Anticipate an Outcome Contingent on Their Own
Action. Infancy, 15(4), 337–361. https://doi.org/10.1111/j.1532-7078.2009.00018.x
Kenward, B. (2010b). 10-Month-Olds Visually Anticipate an Outcome Contingent on Their Own
Action. Infancy, 15(4), 337–361. https://doi.org/10.1111/j.1532-7078.2009.00018.x
Krakauer, J. W., Hadjiosif, A. M., Xu, J., Wong, A. L., & Haith, A. M. (2019). Motor Learning.
In Comprehensive Physiology (pp. 613–663). https://doi.org/10.1002/cphy.c170043
Kyvelidou, A., Koss, K., Wickstrom, J., Needelman, H., Fisher, W. W., & DeVeney, S. (2021).
Postural control may drive the development of other domains in infancy. Clinical
Biomechanics, 82(June 2020), 105273. https://doi.org/10.1016/j.clinbiomech.2021.105273
Landa, R., & Garrett-Mayer, E. (2006). Development in infants with autism spectrum disorders:
A prospective study. Journal of Child Psychology and Psychiatry and Allied Disciplines,
47(6), 629–638. https://doi.org/10.1111/j.1469-7610.2006.01531.x
Lester, B. M., & Tronick, E. Z. (2004). The Neonatal Intensive Care Unit Network
Neurobehavioral Scale Procedures. Pediatrics, 113(Supplement 2), 641 LP – 667. Retrieved
from http://pediatrics.aappublications.org/content/113/Supplement_2/641.abstract
Moore, J. W. (2016). What is the sense of agency and why does it matter? Frontiers in
Psychology, 7(AUG), 1–9. https://doi.org/10.3389/fpsyg.2016.01272
Moraes, Í. A. P. de, Massetti, T., Crocetta, T. B., Silva, T. D. da, Menezes, L. D. C. de,
Monteiro, C. B. de M., & Magalhães, F. H. (2017). Motor learning characterization in
people with autism spectrum disorder: A systematic review. Dementia & Neuropsychologia,
11(3), 276–286. https://doi.org/10.1590/1980-57642016dn11-030010
Ozonoff, S., Young, G. S., Carter, A., Messinger, D., Yirmiya, N., Zwaigenbaum, L., … Stone,
W. L. (2011). Recurrence risk for autism spectrum disorders: A baby siblings research
consortium study. Pediatrics, 128(3). https://doi.org/10.1542/peds.2010-2825
Ozonoff, S., Young, G. S., Goldring, S., Greiss-Hess, L., Herrera, A. M., Steele, J., … Rogers, S.
J. (2008). Gross motor development, movement abnormalities, and early identification of
autism. Journal of Autism and Developmental Disorders, 38(4), 644–656.
https://doi.org/10.1007/s10803-007-0430-0
Piek, J. P. (2002). The role of variability in early motor development. INFANT BEHAVIOR &
DEVELOPMENT, 25(4), 452–465. https://doi.org/10.1016/S0163-6383(02)00145-5
LEARNING CONTINGENCY PARADIGM IN INFANTS

75
Reetzke, R., Iosif, A. M., Hatch, B., de la Paz, L., Chuang, A., Ozonoff, S., & Miller, M. (2022).
Patterns of objectively measured motor activity among infants developing ASD and
concerns for ADHD. Journal of Child Psychology and Psychiatry and Allied Disciplines,
63(6), 663–673. https://doi.org/10.1111/jcpp.13504
Rosander, K., & von Hofsten, C. (2011). Predictive gaze shifts elicited during observed and
performed actions in 10-month-old infants and adults. Neuropsychologia, 49(10), 2911–
2917. https://doi.org/10.1016/j.neuropsychologia.2011.06.018
Rovee-Collier, C. K., Sullivan, M. W., Enright, M., Lucas, D., & Fagen, J. W. (1980).
Reactivation of Infant Memory. Science, 208(4448), 1159–1161. Retrieved from
http://www.jstor.org/stable/1683726
Rovee, C. K., & Gekoski, M. J. (1979). THE ECONOMICS OF INFANCY: A REVIEW OF
CONJUGATE REINFORCEMENT. In Advances in Child Developmentand Behavior (Vol.
13). https://doi.org/10.1017/CBO9781107415324.004
Rovee, C. K., & Rovee, D. T. (1969). Conjugate reinforcement of infant exploratory behavior.
Journal of Experimental Child Psychology, 8(1), 33–39. https://doi.org/10.1016/0022-
0965(69)90025-3
Sacrey, L. A. R., Zwaigenbaum, L., Bryson, S., Brian, J., & Smith, I. M. (2018). The reach-to-
grasp movement in infants later diagnosed with autism spectrum disorder: A high-risk
sibling cohort study. Journal of Neurodevelopmental Disorders, 10(1), 1–11.
https://doi.org/10.1186/s11689-018-9259-4
Sailer, U., Flanagan, J. R., & Johansson, R. S. (2005). Eye-hand coordination during learning of
a novel visuomotor task. Journal of Neuroscience, 25(39), 8833–8842.
https://doi.org/10.1523/JNEUROSCI.2658-05.2005
Sargent, B., Kubo, M., & Fetters, L. (2018). Infant Discovery Learning and Lower Extremity
Coordination: Influence of Prematurity. Physical and Occupational Therapy in Pediatrics,
38(2), 210–225. https://doi.org/10.1080/01942638.2017.1357065
Sargent, B., Reimann, H., Kubo, M., & Fetters, L. (2015). Quantifying learning in young infants:
Tracking leg actions during a discovery-learning task. Journal of Visualized Experiments,
2015(100), 1–14. https://doi.org/10.3791/52841
Sargent, B., Schweighofer, N., Kubo, M., & Fetters, L. (2014). Infant exploratory learning:
Influence on leg joint coordination. PLoS ONE, 9(3).
https://doi.org/10.1371/journal.pone.0091500
Swiney, L., & Sousa, P. (2014). A new comparator account of auditory verbal hallucinations:
How motor prediction can plausibly contribute to the sense of agency for inner speech.
Frontiers in Human Neuroscience, 8(AUG), 1–15.
https://doi.org/10.3389/fnhum.2014.00675
Thelen, E., Kelso, J. A. S., & Fogel, A. (1987). Self-organizing systems and infant motor
LEARNING CONTINGENCY PARADIGM IN INFANTS

76
development. Developmental Review, 7(1), 39–65. https://doi.org/10.1016/0273-
2297(87)90004-9
Thorup, E., Nyström, P., Gredebäck, G., Bölte, S., & Falck-Ytter, T. (2018). Reduced
Alternating Gaze During Social Interaction in Infancy is Associated with Elevated
Symptoms of Autism in Toddlerhood. Journal of Abnormal Child Psychology, 46(7), 1547–
1561. https://doi.org/10.1007/s10802-017-0388-0
Ulrich, B. D. (2010). Opportunities for Early Intervention Based on Theory, Basic Neuroscience,
and Clinical Science. Physical Therapy, 90(12), 1868–1880.
https://doi.org/10.2522/ptj.20100040
Wang, Q., Bolhuis, J., Rothkopf, C. A., Kolling, T., Knopf, M., & Triesch, J. (2012). Infants in
control: Rapid anticipation of action outcomes in a gaze-contingent paradigm. PLoS ONE,
7(2), 1–6. https://doi.org/10.1371/journal.pone.0030884
Wass, S. V., Jones, E. J. H., Gliga, T., Smith, T. J., Charman, T., Johnson, M. H., … Volein, A.
(2015). Shorter spontaneous fixation durations in infants with later emerging autism.
Scientific Reports, 5(Lc), 1–8. https://doi.org/10.1038/srep08284
West, K. L. (2018). Infant Motor Development in Autism Spectrum Disorder: A Synthesis and
Meta-analysis. Child Development, 00(0), 1–18. https://doi.org/10.1111/cdev.13086
Zaadnoordijk, L., Meyer, M., Zaharieva, M., Kemalasari, F., van Pelt, S., & Hunnius, S. (2020).
From movement to action: An EEG study into the emerging sense of agency in early
infancy. Developmental Cognitive Neuroscience, 42, 100760.
https://doi.org/10.1016/j.dcn.2020.100760
Zaadnoordijk, L., Otworowska, M., Kwisthout, J., & Hunnius, S. (2018). Can infants’ sense of
agency be found in their behavior? Insights from babybot simulations of the mobile-
paradigm. Cognition, 181(August), 58–64. https://doi.org/10.1016/j.cognition.2018.07.006
Zaadnoordijk, L., Otworowska, M., Kwisthout, J., Hunnius, S., & Van Rooij, I. (2017). The
mobile-paradigm as measure of infants’ sense of agency? Insights from babybot
simulations. 2016 Joint IEEE International Conference on Development and Learning and
Epigenetic Robotics, ICDL-EpiRob 2016, 41–42.
https://doi.org/10.1109/DEVLRN.2016.7846787

LEARNING CONTINGENCY PARADIGM IN INFANTS

77
APPENDICES
Appendix 1. The timing of gazes on the reinforcement for each infant. The x-axis is time in
seconds. A value of 0 on the x-axis is when the robot would activate. Negative values on the x-
axis mean that the gaze occurred prior to the robot’s activation, while positive values occurred
during its activation. The y-axis is the frequency for that block. The dotted line in each plot
denoted the median for the distribution of data. Non-robot looks are not included in this analysis.
NL1 NL2 L1
L2 L3 L4
NL5 L5 NL6
NL7 NL8 L6
NL9
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6
Time (seconds)
0
20
40
60
80
Count
-0.4 0 0.4 0.8 1.2 1.6 2
Time (seconds)
0
20
40
60
80
Count
LEARNING CONTINGENCY PARADIGM IN INFANTS

78
Appendix 2. Individual plot for each infant, from each group and learning classification. The title
above each graph denotes the infant ID number, learning classification, risk status (superscript),
and their total number of activations (L= Classically defined learner, NL- Classically defined
non-learner, CR= Community risk, ER=Elevated risk). Graph a) is a bar graph that plots the
total number of activations on the y-axis and the minute blocks of the contingency period on x-
axis. Graph b) is a line graph that plots the frequency of the type of gaze in the y axis for the
same minute blocks described in graph a. The colors denote the types of gazes (blue = non-robot
looks, red = reactive looks, yellow = predictive looks).

Community Risk Learners

L1
CR
- 100 activations
a) b)

L2
CR
- 63 activations
a) b)

L3
CR
- 66 activations
a) b)
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
14
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

79
L4
CR
- 87 activations
a) b)

L5
CR
- 77 activations
a) b)

L6
CR
- 41 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
2
4
6
8
10
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
1
2
3
4
5
6
7
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

80
Community Risk Non-Learners

NL1
CR
- 90 activations
a) b)

NL2
CR
- 124 activations
a) b)

NL5
CR
- 43 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
14
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
2
4
6
8
10
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
1
2
3
4
5
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

81
NL6
CR
- 110 activations
a) b)

NL7
CR
- 33 activations
a) b)

NL8
CR
- 72 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
2
4
6
8
10
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
1
2
3
4
5
6
7
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

82
NL9
CR
- 97 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

83
Elevated Risk Learners

L1
ER
- 126 activations
a) b)

L2
ER
- 32 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
30
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
1
2
3
4
5
6
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
1
2
3
4
5
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

84
Elevated Risk Non-Learners

NL1
ER
- 106 activations
a) b)

NL2
ER
- 56 activations

a) b)

NL3
ER
- 59 activations
a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
2
4
6
8
10
12
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

85
NL4
ER
- 114 activations

a) b)

NL5
ER
- 81 activations

a) b)

NL6
ER
- 70 activations

a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
30
35
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
5
10
15
20
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
14
Frequency of Looks
Non-Robot Looks Reactive Predictive
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
12
14
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

86
NL7
ER
- 108 activations

a) b)

1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
Robot Activations
Non-Robot Looks Reactive Predictive
0 2 4 6 8
Minute
0
2
4
6
8
10
Frequency of Looks
Non-Robot Looks Reactive Predictive
LEARNING CONTINGENCY PARADIGM IN INFANTS

87
Appendix 3. Individual bar plot for each infant at community risk for autism depicting the
amount of time spent looking (y-axis) during each minute block (x-axis) of the contingency. The
infant’s ID number is displayed to the left of the infant’s graph. The classification of learning is
denoted as L for learners and NL for non-learners.

NL1 NL2
L1 L2
L3 L4
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
LEARNING CONTINGENCY PARADIGM IN INFANTS

88
NL5 L5
NL6 NL7
NL8 L6
NL9

1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
LEARNING CONTINGENCY PARADIGM IN INFANTS

89
Appendix 4. Individual bar plot for each infant at elevated risk from autism depicting the amount
of time spent looking (y-axis) during each minute block (x-axis) of the contingency. The infant’s
ID number is displayed to the left of the infant’s graph. The classification of learning is denoted
as L for learners and NL for non-learners.

NL1 L1
NL2 NL3

NL4 NL5
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
5
10
15
20
25
30
35
Looking Duration (s)
LEARNING CONTINGENCY PARADIGM IN INFANTS

90
L2 NL6
NL7
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)
1 2 3 4 5 6 7 8
Minute
0
10
20
30
40
50
60
Looking Duration (s)

Abstract (if available)

Abstract This dissertation examines the visual motor learning patterns that infants at elevated and community risk for autism use while learning a contingency paradigm. We used a social assistive robot and wearable sensors to reinforce infant right leg movements. During the paradigm we used a head mounted eye tracker and wearable sensors to examine infant visual motor behaviors. In the first aim, we challenge the classical definition of learning the paradigm. Additionally, we recommend future guidance and questions for examining infant visual motor behavior and contingency learning. In the second aim of the study, we examine visual motor learning differences between infants at elevated and community risk for autism. This aim highlights future areas of study for early motor learning in infants at elevated risk for autism and children with autism spectrum disorder. In summary, the findings from this dissertation support that additional behavioral measure (i.e. measures of visual behavior) can be used in novel ways to gain a better understanding of infant motor learning. Additionally, it advocated for more research when it comes to understanding motor learning in infants at elevated risk for autism and those who eventually receive a diagnose. Through a better understanding of motor learning, we can begin to develop and tailor motor interventions to address motor impairments in individuals with autism.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Learning acceleration control of leg movements during infancy

PDF

Impact of enhanced efficacy expectation on motor learning in individuals with Parkinson’s disease

PDF

The brain and behavior of motor learning: the what, how and where

PDF

Examining parent responsiveness to sensory reactivity and regulation cues of infants with developmental risk factors

PDF

Understanding how infants born full-term and preterm learn, move, and explore

PDF

Relationship between brain structure and motor behavior in chronic stroke survivors

PDF

Neural substrates associated with context-dependent learning

PDF

Infant exploratory learning: influence on leg coordination

PDF

Learning reaching skills in non-disabled and post-stroke individuals

PDF

Brain and behavior correlates of intrinsic motivation and skill learning

PDF

Computational principles in human motor adaptation: sources, memories, and variability

PDF

Behavioral and neurophysiological studies of hand dexterity in health and Parkinson's disease

PDF

The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait

PDF

Relationship between interhemispheric inhibition and bimanual coordination in musicians

PDF

Sociodemographic influences on the early caregiving environment and likelihood of autism

PDF

Modeling dyadic synchrony with heterogeneous data: validation in infant-mother and infant-robot interactions

PDF

Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke

PDF

Developmental trajectories, maternal-infant transactions, and child outcomes of social-emotional competence and sensory processing

PDF

Locomotor skill learning in virtual reality in healthy adults and people with Parkinson disease

PDF

Using ecological momentary assessment to study the impact of social-cognitive factors on paretic hand use after stroke

Asset Metadata

Creator Rosales, Marcelo Ramon (author)

Core Title Learning self-agency during a contingency paradigm in infants at commnity and elevated risk of autism

School School of Dentistry

Degree Doctor of Philosophy

Degree Program Biokinesiology and Physical Therapy

Degree Conferral Date 2023-08

Publication Date 07/05/2023

Defense Date 06/05/2023

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Gaze,Infant,motor learning,OAI-PMH Harvest,robot,visual motor

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Bradley, Nina (committee chair), Baranek, Grace (committee member), Franchak, John (committee member), Smith, Beth (committee member), Winstein, Carolee (committee member)

Creator Email marcelorosales721@gmail.com,mrrosale@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113262028

Unique identifier UC113262028

Identifier etd-RosalesMar-12022.pdf (filename)

Legacy Identifier etd-RosalesMar-12022

Document Type Dissertation

Format theses (aat)

Rights Rosales, Marcelo Ramon

Internet Media Type application/pdf

Type texts

Source 20230706-usctheses-batch-1062 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu