University of Southern California Dissertations and Theses

Homeostatic imbalance and monetary delay discounting: effects of feeding on RT, choice, and brain response

(USC Thesis Other)

Homeostatic imbalance and monetary delay discounting: effects of feeding on RT, choice, and brain response

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

I

HOMEOSTATIC IMBALANCE AND MONETARY DELAY
DISCOUNTING: EFFECTS OF FEEDING ON RT, CHOICE, AND
BRAIN RESPONSE

By

Andrew James Melrose

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Completion of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
PSYCHOLOGY: BRAIN AND COGNITIVE SCIENCES

May 2018

II
Dedication

This thesis is dedicated to my amazing wife Vanessa. I could not have made it this far
without your unconditional love and support. Thank you for always believing in me, even
when I didn’t have it in me to believe in myself when anybody else would have. I don’t
deserve you but I’m thankful every day that I have you as my partner in this crazy
messed up world.

III
Acknowledgments
Without the support of my advisor, Dr. John Monterosso, my fellow lab members,
and the labs of my collaborators, the finishing (or even starting!) of this work would not
have been possible. First and foremost, I owe not only the current work, but my
progression as a scientist and as a critical thinker to the unwavering support of my
advisor John Monterosso. Throughout the course of graduate school he has always
responded to the presentation of a new opportunity or a new challenge not with how can
it benefit the school/lab/himself, but will I get anything useful out of the process. Be it a
collaboration with somebody new, the learning of some new computational skill, or the
onset of a new project, John’s approach has always been to ask himself [and me] what it
is I will gain from the added work/opportunity/responsibility. This legitimate interest in
the development of his students in to well-rounded scientists and human beings is a rare
attribute for a mentor to attain, and is something he has proven time and time again
during my tenure as his graduate student.
Indeed one of my biggest sources of support throughout the finishing of this work,
and graduate school in general, was my fellow lab mate Dr. Eustace Hsu. From the very
start of my time at USC, he has not only proven himself time and again that he can be
relied upon for nearly anything research related, but also as a friend. Without Eustace’s
help with concepts, formulas, coding, experimental design, running studies as a team or
the watching of football/basketball over a beer throughout the years I do not know that I
would have made it to the point of defending a dissertation! Similarly, I owe a huge debt
of gratitude to Xochitl Cordova for her hours of support throughout the whole process of
this work and others like it. Often on unpaid projects she was always willing to help me

IV
with IRB, financial services, managing/training students, scheduling, or even just
listening to me vent! Without her help, this dissertation would likely still be stuck
somewhere dealing with the administrative issues she always handles so well.
In addition, I would like to thank all of my undergraduate research students
throughout my time at USC that helped me with the mechanics of collecting experimental
data for every study I ran during my time at USC. Though I am grateful for the help of all
my students during my time at USC, I would like to give a special thank you to Olivia de
Santis, Olivia Kramer and Madison Norton. Without the help of you 3 wonderful students
I would not have been able to complete this dissertation, and cannot thank you enough for
all of the hard work and dedication you all put forth while also all juggling full class
schedules!
I would also like to thank all of my committee members Drs. Richard John,
Antoine Bechara, Katie Page and Isabelle Brocas for all of your support and guidance
throughout my time at USC. Furthermore, I would like to thank Drs. Page, Bechara, and
Brocas for the opportunity to collaborate with their labs on various projects. Thanks to all
of these opportunities, I was able to continually RA for all but a single semester as a
graduate student, and was able to greatly advance my development as a research scientist
as well as learn so much through my interactions with each of you, and your respective
labs. In addition, I would like to thank th of the USC community who have helped me
with the unexplainable other pieces that go in to a PhD, all of which are important, and
all of which I would not have made it to this point without. Specifically, I would like to
thank Dr. Damien Brevers for being a constant source of inspiration through your
seemingly endless passion for science, and Tasha Poppa for the countless hours of

V
conversation both science-related and not both of whom are from Dr. Bechara’s lab. I
would also like to thank Dr. Shan Luo, Jasmine Alves, Hilary Dorton and Ana Romero
from Dr. Page’s lab for welcoming in to their lab during my tenure in their lab as a
Graduate Research Assistants and helping me develop and fine tune publications,
research ideas and presentations throughout my time at USC. Finally, I would like to
thank Dr. Dalton Colmbs from Dr. Brocas’ lab along with the many other graduate
students and post-docs who attended Dr. Brocas’ lab meetings, Dalton was one of the
first people to encourage me to learn Python-based implementations for my analysis, and
through countless lectures I credit my comfortability with the tenets of behavioral
economics despite being trained as a cognitive scientist on what I learned during these lab
meetings and the ideas they sparked.
Last but not least, I would like to thank my family for all of their love and support
throughout this whole process. I did not make the decision to pursue a PhD lightly, and
was unwavering in my pursuit of this personal goal, and despite being the first member of
my family to seek a graduate degree, I could always feel how proud and supportive my
immediate and extended family have been of me throughout this entire process despite
thousands of miles between us. In particular I would like to thank my wife Vanessa for
being a source of inspiration and support as I often worked to ungodly hours of the night,
and for my parents always supporting my choice to pursue a PhD and always taking an
interest in what it was I was researching simply because I would become so excited about
it. Without Vanessa’s unwavering love and support, I could not have come this far in life
nor would I have even dreamed to one day even start a PhD, never mind actually
completing one!

VI
Table of Contents Page

Dedication………………………………………………………………II

Acknowledgements……………………………………………………..III

Table of Contents……………………………………………………….VI

List of Tables……………………………………………………………VII

List of Figures…………………………………………………………..VIII

Abstract………………………………………………………………….1

Chapter 1-Introduction…………………………………………………..3

Chapter 2- Hunger and Intertemporal MID Task……………………….23

Chapter 3- Hunger, Delay Discounting and the Drift Diffusion Model...48

Chapter 4- General Discussion………………………………………….95

References………………………………………………………………101

Appendices:
Appendix 1………………………………………….…………..116
Appendix 2.1.1 Hunger Ratings MCMC………………………..117
Appendix 2.1.2 Fullness Ratings MCMC………………..……...118
Appendix 2.2.1 Hard vs. Easy ITC RT MCMC ………………..119
Appendix 2.2.2 Fast vs. Fed ITC RT MCMC…………………..120
Appendix 2.3.1 ITC HDDM GAT Model MCMC……….……..121
Appendix 2.3.2 HDDM Posterior Distributions ………………..124
Appendix 2.4 MID RT MCMC…………..……………………..126
Appendix 2.5 MID RT Expanded MCMC ……………………..128

VII
List of Tables Page
Table 1. ……………………………………………………………32
Table 2. ……………………………………………………………33
Table 3. ……………………………………………………………35
Table 4. ……………………………………………………………38
Table 5. ……………………………………………………………39
Table 6. ……………………………………………………………40
Table 7. ……………………………………………………………41
Table 8. ……………………………………………………………43
Table 9. ……………………………………………………………64
Table 10. …………………………………………………………..67
Table 11. …………………………………………………………..69
Table 12. …………………………………………………………..70
Table 13. …………………………………………………………..77
Table 14. …………………………………………………………..78
Table 15. …………………………………………………………..80
Table 16. …………………………………………………………..83
Table 17. …………………………………………………………..84
Table 18. …………………………………………………………..85

VIII
List of Figures Page
Figure 1. …………………………………………………………..28
Figure 2. …………………………………………………………..32
Figure 3. …………………………………………………………..34
Figure 4. …………………………………………………………..35
Figure 5. …………………………………………………………..36
Figure 6. …………………………………………………………..37
Figure 7. …………………………………………………………..39
Figure 8. …………………………………………………………..40
Figure 9. …………………………………………………………..41
Figure 10. ………………………………………………………….42
Figure 11. ………………………………………………………….44
Figure 12. ………………………………………………………….53
Figure 13. ………………………………………………………….65
Figure 14. ………………………………………………………….66
Figure 15. ………………………………………………………….68
Figure 16. ………………………………………………………….68
Figure 17. ………………………………………………………….69
Figure 18. ………………………………………………………….72
Figure 19. ………………………………………………………….72
Figure 20. ………………………………………………………….75
Figure 21. ………………………………………………………….76
Figure 22. ………………………………………………………….77
Figure 23. ………………………………………………………….78
Figure 24. ………………………………………………………….79
Figure 25. ………………………………………………………….80
Figure 26. ………………………………………………………….82
Figure 27. ………………………………………………………….84
Figure 28. ………………………………………………………….85

1
Abstract
Consistency of preferences is a convenient assumption to make when creating behavioral
economic models of choice, but is not a biological imperative of the individual, whereas
maintenance of homeostasis is. At it’s core, homeostasis is the drive of bodily systems to
maintain consistency at each system’s equilibrium point. Maintenance of equilibrium is
essential to survival and the pursuit of consistency can heuristically only be maintained
by dynamic motivational structures that allow inconsistent drives depending on where the
individual is relative to the equilibrium point. Though homeostasis is essential to nearly
all biological systems in the body, my focus will be on maintenance of hunger, and how
dynamic motivations related to hunger status may incidentally ‘spill-over’ to other
motivational systems. Experimental research on biological and behavioral changes
resulting from manipulation of hunger status is dominated by research on responses to
food stimuli. A core hypothesis of the current work is that the cascade of biological
changes in the brain and in the periphery leads to a “bleeding over” of preference shifts
into other reward domains (e.g. monetary) in behavioral economics assessments of these
constructs. In the defense of homeostasis, the body uses hunger and satiety as biological
signals to alter goal-directed action. Satiety is the body’s signal that energy demands have
been met, and homeostasis has been reinstated. In contrast, hunger alters the preferences
of the individual across many domains via mobilization of goal-directed action in the
pursuit of regaining homeostasis. The current work investigated the neurobiology and
related cognitive processes of how experimental manipulation of hunger status affects
intertemporal rewards and choices through the use of two tasks aimed at experimentally
controlling different aspects of intertemporal choice behavior during both a fasted and a

2
fed day during fMRI in a within-subject design. Results of an intertemporal variant of the
monetary incentive delay task indicate that hunger is associated with increased reward-
value of all intertemporal rewards as is evidenced by decreased response times and
increased limbic system activity to potential rewards. In addition, hunger was associated
with faster response times during ‘hard’ intertemporal choices during the completion of
an adaptive intertemporal choice paradigm. Cognitive modeling of response time data
during the adaptive intertemporal choice task through the fitting of a Hierarchical Drift
Diffusion Model provided evidence of decreased reward thresholds and non-decision
times during hunger. The adaptive intertemporal choice task provided evidence of
hunger-based shifts in delay discounting dependent on baseline rates of delay discounting
when delays are fixed with steep discounters becoming steeper, and shallow discounters
becoming shallower. Results of the current work provides evidence that hunger may be
affecting delay discounting through the changing of reward sensitivity, something that is
supported by both the animal and human literature on neurobiological mechanisms of
hunger-based behavior change.

3
Chapter 1: Introduction
Try as we might, humans tend to be dynamically inconsistent among similar
behaviors over time. Ranging from the commonplace failed New-Years diet on the 5
th

day of the new year, to the drug addict who is perpetually taking their last hit, utility
maximization in the neo-classical sense (which is modeled as being static) does not map
well on to actual behavior likely due to constantly varying behaviors and preferences
over time in humans. The foundations of modern-day descriptive models of choice
behavior which rivals the notion of the rational utility maximizing decision maker (often
outperforming it in terms of predictive power) can be traced back to Prospect Theory
which was formalized in the 1970’s and 80’s and focused on irrationality of the decision
maker during decision-making under risk (Kahneman & Tversky, 1984). Though
Prospect Theory’s tenets have more recently been criticized as not accurately describing
individual decision-making (Tversky & Kahneman, 1992), this work was instrumental in
pushing Science away from the rational decision-maker and towards modern-day
Behavioral Economics and Decision Neuroscience which attempt to formalize
inconsistencies within the individual decision-maker.
The scientific study of the systematic irrationality of the decision-maker has
evolved drastically since these early formulations of Framing Effects and Prospect
Theory (Kahneman & Tversky, 1984). In line with Prospect Theory’s description of the
vulnerability of preferences to choice-irrelevant information, there is considerable
evidence that emotions and peripheral bodily signals (or Somatic Markers) though often
instrumental to adaptive choices, can also interfere with decision-making depending on
their context as is formalized in the Somatic Marker Hypothesis (SMH) (for review see

4
Bechara & Damasio, 2005). Unlike prospect theory, the major tenets of the original SMH
have held up against experimental evidence time and again, providing extensive evidence
for the role of bodily signals in determining an individual’s preferences in both adaptive
and maladaptive ways. A seemingly straight forward extension of the SMH could be that
disrupting the body’s optimal biological state (or in other words disturbing the
homeostasis of our well balanced biological systems) will likely affect an individual’s
preferences due in large part to the central and peripheral biological signaling that
accompanies disruptions in homeostasis. This extension of the SMH, however, remains
largely unexplored and has yet to be integrated with most contemporary hypothesis on
the synergy between the body and brain during decision making. Though recent
experimental evidence strongly implies that disruptions in homeostasis (most often
through experimental Hunger or Thirst manipulation) do indeed affect within subject
decision-making, a tenet that needs to be integrated with modern theories.
Homeostasis, in the form in which I intend to use it for this work, is a very old
concept with it’s origin in Biology. According to the theory of Homeostasis, in order to
survive an organism must allocate resources in the defense of the narrow range needed
for survival in the organism’s internal environment (e.g. hydration, blood glucose, blood
sodium, temperature etc.) (Cannon, 1929). Much of this is done automatically within the
organism without any awareness on the part of the organism, however, in complex
organisms homeostasis is also served by directing interaction with the external
environment, including guiding goal-directed action in the defense of the organism’s
desired Homeostatic range (Damasio & Damasio, 2016). It could even be argued that the
ultimate goal of goal-directed action is the defense of homeostasis, and that behavioral

5
changes that are guided by a dynamic internal system in the pursuit of maintaining
homeostasis will best ensure the individual’s continued survival.
In the current work, the focus is going to be on the systematic behavioral and
biological changes associated with imbalances in homeostasis through experimental
manipulation of hunger levels in human subjects. Hunger and it’s antagonist, satiety,
affect the internal milieu in both the CNS and in the periphery (i.e. the rest of the body)
when homeostasis is disrupted. Following meal consumption, ingested calories restores
homeostasis, a process that involves the periphery sending satiety signals to the brain
reflecting the current homeostatic state. These satiety signals are initiated from the
periphery via the release of anorexigenic hormones including insulin, leptin, CCK, PYY,
GLP-1, and GIP in addition to active suppression of the orexigenic hormone Ghrelin (for
review see (Havel, 2001)). Throughout the CNS, introduction of these hormones either
into the bloodstream, or directly into the brain leads to stimulation of hormone-specific
receptors that are distributed throughout the brain. As such, experimental manipulation of
visceral factors related to biological homeostasis, such as increasing hunger or restoring
satiety, have been shown to have widespread effects on a variety of different functional
networks and neurotransmitter systems throughout the brain (for opioids see (Duraffourd
et al., 2012); for opioids and GABA see (Ardianto et al., 2016); for cholinergic neurons
see (Avena, Rada, States, States, & Andes, 2015); for 5-HT see (Voigt & Fink, 2015); for
glutamate see (Shah et al., 2014); for cannabinoids see (Pomorska & Zubrzycka, 2016);
for dopamine see (Araujo, Ferreira, Tellez, Ren, & Yeckel, 2012)).
More generally, when looking across various forms of brain-imaging
methodologies, experimental evidence implies that satiety leads to resting-state down-

6
regulation of various reward-relevant networks and an up-regulation of prefrontal
networks often associated with higher-level cognition (e.g. cognitive control, working
memory etc.). More specifically, post-prandial states tend to decrease resting activity in
the amygdala, orbital frontal cortex, striatum, anterior cingulate (ACC), hippocampus and
surrounding parahippocampus, thalamus and insula, whereas these post-prandial states
are associated with increased activity in the pre frontal cortex (PFC) and inferior parietal
lobule (Li, An, Zhang, Li, & Wang, 2012; Tataranni et al., 1999). In addition to these
changes in resting brain networks, task-based experimental data provides evidence of a
“Spillover” of preference shifts from the food domain, to other reward domains (e.g.
monetary) when homeostasis is disrupted (Fung, Bode, & Murawski, 2017). The
Monetary Incentive Delay task (MID) is a well-validated task that is typically used
during fMRI scanning in order to localize brain networks involved in reward anticipation,
most reliably the ventral striatum (Brian Knutson, Adams, Fong, & Hommer, 2001; Brian
Knutson & Heinz, 2015). Recently, (Simon et al., 2015) provided evidence of both a
shared, and distinct neural circuitry in response to primary and secondary anticipated
rewards using a modified version of the MID task with monetary and food-based
reinforcers. Reward anticipation during the task was associated with increased activity in
the ventral striatum, lateral and medial OFC, and inferior PFC across reinforcer types.
Similarly, neurobiological representation of subjective value during risky choices shares
this similar reward based circuitry in the vmPFC and striatum across reward types (Levy
& Glimcher, 2011). Subjective value during risky choices also has reinforcer specific
activations of the hypothalamus and posterior cingulate cortex (PCC) for food and
monetary reinforcers respectively (Levy & Glimcher, 2011). As is implied by these

7
studies, the neural representation of value needs the ability to compare dissimilar goods
in order to navigate complex environments.
Neuroeconomics has made the study of this internal representation of value a real
focus in recent years, and often describes an interaction between the striatum and the
OFC/PFC (specifically the ventral medial PFC) in the internal representation of reward
value, regardless of reinforcer type enabling the comparison of goods across and within
reward categories (Chib, Rangel, Shimojo, & Doherty, 2009). The up-regulation of both
the ventral striatum and OFC/PFC as the individual falls out of homeostasis and hunger
signals are increased is further supported by animal research showing decreased reward
thresholds in these regions during hunger (Bruijnzeel, Corrie, Rogers, & Yamada, 2011),
which supports a reinforcer invariant shift in reward value. When taken together, this
reinforcer invariant shift in activity and reward thresholds may be the biological
foundation for the myriad of changes in decision making behavior that have been
observed when homeostasis is disturbed through the experimental manipulation of hunger
and satiety.
Despite these conceptual and biological explanations as to how and why
homeostatic imbalance may affect reward responding and decision making more
generally, this field of experimental research is very recent. In a very influential early
paper describing the role of visceral factors in decision making, Loewenstein, 1996 states
“impulsive behavior will occur when visceral factors such as hunger … are intense …
(but) this present-orientation only applies to goods that are related to the visceral factor
… a hungry person would probably make the same choices as a non-hungry person
between immediate and delayed money.” As Loewenstein predicted, homeostatic

8
imabalance (i.e. hunger) does, of course, increase the relative value of food (Raynor &
Epstein, 2003), while also increasing neurobiological biases towards higher calorie foods
during hunger (Goldstone et al., 2009) and behaviorally encouraging people to
impulsively buy more calories (Tal & Wansink, 2013). In addition, hungry people will
tend to choose healthier options for a future date, but when that date arrives people will
often impulsively opt for the unhealthy option (Read & van Leeuwen, 1998). However,
experimental evidence has begun to converge on the notion that not only does
homeostatic imbalance alter domain relevant preferences, but may also affect preferences
across other goods that in the domain of hunger can be broken down in to 4 categories:
Real-world, Social, Risk, and Delay Discounting.
An area of research that is rich with implication and ecological validity despite
being primarily correlational in nature is the effect of hunger on real-world decision
making. In both laboratory and real-world experiments, (Xu, Schwarz, & Wyer, 2015)
provided evidence that hungry individuals are more likely to purchase food-irrelevant
items (e.g. binder clips) when entering a store despite no change in the subjective “liking”
of that item. Despite it’s simple design, this work provides evidence contrary to the
predictions of Loewenstein 1996 in it’s rawest form, that is, hunger promotes the
impulsive purchase of things the individual cannot eat. These results hint that people may
want to be cautious during all hungry shopping, not just grocery shopping. Though
fascinating, the consequences of increased spending when hungry is a relatively
innocuous side effect of homeostatic imbalance, however, correlation evidence implies
that even highly trained decision making is susceptible to homeostatic imbalance. For
example, despite the judicial system being built on the notion of total independence of

9
judicial decisions, such as whether or not to parole an individual, there is correlational
evidence that the duration since a judge’s last food-break correlates with the probability
of a favorable parole decision (Danziger, Levav, & Avnaim-Pesso, 2011). In other words,
as the judge got hungrier he made fewer and fewer favorable parole decisions. Similarly,
the public trading of industrial holdings is built on the foundation of moment-to-moment
and day-to-day independence of stock values, with these values reflecting a time-varying
signal representing a company’s profitability at any given time, with any time-predictable
signal being an exploitable flaw in a Country’s economic structure. In Muslim culture,
however, during the holy holiday of Ramadan (which lasts about a month and occurs
once every year) all devout Muslim individuals must fast while the sun is up. In Middle-
Eastern stock markets where the majority of Stock Brokers are themselves Muslim, some
groups have found decreases in risky trades creating a predictable trend in such markets
during this time of fasting (Bialkowski, Etebari, & Wisniewski, 2012; Seyyed, Abraham,
& Al-hajji, 2005). These studies provide evidence that homeostatic imbalance’s influence
on systematic deviations in decision making has ecological validity during choices of the
upmost importance and gaining a better understanding of how and why should remain an
impetus of the scientific study of decision making.
Moving now to the experimental laboratory setting, homeostatic imbalance,
specifically hunger, has been seen to increase endorsement of social welfare and related
decisions. Early human civilization relied largely on strength in numbers to survive.
Cooperating in attaining food sources when the group is facing starvation likely required
some agreement across various behavioral economic constructs like self-control and risk
aversion. More generally, due to this cooperation-based model for survival in early

10
human civilization, times of increased risk of starvation may increase the endorsement of
sharing resources and often even “tolerated theft” of resources among the group. In line
with this, there is evidence that hunger based disruptions in homeostasis may increase
endorsement of social welfare (Briers, Pandelaere, Dewitte, & Warlop, 2006; Petersen,
Aarøe, Jensen, & Curry, 2014). These preference shifts have been hypothesized as being
mediated specifically by plasma glucose changes, though other feeding related hormones
have not been investigated (Aarøe & Petersen, 2013). Interestingly, despite this increased
endorsement of social welfare amongst hungry individuals, when hungry individuals
complete the dictator game in which they have a chance to turn the endorsement of
sharing assets and fairness during hunger in to action, this increased endorsement of
social welfare does not translate into increased sharing of resources. It appears hunger is
creating a “cheap talk” situation, in which there is increased support of a redistribution of
wealth to those in need when homeostasis is disrupted, despite hungry individuals
reneging on this endorsements and acting only in their own self-interest (Aarøe &
Petersen, 2013). This hypocritical disparity between public endorsement and actual
choice behavior can be conceptualized as an increase in impulsive responding during
homeostatic imbalance as described in (Loewenstein, 1996). More specifically, when an
individual is hungry this will result in increased encouragement of others to share, despite
how when given the chance hungry individuals will selfishly and impulsively distribute
available resources to themselves despite having the power to redistribute resources to
other hungry individuals as they endorsed in the related study. This fascinating
disconnect between words and actions may be the result of alterations in self-control or
impulsivity driven by whether the money was imagined or truly available as a reward.

11
If homeostatic imbalance does indeed increase impulsive responding as discussed
above, a logical extension could be an effect on risk preferences, especially as starvation
draws nearer and the agent is forced to make risky choices that directly influence its
chances of survival. Predatory hunters often live from prey-to-prey in the wild,
consuming large meals when they are able to capture prey, and at times going through
several day periods of fasting in between meals. The pursuit of prey (especially
formidable prey) can be conceptualized as risk taking. As has been formalized in
behavioral models of the risk-taking behavior of predatory animals (Mukherjee &
Heithaus, 2013), a very important factor in deciding when to make riskier decisions is the
hunger state of the animal, which represents how far out of homeostasis the animal is at
the time of decision making. These behavioral models predict that animals will often
display increased risk-taking behavior as they are pushed further away from homeostasis,
and towards starvation.
Current experimental evidence among humans implies that these food-deprivation
based risky choices, which are made specifically in the defense of homeostasis may again
extend to other reward domains. Elicitation of risk aversion across reinforcer types
(including food, water, money) displays a very high correlation within subject (Levy &
Glimcher, 2011). In addition, brain networks involved in risky decision making across
different reward types results in widespread activation of both common and distinct
networks, as discussed above. Among networks responding differentially to reinforcer
type, the hypothalamus was mostly involved in the representation of the subjective value
of options related to risky food decisions, whereas the posterior cingulate (PCC)
represented the representation of the subjective value of risky monetary decisions.

12
Despite this differential response based on reinforcer type in the PCC in the human
neuroimaging literature, the animal neurobiological literature implicates the involvement
of PCC neurons in risky options associated with primary reinforcers (Mccoy & Platt,
2005). Instead of the PCC being money specific, these monkey neurophysiology findings
hint that it is more likely that the posterior cingulate is integrated as part of the neural
networks that play a role in response to risk across reward types, though it may be relied
upon more heavily during evaluation of more abstract reinforcers (like money). In
addition to the PCC, this study also provides evidence that the vmPFC and striatum are
components of common neural networks representing subjective value across reinforcer
types, which are the same item-invariant reward valuation regions described in the last
section, and are in line with the majority of Neuroeconomics literature on the
neurobiological basis of value representation (Chib et al., 2009).
Considering these behavioral and biological similarities among risk preferences
and subjective value representations of gambles across various reward types, if hunger
based disruptions of homeostasis increases risk-taking consistently among animals and
humans when making food-based risky choices, this might again “Spillover” into the
monetary risk-taking domain and lead to similar risk-seeking shifts during times of
increasing hunger. As was previously mentioned, temporary starvation may influence
Middle Eastern stock markets during the holy month of Ramadan, specifically by
decreasing risky stock decisions (Bialkowski et al., 2012; Seyyed et al., 2005). Though
this study suffers from a lack of experimental control (and is confounded by Ramadan
being a religious holiday, a factor that may as well be affecting risk preferences), the

13
ecological validity of this study provides additional support for homeostatic imbalances
influencing monetary risk taking in a behavioral economic setting.
Experimental manipulation of metabolic state and it’s effects on risk aversion is
mixed in the literature, with some prior studies showing increased and others decreased
risk aversion during times of increased hunger signaling (Levy, Thavikulwat, &
Glimcher, 2013; Symmonds, Emmanuel, Drew, Batterham, & Dolan, 2010). Regardless
of the directionality of risk aversion changes, this is not to say that experimental
manipulation of hunger results in poorer risky decisions. To the contrary, there is
evidence in the domain of risky decision making under the condition of ambiguity (the
underlying probability distributions are not known, and must therefore be learned during
the task) using the Iowa Gambling Task (IGT) that deprivation may be associated with
better decision making (Ridder, Kroese, Adriaanse, & Evers, 2014). If hunger is indeed
associated with better decision making under ambiguity, the researchers provide
additional evidence that increased performance on the IGT does not appear to be due to
any systematic increase or decrease in risk-taking under conditions of ambiguity as
measured by the Balloon Analog Risk Task (Ridder et al., 2014).
Even among similar elicitation procedures where the probability distributions of
the options are known (or can be calculated from the information in the question) there
are seemingly conflicting effects of the metabolic states on risk aversion. When risk
preferences are elicited by having participants choose between 2 different rewards with
risk represented by increasing variance in potential rewards, feeding is associated with
decreased risk aversion (Symmonds et al., 2010). In contrast, when risk preferences were
elicited by having participants choose between various gambles and a certain option,

14
hunger was associated with increased risk taking (Levy et al., 2013). A major issue in the
ecological validity of both of these experiments lies in the consequences of losing a
gamble. The situations in which risk preferences became a behavioral and biological
necessity (e.g. the description earlier in this section detailing choice of risky prey among
predators) are often made in the face of increasingly desperate situations, not ones in
which there is a smaller but certain alternative. Considering that the risky choices
considered during times of metabolic imbalance often involve potentially significant loss
and even mortal threat to the decision maker makes replication of the preference shifts
tricky in a lab setting. Estimation of most behavioral models of risk aversion requires
many repetitions, and minimal consequences for gambling and losing creating a potential
barrier between the true nature of the latent process and what can be measured in the lab
setting, which may account for some inconsistent findings in the literature.
As discussed above, directionality of systematic shifts in risk aversion in response
to disruptions of metabolic state remains somewhat elusive in the human literature. Risk
preferences are highly susceptible to context effects (Tversky & Kahneman, 1986), which
leaves open the possibility that slight variations in risk aversion elicitation or some other
anomaly of measuring risky choice in a lab setting is responsible for the inconsistencies
in the literature. Until an experimental examination of the cross-correlation of elicitation-
type and effects of hunger manipulation is completed, this hypothesis about why these
studies have divergent findings remains merely speculation. Replication efforts, or
elicitation of risk preferences using both experimental procedures during manipulation of
metabolic state, may reconcile potential confounding factors in future works.

15
Despite inconsistencies in main effects of metabolic state on risk preferences,
biological and behavioral individual differences have been hypothesized as playing a
significant role determining how hunger affects risk preferences across multiple studies.
(Levy et al., 2013) provides behavioral evidence that individual differences in baseline
levels of risk aversion may play a large role in determining the directionality of hunger
based shifts in risk aversion. Specifically, more risk averse individuals increased risk
seeking behavior when hungry, whereas, more risk seeking individuals display increased
risk aversion when hungry. Biologically, these individual differences in behavioral
effects appeared in this study to be related to higher baseline leptin levels (a hormone
released by adipose tissue and the baseline levels of which likely represents energy stores
(Schwartz, Peskind, Raskind, Boyko, & Porte Jr., 1996) being associated with decreased
risk aversion (Symmonds et al., 2010). In addition, this study found that decreased
suppression of acetyl-ghrelin in response to feeding (an orexigenic hormone secreted by
the stomach, hypothalamus, and limbic regions of the brain) is associated with larger
feeding-based changes in risk aversion. These hormones are, to some degree antagonistic
due to their opposing response to feeding/hunger, so until a study experimentally
manipulates these hormones their role as a biological mechanism remains elusive. These
baseline effects of behavioral and biological traits of the individual illustrate the
importance of metabolic state and individual differences in determining behavioral
response to risky decision making when homeostasis is disrupted.
Investigation into the biological mechanisms mediating risk aversion changes
associated with imbalances in feeding-based homeostasis is currently limited to the
periphery. There has yet to be an investigation into the neural mechanisms associated

16
with hunger-based shifts in risk behavior. Future studies investigating risk aversion shifts
in response to hunger manipulations should combine measurement of plasma changes in
various feeding hormones in combination with risky choice both under ambiguity and
under known risk. In line with the available neurobiological research on risky decision
making, hunger will likely be associated with increased neural response in regions
tracking the homeostatic and interoceptive status of the body (e.g. hypothalamus and
insula respectively), and increased neural responsivity in subjective value tracking
regions (e.g. the OFC and striatum). In addition, there may be increased functional
connectivity among these and other risk-relevant brain regions (i.e. PCC and
hypothalamus). This is a promising area for future research, and furthering scientific
knowledge in this area will help create a more complete profile for the neurobiology of
homeostasis and its relationship to behavioral economic preferences.
Thus far we have examined the experimental and correlational evidence implying
that homeostatic imbalance (specifically hunger) affects real-world decisions, encourages
selfish distribution of resources and decreases risk aversion. In addition, experimental
evidence hints that hunger’s encouragement of impulsive responding manifests itself as
increased delay discounting, which will be the focus of the experimental sections of this
work. This effect can be conceptualized as an extension of Aesop’s Fable of the
hedonistic grasshopper, and the patient ant. While the ant focused on planning for the
winter by storing food, the grasshopper focuses on satisfying only his immediate desires.
When winter inevitably comes, the grasshopper dies from starvation whereas the ant
survives thanks in large part to her earlier self-control. If, however, a severe drought was
to hit near the end of summer, would it behoove the ant to continue her budgeting ways,

17
or should the ant instead deplete her resources to stave off starvation? In the face of
severe imbalances of homeostasis, the priorities of goal-directed action need to be
reevaluated (Damasio & Damasio, 2016), and depending on the level of deprivation, the
homeostatic imbalance can reach a point where no alternative reward can match the
subjective utility of a return to homeostasis (as is described in (Loewenstein, 1996)). In
line with this, Mayack & Naug, 2015 have demonstrated that even short periods of
starvation can increase the choice of smaller immediate meals in place of larger delayed
meals in the simple honey bee. More generally, hunger induced self-control struggles
extends beyond just the domain of food-preferences, with the available experimental
evidence implicating increased immediacy bias, increased convexity of utility functions,
and increased discounting of future rewards.
If hunger modifies goal-directed action towards immediate consumption across
reward domains, increases in monetary delay discounting should be observed within
subject when measured via the administration of an intertemporal choice task. As
discussed above, experimental evidence hints that in addition to hunger signals altering
valuation of food (Xu et al., 2015), hunger also appears to alter the reward valuation of
homeostasis irrelevant objects and prizes. If the individual undergoes a disruption of
homeostasis and this mobilizes the brain’s reward neurocircuitry to be hypersensitive to
reward valuation or to be focused on immediate consumption, it stands to reason that this
could be accompanied by increased discounting of future rewards, and indeed the
experimental evidence in large part supports this notion.
Experimentally induced feelings of hunger encourages steeper discounting of
future rewards and increases immediacy bias in the lab setting (Ashton, 2015; X. T.

18
Wang & Dvorak, 2010), though at least one study does not replicate these findings
(Ridder et al., 2014). The majority of experimental psychology and behavioral economics
literature takes interest specifically in estimating and characterizing state-based changes
in discount functions, which are often assumed to take a hyperbolic shape (often
abbreviated with k (K. N. Kirby, Petry, & Bickel, 1999)) (e.g. (X. T. Wang & Dvorak,
2010)). These single-parameter models of delay discounting implicitly assume that all
inter-subject variation reflects differences in discounting alone. However, differences in
the form of the utility function for amount could also drive individual differences
observed on these tasks. For example, individuals that prefer an immediate $100 over
$150 in four months may do so because they discount steeply. But, alternatively, they
may discount modestly, but exhibit steep diminishing marginal utility so that their
valuation of $150 is not substantially greater than their valuation of $100 (Ho, Wogar,
Bradshaw, & Szabadi, 1997; A. Pine, Shiner, Seymour, & Dolan, 2010; Alex Pine et al.,
2009). In it’s simplest form, hunger may specifically be targeting these monetary utility
functions through ‘Spillover’ from it’s clearer role in how hunger affects food
preferences. When the individual is hungry and is nearing the point of starvation, the
utility of 1 banana from nothing would far outweigh the added utility of a second banana
from one banana. In other words, there is dramatically diminishing marginal returns for
food items after the first food item when the individual is near starvation. Both the
discounting and utility interpretations of the effects of hunger on delay discounting
warrant consideration (Andersen, Harrison, Lau, & Rutström, 2008), though this
dissertation sticks only to the hyperbolic model assuming a linear utility function.

19
Some behavioral economics models of delay discounting enable the dissociation
between state-based changes in the discount function and the utility function, among the
tasks that can dissociate discounting and utility is the convex budget task. In this
experimental variant of the traditional intertemporal choice task (in which a binary choice
is made between some smaller sooner amount, and some larger later amount), the
individual is asked to allocate a budget of tokens between an early and delayed date.
Allocating a coin to each payoff time point is paid off with a different value per token
(conceptualized as an interest rate for the delayed option), and the individual may allocate
coins between the two options in accordance with their preferences. For instance, a
participant might be asked to divide 10 tokens between a now pay-off and a pay-off in
one month, with each token allotted to now worth $2 and each token allotted to 1 month
pay-off worth $2.50. Choices of a complete payoff on either the earlier or later time point
is referred to as a “corner strategy”, whereas some allocation of coins among the two time
points is referred to as an “interior solution”. Using this experimental design, (Ashton,
2015) experimentally manipulated homeostasis via hunger and satiety conditions (in
addition to cognitive load) and found that hungry individuals tended to cash tokens
earlier, an effect further enhanced if the earlier time point was today (referred to as an
immediacy bias). In contrast to the discount function, there was no evidence of increased
curvature of the utility function when compared to the control conditions implicating that
homeostasis disruptions are specifically associated with increased delay discounting
likely via an increased immediacy bias with a similarly shaped utility function in this
study. Utilizing a similar procedure, however, Kuhn, Kuhn, & Villeval, 2014 found that
when compared to the consumption of a sugared drink, deprivation was associated with

20
decreased convexity of the utility function a finding that is supported by the risk taking
literature (Levy et al., 2013). In line with this steeper discounting and increased
immediacy bias on the hunger day, hungry individuals have been seen to respond faster
to intertemporal choice sets when homeostasis is disrupted, further implicating increased
reward-value of the intertemporal choice questions (Wierenga et al., 2015).
As discussed in greater detail earlier in this chapter, meal consumption leads to
satiety signaling via a myriad of central and peripheral biological changes that indicate
the restoration of homeostasis. Experimental evidence manipulating hunger and satiety in
conjunction with elicitation of intertemporal choices has provided initial evidence of a
potential mediating effect of plasma glucose levels underlying metabolic state based
alterations in intertemporal choice preferences. Specifically, (X. T. Wang & Dvorak,
2010) provided evidence that consumption of carbohydrates alone (specifically glucose)
is enough to mirror behavioral effects seen following ingestion of an entire meal, a
finding that is partially corroborated by (Kuhn et al., 2014). Ingestion of glucose induces
sharp spikes in plasma glucose, and Wang & Dvorak hypothesize that this plasma change
mediates hunger-based shifts in delay discounting. In addition, the investigators
hypothesized that the increased change in delay discounting seen amongst higher BMI
individuals in response to consumption was mediated by plasma glucose changes, and
that the previously reported association between higher BMI and greater delay
discounting is likely also mediated by systematic plasma glucose changes (Weller, Cook,
Avsar, & Cox, 2008). However, this mediation effect is challenged by the results of
(Kuhn et al., 2014) in which consumption resulted in decreased delay discounting
following ingestion of any drink (caloric or not). Furthermore, when directly comparing

21
subjective hunger and plasma glucose levels, Skrynka & Vincent, 2017 provide evidence
that hunger better mediates increased impulsive responding than does plasma glucose
levels. In line with this, animal research has implicated that central stimulation of the
VTA by ghrelin (the only hormone that increases in plasma as subjective hunger
increases) is responsible for increased impulsive responding when hungry (Anderberg et
al., 2015), which may implicate that hunger’s influence on impulsivity is likely due to a
combination of biological post-ingestive factors, not a single biological marker like
plasma glucose.
Only one study to date has investigated the neural correlates of manipulating
hunger and satiety on intertemporal choice preferences (Wierenga et al., 2015). Though
we did not observe shifts in delay discounting in response to food deprivation, the
behavioral paradigm used was not optimal for detecting state-related shifts in
discounting. Specifically, we used a fixed set of questions spanning a wide range of
trade-offs. Because inter-subject variability in delay discounting is quite high (spanning
orders of magnitude in model parameter fits, see (Peters & Bu, 2011)) the range of
questions asked in fixed-set discounting procedures is accordingly wide. As a result,
subtle state-related shifts in behavior cannot be observed. Procedures that individualize
questions to the participant’s level of discounting (Shan Luo, Ainslie, & Monterosso,
2014) are more sensitive to state-related changes. Moreover, while main effects of delay
discounting were not observed by Wierenga and colleagues, reaction time and fMRI data
supported the assertions of (Ashton, 2015) of an increased immediacy bias in hungry
individuals via increased reward-circuitry activation (right ventral striatum, dorsal
anterior caudate, rostral and dorsal anterior cingulate, posterior cingulate) to questions in

22
which the earlier option was to be paid immediately, and increased cognitive control
circuitry activation (bilateral insula, bilateral middle frontal gyrus, and vlPFC) to all types
of intertemporal choice questions on the satiety day.
Taken together, the experimental evidence seems to imply that hunger leads to
increased delay discounting, increased immediacy bias, and perhaps also increased
convexity of the utility function. This increased motivational value of sooner options
following experimental manipulation of hunger is supported by faster reaction times and
increased activation of reward circuitry when options are available immediately. Though
identification of a specific peripheral or central mechanism is challenging due to the
complexity of central and peripheral changes in response to disruption of homeostasis,
preliminary evidence indicates that peripheral signals of disrupted or restored
homeostasis are integrated in the brain to modify goal-directed action accordingly.
To date much of the focus of feeding based alterations in preferences has focused
on food preferences (e.g. (Tal & Wansink, 2013; Wansink, Tal, & Shimizu, 2012)) and
functional neurobiology in response to food stimuli when homeostasis is disturbed (e.g.
(Mehta et al., 2012; Uher, Treasure, Heining, Brammer, & Campbell, 2006)).
Experimental work aimed at determining the mechanism of action of how homeostatic
imbalance (specifically hunger) can lead to “Spillover” effects across homeostasis
irrelevant reinforcers in humans is lacking in the literature. This dissertation will
hopefully be instrumental in linking animal and human research that implicates hunger-
based decreases in reward thresholds as a potential mechanism of action with the
behavioral economics and neuroimaging work in healthy human subjects.

23
Chapter 2: Intertemporal Monetary Incentive Delay Task
Background
In decision neuroscience, most modern theories of the neurobiology of choice rely
on the presence of value tracking networks in the brain (e.g. Rangel, Camerer, &
Montague, 2008). Value is often conceptualized as roughly linear when a monetary
reinforcer is used to track value, however, primary rewards (e.g. food or drinks) are often
used in this literature as well. The tracking of value for these primary rewards can be
done through estimation of a willingness to pay for any given food or drink which is
thought to be tracked in the striatum and the medial OFC (Plassmann, O’Doherty, &
Rangel, 2007), or through revealed preferences through repeated choices among available
rewards. An important question in the value-tracking literature is whether the functional
architecture for primary rewards is overlapping or separate from monetary rewards.
In the animal literature, hunger based changes in dopaminergic reward thresholds
in the limbic system can be measured directly using single cell recordings. In the human
literature, reward sensitivity and value tracking can be investigated through the use of
fMRI in conjunction with a task that systematically changes values and rewards. Among
these types of tasks, arguably the most successful task in gaining experimental control
over the neurobiology of value tracking and reward thresholds creating a situation where
values are tracked and reward thresholds can be tested is the Monetary Incentive Delay
(MID) developed originally by (B. Knutson, Westdorp, Kaiser, & Hommer, 2000). This
task is very simple, yet elegant in it’s efficiency estimating value-tracking in the brain.
Each trial starts out with a fixation cross, which is then followed by a reward cue. Each
reward cue represents a different valued reward and represents what the individual is

24
“playing for” on that particular trial. The period of time in which the reward cue is
showing is known as the reward anticipation period, and is primarily what is being
modeled in the brain for value tracking. As soon as the cross comes back up after the
reward anticipation period the individual responds as quickly as possible. The success
rate of the individual is titrated based on their distribution of response times so the
individual wins the prize around 2/3 of the time. As the rewards get bigger or smaller this
varying value is tracked during reward anticipation in the Nucleus Accumbens (Brian
Knutson et al., 2001) and in the mesial Prefrontal Cortex (B. Knutson, Fong, Bennett,
Adams, & Hommer, 2003).
One of the foundational hypotheses of this dissertation is the presence of a
‘Spillover Effect’ between the food reward system and the monetary reward system,
specifically when homeostasis is disrupted in the food reward system this will have
impulsivity effects on rewards in general. For this assumption to hold, the neurobiology
of food and monetary rewards would likely need to share some network nodes for there
to be a biological basis for ‘Spillover’. Simon et al., 2015 compared monetary and food
rewards during a modified MID task. Results indicated that both types of task activated
the ventral striatum during anticipation of reward. These findings provide a potential
mechanism by which valuation of rewards may result in a neurobiological ‘Spillover
Effect’, namely value tracking in the ventral striatum. Though without specific
experimental manipulation, ‘Spillover’ cannot be dissociated from just a singular value-
tracking across reward types.
The original version of the MID task offers only simple food or monetary rewards
often looking for linear relationships between the increase in offered monetary reward

25
values and similar increases across value-tracking regions in the brain. Intertemporal
Choice can be conceptualized as a decision between larger monetary value and shorter
delay to reward. Similar to the traditional MID in which larger monetary rewards are
associated with decreased response times, past work has implicated that shorter delays
and larger rewards are both also associated decreased response times (S. Luo, Ainslie,
Giragosian, & Monterosso, 2009). In addition, the fMRI contrast comparing immediate
vs delayed rewards during reward anticipation showed increased activation across the
limbic system, specifically in the caudate, putamen, insula, pallidum, supramarginal
gyrus and anterior cingulate cortex. The magnitude tracking contrast, comparing high and
low reward magnitudes in the intertemporal variant of the MID task similarly showed
increased limbic system activity in the caudate, thalamus, lateral occipital cortex and
occipital pole.
One of the foundational hypotheses of the current work is that the disruption of
homeostasis through the experimental manipulation of hunger affects delay discounting
through hunger’s actions altering reward sensitivity in general. Through the experimental
manipulation of hunger within subject, the effects of hunger on magnitude sensitivity and
delay sensitivity can be investigated through the use of the intertemporal variant of the
MID task. If hunger increases reward sensitivity then there should be increased reward-
related brain activity to all rewards when compared to baseline. If hunger is associated
with increased immediacy bias or delay effects this may lead to increased reward related
activity and decreased RT when rewards are available immediately when compared to
rewards available in 1 month. Finally, if hunger is associated with alterations in the

26
convexity of utility functions then the hunger day may be associated with decreased
differential response to larger sized monetary rewards across delays.
Methods
Subjects
A total of 28 right handed subjects were enrolled in the study. Of these 28
subjects, 27 of them completed the behavioral session. Of those 27 subjects, 5 were
excluded due to the subject being outside of the healthy BMI range required to enroll in
this study. All 22 of the remaining subjects completed their first fMRI session. Two of
the remaining 22 subjects did not show up for their second fMRI session leaving 20
subjects who completed all sessions of the study. Examination of behavioral data
indicated that one subject either did not understand the instructions, or was responding
randomly across all of the tasks so this subject was excluded from all behavioral and
fMRI analysis. Among the 19 subjects who completed all sessions of the study, and were
included in the analysis, the mean age was 21.32 (SD = 3.34), 10 of the 19 study
completers were women, and the mean number of years of education across the sample
was 14.68 (SD = 2.79). All participants received a $25 show-up fee for the behavioral
session, and received a $50 show-up fee for each of the fMRI sessions. In addition, on
each fMRI session subjects were informed that one randomly selected trial from each day
would be carried out as bonus payment for the subject. They were told all tasks they
completed on the scan day was eligible for random selection to ensure real choices were
made across all tasks.
Intertemporal MID Task
Participants completed two runs of a 75 question intertemporal variant of the

27
Monetary Incentive Delay task (MID) paired with fMRI. At the onset of each trial of the
MID task, one of five symbols was presented, indicating the potential prize for the round
($5 Today, $25 Today, $5 in 1 month, $25 in 1 month, or no prize). Participants learned
the pairing of symbols with rewards prior to fMRI testing, through a brief training
protocol used by our team in similar studies (Shan Luo, Ainslie, Giragosian, &
Monterosso, 2009). During each trial, after the possible prize is presented, participants
wait a random time iteration between 3.5 and 4.5 sec for a target to appear, this time-
period is referred to as the reward anticipation period. When the target appears, the
participant responds as quickly as possible to maximize their likelihood of winning that
trial. Feedback indicates whether their response is fast enough to win the prize indicated
for the round. At the end of each of two task runs, one trial is selected to be “real”, and if
the participant won on that trial, he or she gets the specified prize (delivered via Amazon
Allowance). If the participant did not win on the randomly selected trial, no bonus prize
is received for that run. In this way, participants are incentivized to try to win each round,
but the magnitude of the incentive is a function of his or her valuation of the available
prize for the round (see Figure 1).

28
Figure 1. Visualization of the custom-designed intertemporal variant of the monetary incentive delay task.
Participants started by seeing a fixation cross for a mean of 2 seconds (exponential distribution with a mean
of 2 in an attempt to optimize experimental design with fMRI scanning). Participants then saw 1 of 5
different reward cues for a randomized duration between 3.5-4.5 seconds. The four reward cues included
$5 Today, $5 in one month, $25 Today and $25 in one month, and as a control condition, there was a fifth
cue that indicted the current trial would not be rewarded. When the blue ‘X’ appeared over the reward cue,
participants were told to respond as quickly as possible to ensure they have the best chance at winning the
available reward. Winning was kept at approximately 60% in line with past works (Brian Knutson et al.,
2001).

Behavioral Analysis
Feeding Manipulation Check
On both scanning days of the experiment subjects arrived following a minimum
of a 12 hour fast, on the Fed day participants were provided with a personal sized pizza
and asked to eat until they felt “comfortably full”. As such, the experimental
manipulation requires that self-reported hunger decreases on the Fed day, and self-
reported satiety increases on the Fed day to consider the experimental manipulation a
success. These hypotheses were tested by utilizing the BAMBI package in python to fit
Jittered ITI (exponential mean=2s)
Response Period
(0.1-0.5s)
Outcome
(1.5-2s)
Reward Anticipation
(3.5-4.5s)
Practice
$5
In a Month
$25
Today
$25
In a Month
$5
Today
No
Reward

29
hierarchical regression models utilizing and MCMC procedure through the PYMC3
package.
Collapsed Reward Cues Reaction Time Analysis
Hierarchical Bayesian models were fit using an MCMC algorithm through the
BAMBI python package. To test for this interaction between Cue type (all rewards vs. no
reward cue) and hunger status on response time, a Hierarchical Bayesian was fit via the
MCMC chains and posterior distributions were generated from a naïve prior. To test for
significance a threshold of 95% High Probability Density was set for model parameters
of interest.
Expanded Reward Cues Reaction Time Analysis
Similar to the last section, a Hierarchical Bayesian model was fit using an MCMC
algorithm again through the BAMBI python package. These models were fit to examine
if the interaction effects between reward cue and hunger status is dependent on either
magnitude or delay of reward cue. This was done by specifying reward type (immediate
vs. delayed; $5 vs. $25) in the fitting of the interaction term for predicting response times.
Significance for the posterior distribution for the interaction effects was set at a threshold
of 95% High Probability Density.
MRI Data Acquisition
Neuroimaging data was collected using a 3T Siemens MAGNETOM Tim/Trio
scanner at the Dana and David Dornslife Cognitive Neuroscience Imaging Center at the
University of Southern California with a 32-channel head-coil. Participants laid supine on
a scanner bed, viewing stimuli through a mirror that is mounted on the head coil. Blood
oxygen level-dependent (BOLD) response was measured via an echo planar imaging

30
(EPI) sequence with PACE (prospective acquisition correction) turned on (TR=2s,
TE=25ms, flip angle=90, resolution=3mm
3
isotropic, 64 x 64 matrix in FOV=192mm). A
total of 41 axial slices, each 3mm in thickness were acquired in an ascending interleaved
fashion to cover the whole brain. The slices were tilted to align the axial slices with each
individual’s AC-PC plane to minimize signal dropout in the orbitofrontal cortex
(Deichmann, Gottfried, Hutton, & Turner, 2003). Anatomical images were collected
using a T1-weighted three-dimensional magnetization prepared rapid gradient echo
(MPRAGE with TI=900ms, TR=1.95s, TE=2260ms, flip angle=9 degrees,
resolution=1mm, 256 x 256 matrix in FOV=256mm) for alignment of each individual’s
structural space to standardized space. These scans were co-registered to the individual’s
mean EPI images.
fMRI Analysis
Preprocessing
fMRI data preprocessing was conducting utilizing the python-based neuroimaging
data processing package nipype (Gorgolewski et al., 2011) to call preprocessing functions
from FSL, a part of the Oxford University Centre for functional MRI of the Brain
(FMRIB) Software Library (www.fmrib.ox.ac.uk/fsl). Each individual’s structural and
functional data was transformed to MNI space by first linearly transforming structural
image to MNI space, and linearly transforming EPI to structural space via FSL’s FLIRT
function (Jenkinson, Bannister, Brady, & Smith, 2002; Jenkinson & Smith, 2001). FSL’s
FNIRT was then used to non-linearly transform the structural data to standard MNI
space. The functional images were motion and slice-time corrected. They were then
spatially smoothed using a Gaussian Kernel with a full width at half-maximum of

31
5mm, and were submitted to a high-pass temporal filter using a filter width of 120s.
Whole Brain Analysis
As has been recently pointed out, cluster correction for whole-brain voxelwise
analysis has received some criticism for potentially having inflated type-1 errors when
utilizing any of the major fMRI analysis softwares (Eklund, Nichols, & Knutsson, 2016).
In light of these cluster correction issues, whole brain analysis were carried out using
nonparametric permutation inference via FSL’s randomise algorithm (Winkler, Ridgway,
Webster, Smith, & Nichols, 2014). These methods are used for inference on static maps
when the null distribution is not known. In order to reduce the number of comparisons
made, a gray matter mask was applied prior to the running of randomise since signal
changes in white matter are likely to be Type 1 error. The raw statistic contrast maps
were then converted to cluster-like activity in contiguous voxels via Threshold-Free
Cluster Enhancement. Corrected p-value maps were then thresholded at 0.05.
ROI Analysis
My approach to investigating my a-priori regions of interest was to use the same
strategy taken as in the 2009 Journal of Neuroscience Paper by Shan Luo et al.,. They
took their ROIs from those previously associated with the MID task (Brian Knutson et
al., 2001; Brian Knutson, Taylor, Kaufman, Peterson, & Glover, 2005). This consisted of
recreating 4 spheres of 5mm diameter of the bilateral anterior Insula, the right
supplementary motor area (SMA) and midbrain based on the coordinates used by (S. Luo
et al., 2009).Whereas subcortical structural ROIs were taken from the probabilistic
Harvard-Oxford Subcortical Structural Atlas, where the bilateral putamen, caudate,
thalamus and accumbens were defined as separate ROIs.

32
Results
Hunger Manipulation
As expected there was a significant reduction in self-reported hunger ratings
following feeding on the Fed day (see Figure 2 and Table 1 below).
Hierarchical Bayesian Hunger Chains and Estimates
Mean SD 0.95 HPD
lower
0.95 HPD
upper
Effective n Gelman
Rubin
Fast Arrival -4.91 5.61 -16.21 5.92 10206 0.99
Fast Scan 1
(Intercept)
63.68 6.78 50.76 77.17 5685 1.00
Fast Scan 2 14.78 5.53 3.67 25.44 10098 1.00
Fed Arrival 52.69 8.07 36.98 68.55 9171 0.99
Fed Scan 1 -54.61 5.86 -66.30 -43.29 8330 1.00
Fed Scan 2 0.06 8.07 -14.96 16.45 9457 1.00
Table 1. Outputs of the hierarchical Bayesian estimation of the mixed-effects model used to estimate the
effects of the feeding manipulation on self-reported hunger. Fasting was associated with little to no change
during the experimental procedure, whereas self-reported hunger was reduced an average of 54.6 points
following feeding.

33
Figure 2. Self-reported hunger levels split apart by experimental condition (i.e. fast and fed). Feeding was
associated with a large drop in self-reported hunger ratings immediately after eating with this hunger
reduction remaining significantly below the fasted condition up to an hour after feeding.

Similarly, there was a significant increase in self-reported fullness on the Fed day
following meal consumption (see Table 2 and Figure 3 below). Taken together, these
findings strongly imply that the experimental hunger manipulation (12 hour fast and then
continued fasting or feeding to comfortable satiation) was a success.
Hierarchical Bayesian Fullness Chains and Estimates
Mean SD 0.95 HPD
lower
0.95 HPD
upper
Effective n Gelman
Rubin
Fast Arrival -4.91 5.61 -16.21 5.92 10206 0.99
Fast Scan 1
(Intercept)
7.64 6.00 -3.80 19.63 2502 1.00
Fast Scan 2 -3.21 5.34 -13.68 7.11 5897 1.00
Fed Arrival 3.82 5.27 -6.51 14.07 5867 1.00
Fed Scan 1 68.91 4.93 59.70 78.78 4570 1.00
Fed Scan 2 -16.65 7.07 -30.47 -2.73 5303 1.00
Table 2. Outputs of the hierarchical Bayesian estimation of the mixed-effects model used to estimate the
effects of the feeding manipulation on self-reported satiety. Fasting was associated with little to no change
during the experimental procedures, whereas self-reported satiety was increased an average of 68.9 points
following feeding.

34

Figure 3. Self-reported satiety levels split apart by experimental condition (i.e. fast and fed). Feeding was
associated with a large increase in self-reported satiety immediately after eating with this satiety increase
remaining significantly above the fasted condition up to an hour after feeding.

Collapsed Reward Cues Reaction Time Results
In this first model, all rewarded cues were collapsed to examine if there was a
significant interaction between Cue type and homeostatic condition. Results of the
Hierarchical Bayesian model indicate that there is a significant interaction between cue
type (rewarded vs. non-rewarded) and hunger status on RT, specifically there is a
speeding of response times on the fasted day in response to rewarded cues
(M[Day(Fed)*Cue(Reward)]=0.02; 95% HPD [0.003, 0.03]; see Figure 4 and Table 3).

35

Figure 4. Visualization of the posterior distributions generated from the hierarchical Bayesian model that
shows there is a significant decrease in response times to rewarded cues on the fasted day only
(M[Day(Fed)*Cue(Reward)]=0.02; 95% HPD [0.003, 0.03]). The figure above is comparing rewarded and
unrewarded RTs by collapsing across the 4 different reward cue trial types ($5 Today, $25 Today, $5 in
one month, $25 in one month) and comparing these trials with the non-rewarded trial response times.

Hierarchical Bayesian MID RT Chains and Estimates
Mean SD 0.95 HPD
lower
0.95 HPD
upper
Effective
n
Gelman
Rubin
1|Subject SD 0.04 0.009 0.03 0.06 1194 1.00
Cue Duration -0.025 0.004 -0.03 -0.02 5026 1.00
Day (Fed) -0.011 0.006 -0.02 0.001 20000 0.99
Day(Fed) x Reward
Anticipation (Reward)
0.016 0.007 0.003 0.03 20000 0.99
Intercept 0.401 0.022 0.36 0.44 3233 1.00
RT SD 0.094 0.001 0.09 0.10 19574 1.00
Reward Anticipation
(Reward)
-0.067 0.005 -0.08 -0.06 20000 1.00
Run 0.006 0.001 0.004 0.008 20000 1.00
Trial 0.001 0.0001 0.001 0.001 20000 0.99
Table 3. Outputs of the hierarchical Bayesian estimation of the mixed-effects model used to estimate the
effects of the feeding manipulation on reward-cue condition reaction time. This model provided evidence
of a significant interaction between feeding condition and reward type. Hunger was associated with a
speeding of an average of 0.067 seconds when cues were rewarded, whereas this speeding was slowed
down by an average of 0.016 seconds on the fed day.

Expanded Reward Cues Reaction Time Results

36
This analysis is a further extension of the last model in which the interaction
between cue type and homeostatic state is extended to estimate a coefficient for each of
the 5 different cue types. Results indicate that the interaction between hunger condition
and cue condition is driven by the difference between rewarded and non-rewarded cues,
not necessarily by any specific delay or magnitude of reward cue (see Figures 5 & 6 and
Table 4).

Figure 5. Visualization of the posterior distributions of the hierarchical Bayesian model of the different
reward cue conditions for the Fasted day only. As can be seen, all 4 rewarded cue conditions were
associated with decreased response times when compared to the non-rewarded condition on the fasted day.

37

Figure 6. Visualization of the posterior distributions of the hierarchical Bayesian model of the five different
reward conditions on the fed day only. The data provided evidence that there was no difference in either
direction of reaction times when comparing any rewarded cue condition with the no-reward cue condition.

38
Expanded Hierarchical Bayesian MID RT
Chains and Estimates
Mean SD 0.95
HPD
lower
0.95
HPD
upper
Effective
n
Gelman
Rubin
1|Subject SD 0.04 0.009 0.03 0.06 1279 1.001
Condition ($25 Month) -0.06 0.006 -0.07 -0.05 14895 1.001
Condition ($25 Today) -0.07 0.006 -0.08 -0.06 15787 1.001
Condition ($5 Month) -0.06 0.006 -0.07 -0.05 15128 1.001
Condition ($5 Today) -0.07 0.006 -0.08

-0.06 15284 1.001
Cue Duration -0.03 0.004 -0.03 -0.02 3875 1.000
Day (Fed) -0.01 0.006 -0.02 0.001 11056 1.001
Day (Fed) x
Condition ($25 Month)
0.02 0.009 0.003 0.037 14555 1.000
Day (Fed) x
Condition ($25 Today)
0.01 0.008 -0.006 0.027 14395 1.000
Day (Fed) x
Condition ($5 Month)
0.01 0.009 -0.005 0.029 14233 1.000
Day (Fed) x
Condition ($5 Today)
0.02 0.008 0.004 0.037 13982 1.001
Intercept
(Fast Practice Condition)
0.40 0.02 0.357 0.441 2604 0.999
RT SD 0.09 0.001 0.092 0.095 18148 1.000
Run 0.006 0.001 0.004 0.008 20000 1.000
Trial 0.001 0.001 0.001 0.001 20000 0.999
Table 4. Outputs of the hierarchical Bayesian estimation of the mixed-effects model used to estimate the
effects of the feeding manipulation on reward-cue condition reaction time. This model is an expansion from
the model shown in Table 3. This model provided evidence that fasting was associated with a speeding of
all rewarded cues regardless of magnitude or delay, whereas the feeding condition was not associated with
a speeding of any reward cue conditions.

RT Magnitude Effects
Direct comparison of RT distributions for the magnitudes of $25 and $5 between
the homeostatic conditions of hunger and satiety show no significant differences (M
[Fed:$5] = 0.0014, 95% HPD = [-0.009, 0.0122]).
RT Delay Effects

39
Direct comparison of RT distributions for the reward delays of Today and 1
month between the homeostatic conditions of hunger and satiety show no significant
differences (M[Day(Fed)*Delay(Today)] = -0.0001, 95% HPD = [-0.011, 0.0107]).
fMRI Results
Main Effects
$25 Cues > $5 Cues
Collapsing across days, there was a main effect of reward magnitude in the
Occipital Lobe (p<0.001) mainly in early visual cortices (see Figure 7 and Table 5).

Figure 7. Whole-brain activation map for the main effect of $25 cues greater than $5 cues across both days.
Significant activation for this contrast was evident only in the visual cortex. Thresholding for this map was
set to p<0.05 using the corrected p-value output of FSL’s randomise permutation algorithm that utilizes
threshold-free cluster enhancement to search for ‘cluster-like’ behavior in voxels.

Cluster Voxels Max
(1-p)
Side Brain Region Max X Max
Y
Max Z
1 20572 >0.999 L/R Visual Cortex 49 19 24
Table 5. Significant whole brain cluster for the main effect of the $25 > $5 contrast averaged across the two
experiment days. Only a single cluster met significance in the visual cortex with the minimum p-value for
this cluster reaching p<0.001.

Today Cues > 1 Month Cues
There was a significant main effect of delay when collapsing across hunger
conditions in the caudate, temporal gyrus, bilateral thalamus, several bilateral anterior
Insula clusters, visual cortex and cerebellum (see Figure 8 and Table 6).

40
Figure 8. Whole-brain activation map for the main effect of Today cues greater than 1 month cues across
both days. Significant activation for this contrast appeared in the striatum, insula, visual cortex and
cerebellum. Thresholding for this map was set to p<0.05 using the corrected p-value output of FSL’s
randomise permutation algorithm that utilizes threshold-free cluster enhancement to search for ‘cluster-
like’ behavior in voxels.

Cluster Voxels Max
(1-p)
Side Brain Region Max
X
Max
Y
Max Z
1 12787 >0.999 L/R Visual Cortex &
Temporal Gyrus
60 47 20
2 1212 0.988 L/R Bilateral Thalamus &
Caudate
53 47 34
3 1149 0.992 R Anterior Insula 25 73 33
4 55 0.961 L Anterior Insula 3 62 69 30
5 29 0.956 L Anterior Insula 2 61 70 37
6 5 0.952 L Anterior Insula 60 77 32
7 3 0.951 R Visual Cortex 17 23 32
8 2 0.951 R Cerebellum 19 37 19
9 1 0.95 R Caudate 39 71 42
Table 6. Significant whole brain cluster for the main effect of the Today > 1 Month contrast averaged
across the two experiment days. There were significant main effects across the striatum, insula, visual
cortex and cerebellum. Thresholding for this map was set to p<0.05.

Reward Cues > Baseline

41
Investigation of the main effect of rewarded cues greater than baseline collapsing
across hunger status resulted in significant activation of the occipital lobe, specifically in
early visual cortices (see Figure 9 and Table 7).
Figure 9. Whole-brain activation map for the main effect of Reward cues greater than baseline across both
days. Significant activation for this contrast appeared only in the visual cortex. Thresholding for this map
was set to p<0.05 using the corrected p-value output of FSL’s randomise permutation algorithm that
utilizes threshold-free cluster enhancement to search for ‘cluster-like’ behavior in voxels.

Cluster Voxels Max
(1-p)
Side Brain Region Max X Max
Y
Max Z
1 9995 >0.999 L/R Visual Cortex 28 23 23
Table 7. Significant whole brain cluster for the main effect of reward cues > baseline contrast averaged
across the two experiment days. Only a single cluster met significance in the visual cortex with the
minimum p-value for this cluster reaching p<0.001.

Hunger Differences
Reward Cues > Baseline Day Differences
No significant clusters for this contrast meet thresholding.
$25 Cues > $5 Cues Day Differences
No significant clusters for this contrast meet thresholding.
Now Cues > 1 Month Cues Day Differences
No significant clusters for this contrast meet thresholding.
ROI Analysis
Practice Cues – Implicit Baseline Contrast

42
Plotting out the a-priori contrasts of interest led to an abnormal discovery, there
were very large negative fluctuations when comparing the no reward cue with implicit
baseline. Further investigation of these abnormal coefficients led to the discovery of large
differences across limbic ROIs due to metabolic state. Specifically, the Fed condition is
associated with a large drop in brain response (relative to implicit baseline) during the no
reward anticipation period (M[Fed] = -7.925, 95% HPD = [-10.302 - -5.532]; see Figure
10). Due to this complication, the primary contrast of interest used to examine the effects
of metabolic state on ROI response across all rewarded cue anticipation periods was
compared with the implicit baseline of the model, not the no reward anticipation period.
It is worth noting, however, that the use of the implicit baseline instead of the control
condition is not ideal. There could be differences in baseline response in regions of
interest, future works should explore whether there are indeed network differences in
baseline response due to hunger status so an ideal contrast can be designed.
Figure 10. Limbic ROI fast – fed difference of No Reward cues greater than baseline. Significant activation

43
for this contrast appeared across bilateral nucleus accumbens and bilateral caudate. The presence of
significant activation was determined through the fitting of a hierarchical Bayesian model across all ROIs
(M[Fed] = -7.925, 95% HPD = [-10.302 - -5.532]).

All Reward Cues > Baseline
There is a significant main effect of hunger condition when comparing all
rewarded cue anticipation periods with baseline across the combined chosen limbic ROIs
(M[Fed] = -5.034, 95% HPD [-7.286, -2.839]; see Figure 11). Follow-up t-tests indicate
increased fasting differential brain response to reward cues > baseline across the limbic
ROIs, with only the bilateral Caudate providing evidence of a trend (see Table 8).
Limbic ROIs Reward Cues > Baseline Follow-up t-tests
ROI t-stat (Fast>Fed) p-value
Left Insula 1.6228 0.122
Left Accumbens 1.286 0.15989
Left Caudate 2.021 0.0584
Left Putamen 1.2204 0.2381
Thalamus 0.911 0.375
midbrain -0.3045 0.7643
Right Insula 1.345 0.195
Right Accumbens 1.598 0.127
Right Caudate 2.4597 0.0243
Right Putamen 1.234 0.233
Right Thalamus 1.1195 0.2776
Right SMA 0.5244 0.6064
Table 8. Outputs of the paired samples t-tests across the limbic ROIs comparing fast – fed days.
Hierarchical Bayesian modeling for the main effect of day on the Reward Cues – Baseline contrast (M[Fed]
= -5.034, 95% HPD [-7.286, -2.839]) with the outputs of the follow-up t-tests shown above indicating that
this effect may be driven by the bilateral caudate.

44
Figure 11. Limbic ROI fast – fed difference of Reward cues greater than baseline. Significant activation for
this contrast appeared across bilateral caudate. The presence of significant activation was determined
through the fitting of a hierarchical Bayesian model across all ROIs (M[Fed] = -5.034, 95% HPD [-7.286, -
2.839]).

Today Cues > 1 Month Cues
There is no evidence of a significant interaction between reward immediacy and
hunger status across the limbic ROIs (M[Fed] = -2.282, 95% HPD = [-5.193, 0.57]).
$25 Cues > $5 Cues
There is no evidence of significant interaction between reward magnitude and
hunger condition across the limbic ROIs (M[Fed] = 0.547, 95% HPD = [-2.611, 3.528]).
Discussion
Use of the Monetary Incentive Delay task has historically been used in
experimental research for value tracking in the nucleus accumbens (Knutson et al., 2001)
and in the ventral medial PreFrontal Cortex (Knutson et al., 2003). However, recent work
has implied that in addition to MID task being used for tracking the magnitude of

45
rewards, the MID task can be extended to investigate behavioral and neurobiological
makers of delay discounting. Specifically, when playing for either an immediate reward
or a preference matched reward that was delayed by 4 months, immediate rewards were
associated with increased responding in the bilateral putamen, bilateral anterior insula
and midbrain, with response times appearing to be significantly faster during immediate
rewards when compared to preference matched delayed rewards (S. Luo et al., 2009).
Nicotine dependence has a strong relationship in the experimental literature with
delay discounting (Bickel, Odum, & Madden, 1999). Nicotine’s primary neurochemical
actions are to decrease the activity of GABA (the brain’s most prevalent inhibitory
neurotransmitter) and to increase the activity of glutamate (an excitatory
neurotransmitter). Likely through it’s action on GABA or glutamate, nicotine has been
seen in animal studies to decrease reward stimulation thresholds (Huston-Lyons &
Kornetsky, 1992). Due to the decrease in reward thresholds and the increase in delay
discounting, Shan Luo, Ainslie, Giragosian, & Monterosso, 2010 completed the
intertemporal variant of the MID task during fMRI scanning to investigate whether
reward sensitivities to magnitude, delays, or both were responsible for the increased
discounting seen among smokers. Results indicated activation of magnitude tracking
regions like the dorsal and ventral striatum to immediate rewards with little to no
reactivity to delayed rewards among smokers.
The current work sought to investigate how the experimental manipulation of
hunger status (by either having participants remain fasted, or by feeding them until they
felt ‘comfortably full’) affects the behavioral and neural response to rewards of different
delays and magnitudes. Results of my manipulation check indicated that unsurprisingly

46
the primary experimental manipulation of a 12-hour fast followed by either feeding or
continued fasting (in a randomized order across subjects) led to dramatic differences in
both self-reported hunger and fullness. In this intertemporal variant of the MID task,
hunger was associated with significantly faster response times across all rewarded cues
when compared to the non-reward cues, whereas when ‘comfortably full’ there did not
appear to be RT differences by cue condition. This finding may represent either a
generalized decrease in reward reactivity due to feeding’s restoration of homeostasis or
hunger’s disruption of homeostasis increasing reward sensitivity and reactivity.
Main effects of experimental contrasts averaging across the two days did not
reach significance in the same contrasts as the original intertemporal MID study, which
saw widespread activation in value-tracking regions in response to different reward
magnitudes (S. Luo et al., 2009), though this may be due to sample size differences. Only
the visual cortex met whole-brain thresholding for the magnitude contrast ($25 > $5), this
is likely due to visual and luminance differences between the different reward magnitudes
as opposed to this being due to actual magnitude tracking in the visual cortex. Delay
tracking for the contrast of Today > 1 Month averaged across the fasted and fed day
showed widespread differences in functional neurobiology. In line with past works,
immediate rewards led to increased activation in limbic structures including the caudate,
putamen, anterior insula, and the thalamus when averaged across the fasted and fed days.
Finally, rewarded cues led to only significant increases in visual cortex activation when
compared to the baseline condition, no value tracking regions met whole brain
thresholding as would be expected by a comparison of a rewarded and non-rewarded
trial.

47
Hunger status did not meet whole brain thresholding for any of the contrasts of
interest. As was mentioned, there were, however, dramatic differences between
experiment days when the non-rewarded cue was contrasted with implicit baseline. Due
to this complication, neural changes in response to rewarded cues was completed by
comparing all rewarded cues to implicit baseline as was done in the whole-brain analysis
above. Looking across the limbic ROIs used for this task, there appeared to be an effect
of hunger status on neural response to rewarded cues. Follow-up tests indicated that this
effect was primarily driven by the bilateral caudate. This increase in dorsal caudate
response when hungry matches results from the only other study to investigate the
neurobiology of hunger’s effect on delay discounting processes (Wierenga et al., 2015).
This experimental procedure expands on prior work in several ways, however it
does have some limitations that are important to address. This experimental design
allowed for a direct comparison of the behavioral and neurobiological reward signals
associated with immediate and delayed rewards which is a novel extension from past
works. However, the control condition in which there was no reward associated with
performance on that trial showed dramatic differences across the two days. This led to the
necessity of comparing the rewarding food-cues to the implicit baseline of the model
instead of the non-reward condition. In addition, the use of a within subject design
enabled the best use of available resources, however, the sample size for this experiment
was still only 19. Future works should seek to increase the number of subjects to reduce
across subject variance in order to better estimate changes due to homeostatic condition.

48
Chapter 3: Adaptive Intertemporal Choice Task
Introduction
Relative to the rest of the animal kingdom, humans have an exceptional ability to
forego immediate rewards in favor of larger delayed rewards. Systematic choice of
intertemporal choice questions and the fitting choice behavior to a chosen computational
model enables the estimation of discounted utility functions, which can be used to infer
an individual’s delay discounting. Historically, experimental research on delay
discounting has focused on trait level variance across individuals due to this aspect of
estimated delay discounting’s strong test-retest correlation of 0.71 over the course of a
year (K. N. Kirby, 2009). More recently, however, a large number of studies have taken
an interest specifically in working out the mechanisms of systematic stochasticity around
an individual’s estimated trait level discounting. In other words, if an individual’s
baseline level of discounting is X, will an experimental manipulation such as emotion
induction lead to a systematic increase or decrease in discounting around X within a
randomly selected subject (for review of state-like effects on delay discounting see
Lempert, Glimcher, & Phelps, 2015).
Experimental research aimed at investigating the “malleability” of intertemporal
choice preferences must start by picking a model for the shape of the discounting
function. Early research on delay discounting utilized a simple exponential form to
estimate discounting, however, the inability of exponential functions to explain shifting
preferences as time moves forward led to the emergence of the hyperbolic discounting
function as the most common model across the experimental literature (for review of
potential models see Doyle, 2013). In addition to the hyperbolic model better fitting

49
experimental data, hyperbolic estimation of delay discounting is done via a single free
parameter, k, as shown in Formula 1 below:
𝑉 =
𝐴
1+𝑘𝐷

Formula 1.
Using this hyperbolic form, the current value, V, of a delayed option is equal to the dollar
amount, A, divided by one plus the free parameter, k, representing the individual’s rate of
discounting multiplied by the number of days, D, until the delayed reward (K. Kirby,
Petry, & Bickel, 1999). Taking advantage of the flexibility of such a simple discounting
function, recent studies have utilized an adaptive procedure in which estimated k-value
has the ability to track changes in preferences trial by trial (Shan Luo et al., 2014). By
utilizing this adaptive estimation procedure, presentation of intertemporal choice sets can
be kept as close to the indifference point while allowing state-based alterations in
discounting which enables efficient estimation of systematic state-based stochasticity of
choice and associated brain networks associated with “hard” intertemporal choices.
The choice of a discount function to estimate delay discounting enables the study
of the neurobiology of intertemporal choice preferences. Across the neuroimaging
literature researchers either choose to give a wide array of interest rates to ensure they hit
the individual’s discounting in the presented intertemporal choices (e.g. McClure,
Laibson, Loewenstein, & Cohen, 2004) or they use a more sophisticated adaptive
algorithm and experimentally control easy and hard choice sets (e.g. Shan Luo et al.,
2014). Early neuroimaging work hypothesized the existence of two separate but
competing functional networks in the brain, namely the beta and delta systems (Samuel
M. McClure et al., 2004; S.M. McClure, Ericson, Laibson, Loewenstein, & Cohen,

50
2007). The beta network is comprised of limbic and paralimbic structures rich in
dopaminergic innervation (e.g. ventral striatum, medial OFC and medial PFC) and is
most responsive to the smaller sooner option when it is available today. The delta system
on the other hand is responsive to all choice sets independent of the duration to the first
reward and includes the visual cortex, premotor area, the left and right intraparietal
cortex, right dorsolateral PFC, right ventrolateral PFC and lateral OFC. More recently,
some researchers in the field of Decision Neuroscience have suggested that difficult
decisions may not involve two antagonistic networks. Instead, single system theories
posit that limbic reward networks and frontoparietal control networks work as one
dynamic valuation system that enables complex choices (Hare, Hakimi, & Rangel, 2014;
Monterosso & Luo, 2010), however this remains a hot topic in the theoretical literature
with no clear biological model proving to be superior.
Even when viewed as a single functional network spanning across cognitive
control and limbic regions, there does appear to be regional specialties among the
dynamic network. Specifically, delay to reward is tracked in the dlPFC, PPC and mPFC
(Ballard & Knutson, 2009; Kable & Glimcher, 2007), whereas the value of the delayed
option is tracked in the anterior dmPFC, PCC, ventral striatum, specifically the nucleus
accumbens (Ballard & Knutson, 2009; Kable & Glimcher, 2007; Q. Wang et al., 2014).
The value of the smaller sooner option is tracked in the posterior dmPFC (in contrast to
the anterior dmPFC for the delayed option) whereas the relative value of the options is
tracked in the vmPFC, nucleus accumbens and PCC (Q. Wang et al., 2014). During
“hard” intertemporal choices, choice of the smaller sooner option is associated with
decreased activity of the left dlPFC (Figner et al., 2010), less frontoparietal activation

51
along with increased connectivity of the anterior insula with the frontoparietal network
(Clewett et al., 2014). In contrast, larger later choice is associated with increased activity
of the vmPFC, ACC, frontal pole, left dlPFC and left insula (Figner et al., 2010; Shan
Luo, Ainslie, Giragosian, & Monterosso, 2011), and with increased connectivity between
the dlPFC and the vmPFC (Hare, Camerer, & Rangel, 2009; Hare et al., 2014).
A recent meta-analysis utilized an ALE procedure across the functional peaks for
13 different delay discounting reports in an attempt to objectively summarize the
neuroimaging literature (Carter, Meyer, & Huettel, 2010). This meta-analysis found 25
total clusters that spanned the functional neurobiology of delay discounting. These 25
clusters are broken up in to 3 categories in this report. Core ROIs that support prospective
processes including future planning, autobiographical memory and theory of mind.
Valuation ROIs on the other hand span a network of regions that are thought to be
sensitive to value. Finally, some of the ROIs were categorized as not involving future
orientation or valuation, so were considered to support what are potentially non-
interesting aspects of the decision-making process not specific to intertemporal choice.
As discussed in the introduction, the effects of homeostatic imbalance on
intertemporal choice preferences, specifically hunger, has been looked at almost
exclusively from a behavioral and endocrine point of view. Behaviorally there is
evidence that the visceral impulsivity that accompanies hunger has a “Spillover” effect on
intertemporal choice preferences and leads to steeper discounting (Ashton, 2015; X. T.
Wang & Dvorak, 2010) though this may be a subtle effect due to one experimental test
not finding delay discounting effects (Ridder et al., 2014). In addition, hunger has
previously been associated with increased convexity of the utility function and increased

52
immediacy bias (Ashton, 2015; Kuhn et al., 2014). The immediacy bias “Spillover” also
extends to decreased reaction time on the hunger day, and increased response in reward
networks when the smaller sooner option is today in the brain among healthy women
(Wierenga et al., 2015).
This experiment was designed to expand on the one study to date that has looked
at the neural mechanisms of homeostatic imbalance on intertemporal choice preferences.
In Wierenga et al., 2015, a wide array of intertemporal choice questions were used with
no regard for individual differences in what constitutes hard or easy decisions. By
utilizing an adaptive procedure to ensure hard and easy choices are under experimental
control, and having fixed delays of today and in a month reaction time effects and the
neural mechanisms of magnitude and hard choices can be expanded upon.
Methods
Subjects
As was described in Chapter 2, 19 individuals completed the pre-screening
session and both of the fMRI sessions for the study. The use of the adaptive intertemporal
choice task was interleaved with the intertemporal MID task in a pseudo-randomized
order across subjects.
ITC Task
Participants completed two runs of an 80-question adaptive intertemporal choice
task paired with fMRI. On each trial, participants chose between an immediate amount of
money (“Today”) and a delayed amount of money (delay always of 30 days, and amount
of the delayed amounts of $20-65, see Figure 12.). Participants were instructed that all
choices would be eligible to be randomly selected for real payout at the end of each study

53
session, which was made by Amazon Allowance (activated either immediately or at the
specified delay). Immediate amounts offered were generated and adapted so as to narrow
in on a set of difficult intertemporal choices for each participant (i.e., sooner smaller and
later larger alternatives that are similarly attractive). We used adapting “ladder
parameters” to accomplish this. In actuality, when the participant chose the immediate
alternative, the immediate alternative was made slightly less attractive on the next trial.
Conversely when the participant chose the delayed alternative, the immediate option was
adjusted to be more attractive on the next trial. The interleaving of multiple ladder
parameters obscured the connection between decisions and subsequent alternatives. As a
control condition for fMRI contrasts, 20 questions were designed to be “easy” by having
either both options being available today with differing amounts, or having both options
be equal in amount but having a different delay to reward receipt counterbalanced across
subjects.
Figure 12. Visual representation of the adaptive intertemporal choice task used during fMRI scanning.
Similar to the intertemporal MID task, each trial started with a random jittered ITI pulled an exponential
Jittered ITI (exponential mean=2s)
Post-choice (0.5s)
Choice Period (1s-RT, max of 5s)

54
distribution with a mean of 2. The task was given as a starting estimate either the behavioral day k-value or
the k-value estimated from the last trial of the last run if the second run of the task was being started. Based
on the estimated level of discounting for the person, a randomized larger later value was generated which
was then matched with a smaller sooner option based on estimated delay discounting. Subjects were given
a maximum of 5 seconds to respond. Following choice, participants were shown which option they had
chosen for half a second prior to the start of the next trial. Participants completed 2 runs of 80 trials on each
experimental day of the study, consisting 60 ‘hard’ and 20 ‘easy’ intertemporal choice sets.

MRI Data Acquisition
Neuroimaging data was collected using a 3T Siemens MAGNETOM Tim/Trio
scanner at the Dana and David Dornslife Cognitive Neuroscience Imaging Center at the
University of Southern California with a 32-channel head-coil. Participants laid supine on
a scanner bed, viewing stimuli through a mirror that is mounted on the head coil. Blood
oxygen level-dependent (BOLD) response was measured via an echo planar imaging
(EPI) sequence with PACE (prospective acquisition correction) turned on (TR=2s,
TE=25ms, flip angle=90, resolution=3mm
3
isotropic, 64 x 64 matrix in FOV=192mm). A
total of 41 axial slices, each 3mm in thickness were acquired in an ascending interleaved
fashion to cover the whole brain. The slices were tilted to align the axial slices with each
individual’s AC-PC plane to minimize signal dropout in the orbitofrontal cortex
(Deichmann et al. 2003). Anatomical images were collected using a T1-weighted three-
dimensional magnetization prepared rapid gradient echo (MPRAGE with TI=900ms,
TR=1.95s, TE=2260ms, flip angle=9 degrees, resolution=1mm, 256 x 256 matrix in
FOV=256mm) for alignment of each individual’s structural space to standardized space.
These scans were co-registered to the individual’s mean EPI images.
Analysis 1: Effects of Hunger on Response Times
Easy vs. Hard Choice Reaction Time Analysis

55
In order to validate the control condition as being “easier”, hierarchical Bayesian
estimation of RT differences across choice types was fit again utilizing a regression
framework fitting using an MCMC procedure through the BAMBI package in python.
Hard Choice Reaction Time Analysis
Hierarchical Bayesian estimation of RT differences across the two experimental
sessions was fit again utilizing a regression framework utilizing only the “hard” choices.
In addition, the magnitude of the larger later amount and the log of k of the offered
choice set was included as covariates to ensure RT effects across days were not due to
latent variables of the offered choice sets.
Analysis 2: Hierarchical Drift Diffusion Model
Introduction to the HDDM
As we move about world, humans are constantly faced with a wide variety of
decisions. Most of these real-world decisions must be completed in some reasonable
amount of time, depending on the specifics of the choice. Considering how nearly all
choices require a decision to be made in some timely manner, choices require both the
decision to stop deliberating as well as the decision or act itself. The Drift Diffusion
Model is a computational model used in the cognitive sciences and behavioral economics
as a descriptive model for how individuals make decisions in a noise-filled world. This
model, which is among the class of sequential sampling models (that also includes the
Ballistic Accumulator Model, among others), assumes that when a choice needs to be
made, the decision maker accumulates noisy information from the environment until a
threshold of evidence is reached, triggering the decision to stop deliberating and make a
choice or complete the act (Forstmann, Ratcliff, & Wagenmakers, 2016).

56
Due to the environment being noisy and decisions generally needing to be made
in a timely manner, choices are often plagued by the speed-accuracy trade-off. More
specifically, this trade-off is between how continuation of sampling from the environment
will enable a better decision, yet speedy decision-making is also a necessity. Often in
computational models of choice, this tradeoff is side-stepped with choices being analyzed
separately from response times, and with errors being excluded. The Drift Diffusion
Model offers an explanation of choices, reaction time and errors. Specifically, errors can
be explained in the DDM due to noise in the value accumulation process occasionally
leading to errors through an accumulation of evidence to the “wrong” threshold, it is
posited by the DDM that this same noisy accumulation of value is mechanism underlying
within subject variance in response times.
Across the literature there are a number of variations of the DDM, and several
open-source analysis packages for implementing these models. Some models enable
added free parameters, such as shrinking decision-thresholds as response time increases
(see Voskuilen, Ratcliff, & Smith, 2016). However, the best-established formulation of
the DDM involves the estimation of 5 parameters. First, is the bias parameter or g
parameter which enables the model to estimate a decision bias among the data. In other
words, if across all choices one option is chosen more often than the other, than the
model will best fit if the starting point for the sequential sampling is closer to the option
that is picked most frequently. Second, is the non-decision time parameter or t parameter
which is an estimation of the amount of time at the start of the sequential sampling
process prior to the actual sampling of the environment reflecting encoding, motor
responses and other unknown factors that may affect the duration until sampling can

57
begin. Third, is the drift-rate parameter or v parameter which is an estimation of the rate
(or slope of the value-accumulation line if noise were removed) at which value can
accumulate in the model in the direction of which available option is currently being
sampled. Fourth, is the decision-threshold parameter or a parameter which determines
how far from the midpoint the boundaries of the value accumulation function are, and
when the value accumulation process hits one of these boundaries (regardless of whether
this is due to value accumulation or noise), passing this threshold triggers the decision to
stop sampling and make the choice. Finally, a noise parameter is estimated along with all
the other parameters in a manner that enables the best fit of the computational model to
the data based on “errors” and variance in within-subject response time distributions.
In the computational modeling literature, there is some evidence of the efficacy of
sequential sampling class models in explaining intertemporal choice behavior, though to
date this has only been tested through ballistic accumulator models (Rodriguez, Turner,
Van Zandt, & McClure, 2015). There is evidence, however of increased efficacy and
power of the hierarchical class of the drift diffusion model so long as there is variability
in RT where individuals make choices between two options (Johnson, Hopwood, Cesario,
& Pleskac, 2017). Fitting of experimental datasets to the DDM enables the unpacking of
the cognitive processes underlying within or between subject differences in response
times. Past work has provided evidence of a slowing of response times following feeding
and through the fitting of the HDDM during intertemporal choices provides an
opportunity to unpack potential cognitive processes underlying the hunger-based
difference in RT (Wierenga et al., 2015).
Intertemporal Choice Data

58
The behavioral data used for the fitting of the Hierarchical Drift Diffusion Model
was the same data described earlier in this chapter in which participants completed 2 runs
of 80 intertemporal choice sets on both a day in which they have fasted for at least 12
hours, and another day in which they have been recently fed. In the first fitting of the
HDDM, the 60 ‘hard’ and 20 ‘easy’ trials per run across both days were all used in an
attempt to verify that differences between easy vs. hard intertemporal choice response
times matches the past findings of this type of contrast in sequential sampling models in
an attempt to verify the use of the HDDM in intertemporal choice sets. Next, only the
‘hard’ trials from each run were used to investigate the cognitive processes underlying
the hunger-driven homeostatic state RT changes.
Model Estimation
Both sets of DDM models were fit using the HDDM toolbox in python (Wiecki,
Sofer, & Frank, 2013). This package enables hierarchical Bayesian estimation of all
model parameters, and provides an easy platform for model fitting and testing. The
standard DDM (in which maximum likelihood estimation is most often used instead of
Hierarchical Bayes) estimates parameters one subject at a time, and in so doing typically
requires hundreds of trials for accurate estimation to generate an RT distribution for each
subject. The HDDM python package instead takes a Hierarchical Bayesian approach
producing estimates at both the individual and condition levels while also exploiting
shrinkage of model estimates with low variability across subjects moving individual
estimates towards group means. The model is structured hierarchically because the data
for each subject are constrained by a higher order condition-level distribution (Wiecki et
al., 2013) which results in an accurate ability to estimate condition-level parameter

59
changes with far fewer trials than the older DDM models that require within subject RT
distributions. Posterior distributions produced by the HDDM package are estimated using
Markov Chain Monte Carlo (MCMC) methods by estimating distributions by repeatedly
drawing samples from it. For the HDDM, estimation of condition-level parameters are
simply given uninformative prior distributions which are then updated by sampling the
data resulting in posterior distributions that can be used to test inferences.
Investigation of the effects of choice difficulty on intertemporal choice RT, and of
homeostatic condition on intertemporal choice RT involved generating 9 different
potential models. Starting with a null model in which the condition of interest (i.e. choice
difficulty or homeostatic condition) was not included in the model, to the most detailed
model that had decision bias turned on, and condition-based alterations in non-decision
time, threshold, and drift rate. All possible iterations of condition-based changes in these
parameters were tested in searching for the best-fitting model.
Model Comparisons
For both sets of 9 HDDM models that were fit for the two comparisons of interest,
the best fitting models were initially chosen based on each individual model’s DIC score.
Model selection through the minimization of the DIC score was chosen due to the DIC
score being based on the effective number of parameters in the model since increased
model complexity should by chance alone increase explained variance in the model. By
penalizing more complicated models for having more DOF this method enables the
comparison of models of varying complexity. Past research has found that the DIC can in
some situations slightly favor more complex models (Wiecki et al., 2013), so to verify

60
model selection the 3 best-scoring DIC models for the contrast of greatest interest were
further examined by comparing simulated to experimental data.
Simulation Tests
To further validate model selection procedures, the 3 lowest scoring DIC scores
across the 9 tested models were submitted to simulation tests. More specifically, each of
the 3 best-fitting models were used to generate 500 simulated datasets that were then
compared to the experimental data. To assess how well simulated datasets matched
empirical data, the Mean, Mean Squared Error (MSE), and Mahalanobis Distance was
calculated to determine how well the simulated response distributions matched the
empirical data. Unlike the use of model-performance metrics like the MSE when
attempting to best explain empirical data (searching for the model with the absolute
lowest MSE for example), model performance in this case is considered best when the
simulated data has an MSE, Mahalanobis Distance, and Mean as close to the empirical
data as possible.
Parameter Inference
Due to the use of Hierarchical Bayesian estimation of posterior distributions for
condition effects of interest, inferences about differences between parameters can be done
by directly comparing posterior distributions. Based on the results of the model
comparisons via DIC and simulation tests, the best fitting model can be used to compare
which model parameters are affected by homeostatic state by calculating the probability
of one posterior distribution being larger for one condition than for the other for the
conditions of interest (in this case easy vs. hard questions and fasted vs. fed).
Analysis 3: Effects of Hunger on Intertemporal Choices

61
Fast Vs. Fed Choice Effects
To examine the effect of the experimental hunger manipulations on the tendency
to choose immediate or delayed options, hierarchical mixed-effects logistic regressions
were used via the glmer function (a part of the lme4 package in R), and the anova
function (built-in R function). Due to the adaptive algorithm used in the intertemporal
choice task completed, simple SS/LL ratio questions are not relevant as have been used in
prior research (Shan Luo et al., 2014). Instead, the question asked is whether on a person-
by-person does the logistic curve based on SS/LL choice systematically shift as a
function of hunger condition when plotted against the log of the offered k-value.
First, the best mixed-effects model structure for the measured variables of no-
interest was determined, as is recommended in Gelman & Hill, 2007. To determine the
best fitting baseline model, variables of no-interest were added as both fixed effects, and
as random slopes one at a time, and model selection was evaluated via the anova function
which evaluates model fit by utilizing a Chi-squared value that is the coefficient of the
likelihood ratio test (twice the difference in log likelihood across the two models). The
best-fitting base model looked as follows:
𝐶ℎ𝑜𝑖𝑐𝑒 ~ 𝑙𝑜𝑔𝐾+𝑆𝑆 𝑎𝑚𝑜𝑢𝑛𝑡+𝐷𝑎𝑦+ (1+𝑙𝑜𝑔𝐾+𝑅𝑢𝑛+𝑆𝑆 𝑎𝑚𝑜𝑢𝑛𝑡+𝑇𝑟𝑖𝑎𝑙+𝐷𝑎𝑦|𝑆𝑢𝑏𝑗𝑒𝑐𝑡)
Formula 2.
In other words, fixed effects of the log of K, a z-scored version of SS amount offered and
experiment day. A random intercept of Subject was included, along with random slopes
of the log of K, run of the task, z-scored SS amount offered, z-scored Trial number, and
Day (hunger condition).
To examine the primary hypothesis of a hunger-based shift in choice behavior, the
above Formula 2 is taken as a baseline model and is compared to a similar model

62
(Formula 3) that has the interaction between the log of K and the experimental day via
the anova function as in R.
𝐶ℎ𝑜𝑖𝑐𝑒 ~ 𝑙𝑜𝑔𝐾+𝑆𝑆 𝑎𝑚𝑜𝑢𝑛𝑡+𝐷𝑎𝑦+𝑙𝑜𝑔𝐾:𝐷𝑎𝑦+ (1+𝑙𝑜𝑔𝐾+𝑅𝑢𝑛+𝑆𝑆 𝑎𝑚𝑜𝑢𝑛𝑡+𝑇𝑟𝑖𝑎𝑙+𝐷𝑎𝑦|𝑆𝑢𝑏𝑗𝑒𝑐𝑡)
Formula 3.
Analysis 4: fMRI Analysis
Preprocessing
fMRI data preprocessing was conducting utilizing the python-based neuroimaging
data processing package nipype (Gorgolewski et al., 2011) to call preprocessing functions
from FSL, a part of the Oxford University Centre for functional MRI of the Brain
(FMRIB) Software Library (www.fmrib.ox.ac.uk/fsl). Each individual’s structural and
functional data was transformed to MNI space by first linearly transforming structural
image to MNI space, and linearly transforming EPI to structural space via FSL’s FLIRT
function (Jenkinson et al., 2002; Jenkinson & Smith, 2001). FSL’s FNIRT was then used
to non-linearly transform the structural data to standard MNI space. The functional
images were motion and slice-time corrected and were spatially smoothed using a
Gaussian Kernel with a full width at half-maximum of 5mm. We also applied a high-
pass temporal filtering using a filter width of 120s.
Whole Brain Analysis
As has been recently pointed out, cluster correction for whole-brain voxelwise
analysis has received some criticism for potentially having inflated type-1 errors when
utilizing any of the major fMRI analysis softwares (Eklund et al., 2016). In light of these
potential cluster correction issues, whole brain analysis were carried out using
nonparametric permutation inference via FSL’s randomise algorithm (Winkler et al.,

63
2014). These methods are used for inference on static maps when the null distribution is
not known. In order to reduce the number of comparisons made, a gray matter mask was
applied prior to the running of randomise since signal changes in white matter are likely
to be Type 1 error. The raw statistic contrast maps were then converted to cluster-like
activity in contiguous voxels via Threshold-Free Cluster Enhancement. Corrected p-value
maps were then thresholded at 0.05.
ROI Analysis
Regions of interest were chosen based on a recent meta-analysis done by Carter et
al., 2010. Regions in this meta-analysis were categorized by the authors in to 3 categories
as was described above: valuation (supporting valuation processes of immediate and
delayed options), core (supporting prospective processes like autobiographical memory,
theory of mind and planning for the future) or neither. For dimensionality reduction for
the ROI analysis, only the valuation and the core ROIs were analyzed. In addition, data
was first fit to a hierarchical mixed effects model by fitting regression parameters with
the BAMBI module in python via MCMC estimation through PYMC3 looking
specifically for the effects of hunger across each category of ROI. Significant main
effects of day were then followed up upon via paired samples t-test to determine which
ROIs among each class are driving any main effects of hunger condition on ROI
response.
Results
Analysis 1: ITC Reaction Time Results
Easy vs. Hard Choice Reaction Time Results

64
In order to validate the control condition as being “easier”, hierarchical Bayesian
estimation of RT differences across choice types confirms that “hard” choices are indeed
associated with longer response times across days (M [Hard] =0.387, 95% HPD [0.352,
0.423], for Chains and Posterior Distributions see Appendix 2.2.1).
Fast vs. Fed All Choices Reaction Time Results
Replicating prior findings showing that there is slowing of RT after feeding on
both easy and hard trials there was a significant increase in RT on the Fed day (M [Fed] =
0.091, 95% HPD [0.059, 0.122]; for MCMC Chains see Appendix 2.2.1) (Wierenga et
al., 2015).
Fast vs. Fed Hard Choice Reaction Time Results
Hierarchical Bayesian estimation of RT differences across the two experimental
sessions indicates that there is a significant increase in RT on the Fed day (M [Fed] =
0.1371, 95% HPD [0.098, 0.176]; see Table 9 and Figures 13 and 14; for MCMC Chains
see Appendix 2.2.2) on only ‘hard’ trials extending past findings (Wierenga et al., 2015).
Hierarchical Bayesian ITC RT Chains and Estimates
Mean SD 0.95 HPD
lower
0.95 HPD
upper
Effective n Gelman
Rubin
1|Subject 0.39 0.07 0.27 0.54 2923 1.00
Day (Fed) 0.14 0.02 0.10 0.18 20062 0.99
Intercept
(Fast)
2.52 0.12 2.28 2.75 3411 1.00
LL
Amount
-0.002 0.0008 -0.003 0.00001 10181 1.00
RT SD 0.64 0.007 0.63 0.66 24000 0.99
Log of K 0.07 0.01 0.04 0.10 4676 1.00
Table 9. Posterior estimates for the hierarchical Bayesian model estimating the main effect of experiment
day on response times. Results indicate that there is a significant main effect of day on RT, with the fed day
being approximately 0.14 seconds (95% HPD [0.10, 0.18]) slower than the mean of 2.52 seconds on the
fasted day (95% HPD [2.28, 2.75]).

65
Figure 13. Visualization of the posterior estimates of response times on the adaptive intertemporal choice
task when split by experiment day (i.e. fast and fed). Feeding was associated with significant slowing of
response times of an average of 0.14 seconds (95% HPD [0.10, 0.18]).

66
Figure 14. Visualization of the RT distributions across all subjects on the fasted and fed days. From this
histogram is appears that feeding is indeed associated with increased response times as is confirmed by the
hierarchical Bayesian model that estimated an average increase of 0.14 seconds above.

Analysis 2: HDDM Results
Easy vs. Hard Decisions
DIC Comparisons
Comparison of the DIC score of the 9 possible models investigating which of the
HDDM free-parameters are influenced by choice difficulty led to the selection of the
model which allowed decision bias across conditions (g), and allowed difficulty-based
modulation of the threshold (a), drift rate (v), and non-decision time (t) parameters (See
Table 10).

67
Choice Difficulty Changes in Drift Diffusion
Model Parameters
Model
Name
DIC Score Decision
Bias on (g)
Decision
Threshold
(a)
Non-decision
Time (t)
Drift
Rate (v)
Null 18652.37
G 18563.11 X
GA 18263.01 X X
GV 16452.77 X X
GT 17961.20 X X
GAV 16417.53 X X X
GAT 17713.57 X X X
GVT 15983.17 X X X
GAVT 15827.25 X X X X
Table 10. Description of the 9 different models fit for the hierarchical drift diffusion models fit in an
attempt to investigate what parameters are altered by choice difficulty. Models were compared based on
DIC score, which penalizes more complicated models enabling the comparison of models of varying
complexity. The best-fitting model was the GAVT model, which allowed increased probability of choosing
the smaller sooner or larger later within subject across days (via the g parameter) along with experimental
day based variation in decision threshold (a parameter), non-decision time (t parameter), and drift rate (v).

Parameter Inference
Parameter inference using the posterior distributions of the GAVT model (chosen
based on DIC score), that enables decision thresholds, drift rate and non-decision time to
vary as a function of question difficulty implies that the decision threshold and non-
decision time both differ significantly based on trial difficulty (P(a(Easy) < a(Hard))= 1;
P(t(Easy) < t(Hard)) = 0.021; see Figures 15 and 16 respectively), whereas drift-rate
showed much less evidence of difference in Easy vs. Hard posterior distributions
(P(v(Easy) < v(Hard)) = 0.282; see Figure 17).

68

Figure 15. Posterior distributions for the effect of choice difficult (easy vs. hard) on decision threshold (a
parameter). Comparison of these distributions directly indicate a significant effect of choice difficulty on
posterior estimates for the best fitting model (P(a(Easy) < a(Hard)) = 1).

Figure 16. Posterior distributions for the effect of choice difficult (easy vs. hard) on non-decision time (t
parameter). Comparison of these distributions directly indicate a significant effect of choice difficulty on
posterior estimates for the best fitting model (P(t(Easy) < t(Hard)) = 0.021).

69

Figure 17. Posterior distributions for the effect of choice difficult (easy vs. hard) on drift rate (v parameter).
Comparison of these distributions directly indicate there was not a significant effect of choice difficulty on
posterior estimates for the best fitting model (P(v(Easy) < v(Hard)) = 0.282).

Hunger Condition
DIC Comparisons
Comparison of the DIC score of the 9 possible models investigating which of the
HDDM free-parameters are influenced by homeostatic condition led to the selection of
the model which allowed decision bias across conditions (g), and allowed difficulty-
based modulation of the threshold (a) and non-decision time (t) parameters (See Table
11). However, the DIC score for the GT and GAVT model were reasonably close to the
GAT model, so these three models were chosen to be passed to the simulation tests
completed next.
Feeding-Based Changes in Drift Diffusion Model
Parameters
Model
Name
DIC Score Decision
Bias on (g)
Decision
Threshold
(a)
Non-decision
Time (t)
Drift
Rate (v)
Null 14092.83
G 13995.35 X
GA 13887.34 X X

70
GV 14010.42 X X
GT 13692.99 X X
GAV 13904.26 X X X
GAT 13650.18 X X X
GVT 13705.58 X X X
GAVT 13666.40 X X X X
Table 11. Description of the 9 different models fit for the hierarchical drift diffusion models fit in an
attempt to investigate what parameters are altered by the experimental manipulation of hunger. Models
were initially compared based on DIC score, which penalizes more complicated models enabling the
comparison of models of varying complexity. This comparison was used to identify the three best fitting
models which were then submitted to simulation analysis to better ensure the best model was chosen. The
best-fitting model based on DIC alone was the GAT model (DIC=13650.18), with the GAVT
(DIC=13666.4) and GT DIC=13692.99) models performing second and third respectively.

Simulation Tests
Datasets were simulated with the 3 best scoring models based on DIC, which
included the GT model (allowing non-decision time to vary based on feeding condition),
the GAT (allowing decision threshold and non-decision time to vary based on feeding
condition) model, and the GAVT model (allowing decision threshold, drift rate, and non-
decision time to vary based on feeding condition). This enables the estimation of
accuracy-related parameters comparing each of the 3 best-scoring models with the actual
collected data. Results indicate that all 3 of the best scoring models have only negligible
differences in prediction accuracy metrics including mean accuracy, mean squared error
(MSE) and mahalanobis distance (see Table 12 below). As would be expected, the model
with the largest number of parameters (the GAVT model) has the best prediction
accuracy by a razor thin margin, but since these error-based statistics do not penalize
model complexity as does DIC, it appears the most likely model remains the GAT model
(See Table 12; For simulated distributions see Appendix 2.3.1).
Model Comparisons HDDM Simulated Datasets
Model Descriptive
Statistic
Accuracy Mean UB Mean LB

71
Observed
NA 0.434 2.11 -2.18
GT
Mean 0.427 2.12 -2.25
MSE 0.007 0.133 0.14
Mahalanobis 0.082 0.026 0.183
GAT
Mean 0.427 2.12 -2.25
MSE 0.008 0.137 0.15
Mahalanobis 0.075 0.024 0.173
GAVT
Mean 0.427 2.12 -2.25
MSE 0.007 0.14 0.15
Mahalanobis 0.08 0.034 0.186
Table 12. Model comparison statistics comparing the empirical data with the three best performing models
when comparing DIC score. Results indicate that all 3 models perform very similarly despite differing
numbers of free parameters. Of note, the best performing model based on DIC comparison ( the GAT
model) also was associated with the lowest Mahalanobis distance (0.075) when comparing distance of
centroid from the empirical data. These simulations do not eliminate any simulated model due to poor
performance, and agree with the DIC finding that the GAT model is the best fitting based on the collected
empirical data.

Parameter Inference
Parameter inference using the posterior distributions of the GAT model (chosen
based on DIC score and Simulation Tests Above), that enables decision thresholds and
non-decision time to vary as a function of homeostatic state implies that the decision
threshold and non-decision time both play a minor role in hunger-based differences in RT
distributions (P(a(Fast) < a(Fed) = 0.716; P(t(Fast) < t(Fed) = 0.901; see Figures 18 and
19 respectively), however, neither comparison of posterior distributions met the pre-
defined cut-off of a 95% difference. Thus while the null hypothesis that neither of these
parameters differed across the hunger manipulation could be rejected with confidence,
the same null hypothesis for either of these parameters alone could not be rejected.

72

Figure 18. Posterior distributions for the effect of the experimental manipulation of hunger (fasted vs. fed)
on decision threshold (a parameter). Comparison of these distributions directly do not provide evidence of a
significant effect of feeding on posterior estimates (P(a(Fast) < a(Fed) = 0.716). Directionality of these
posterior estimates do, however, imply there may be increased decision-thresholds following feeding.

Figure 19. Posterior distributions for the effect of the experimental manipulation of hunger (fasted vs. fed)
on non-decision times (t parameter). Comparison of these distributions directly do not provide evidence of
a significant effect of feeding on posterior estimates (P(t(Fast) < t(Fed) = 0.901). Directionality of these
posterior estimates do, however, imply there may be increased non-decision times following feeding.

Analysis 3: ITC Choice Results
Fast vs. Fed Choice Results

73
Model comparisons for the mixed effects logistic regressions resulted in a
significant interaction effect between log of K and experimental day (c
2
=8.7166, DF=1,
p=0.0032). Looking at this final model, people tended to have an increased probability of
choosing the more delayed option as the log of K increased (b=6.873, Standard Error =
1.2, p<0.0001), this verifies that as intertemporal choice sets better favored the delayed
option within a subject by giving the delayed option a larger per-day interest rate people
were more likely to pick that option over the smaller sooner option. Alternatively, when
people were making decisions during the fed condition, they were slightly more likely to
choose the immediate option when compared to the Fasted day (b=-2.81, Standard Error
= 1.12, p=0.0128, in conflict with some prior works). It is worth noting that when this
model is compared to a base model without a fixed effect of day, the main effect of day is
no longer significant. This implies that it is not the Day effect alone driving this effect,
but is instead the interaction among day and the offered choice set. When an interaction
term between homeostatic condition and the log of K is added, people in the fed
condition had a smaller Beta coefficient for the log of k than they did during the fasted
day indicating that as the delayed option is made more appealing, individuals had a
higher probability of choosing the smaller sooner option on the fed day as the delayed
option was titrated to be more attractive than when they were in the fasted condition (b=-
0.587, Standard Error = 0.199, p=0.003). Conceptually this effect should be driven by a
change in indifference point, however, if hunger is altering stochasticity in some way this
could be involved in some way in the presence of an interaction. If homeostatic state
changes the inflection point of the logistic fit to the choice data, this would indicate a
change in the indifference point due to homeostatic condition. Alternatively, homeostatic

74
state may instead be shifting the log likelihood of choosing the larger later option as the
log of k changes which may be playing a role in the presence of an interaction effect.
In an attempt to unpack the interaction seen in the mixed-effects logistic
regression model post-hoc, a simple GLM was run on each individual subject’s choice
data on each experimental day using a binomial distribution through the glm function in
R statistical software. The fitting of a per-day model led to the outputting of both a log-
likelihood score for the log of k, and calculation of the inflection point for the logistic fit
(via the predict() function in R) for each day for each subject. Due to the completion of
only 1 run on the fed day, one subject was underpowered for use of the predict function
in determining an inflection point, so was excluded from later analysis. Paired samples t-
tests (via the t.test() function in R with the paired flag turned on) provided no evidence
that the mixed-effects interaction was driven by homeostatic differences in either
stochasticity (t = -0.93, df = 18, p-value = 0.367) or indifference point (t = 0.85, DF = 17,
p-value = 0.405) across all subjects. Due to this lack of main effect, homeostatic changes
in both stochasticity and indifference point were compared with trait-level discounting
rate (collected during the screening day to provide adaptive algorithm with an accurate
starting point) via linear regression using the lm() function in R to test for the role of
individual differences in discounting rate and how that relates to homeostasis-based
changes in choice behavior. Results of these regressions indicated that homeostatic state
differences in indifference point (F(1,16)=4.402, p=0.052, see Figure 20 below) but not
stochasticity (F(1,17)=0.204, p=0.657) are related to individual differences in
discounting rates. Conceptually, this linear relationship between trait-level discounting
and indifference point implies a polarization of choice behavior in response to

75
deprivation states thus making shallow discounters shallower, and steep discounters
steeper. In an attempt to validate the last finding of changing indifference points based on
hunger condition, the log of baseline discounting was compared to the mean of estimated
log of discounting on each experimental day (by averaging log of k for offered options
across the 160 questions on each day). Results of this validation attempt indicated that
there was a significant trend in the same direction (F(1,17)=3.555, p=0.076; see Figure
21).

Figure 20. Visualization of the marginally significant post-hoc finding that the log of k as estimated on the
screening day is related to the difference in estimated indifference points on the scale of log of k from the
160 trials on each experiment day (F(1,16)=4.402, p=0.052). Results indicate that shallow discounters
(more negative baseline log of k) become shallower, whereas steeper discounters (closer to zero on baseline
log of k) appear to become even steeper in their discounting as estimated by inflection point.
−2
−1
0
1
−7 −6 −5 −4 −3
Baseline log of Discounting (k)
Fast − Fed log(K) Inflection Point
Trait Discounting Predicts Homeostatsis Changes
in Logistic Inflection Points

76

Figure 21. Follow-up post-hoc test investigating the role of the log of baseline discounting acquired on the
screening session with the difference in estimated log of k by averaging across the 160 trials on each
experimental day (i.e. fast and fed). Results of this comparison provide further evidence for the association
seen between baseline discounting and estimated indifference points through a similar trending pattern
(F(1,17)=3.555, p=0.076) where steep discounters get steeper and shallow discounters get shallower in
their estimated discounting.

Analysis 4: fMRI Results
Main Effects
Choice > Baseline
There was widespread activation during the Choice > Baseline contrast. Due to
the high level of significance, and more conservative corrected p-value of 0.001 was used
to threshold the whole-brain contrasts. At this conservative threshold, there remained
significant activation across several regions previously shown to be involved in
intertemporal choices such as in the visual cortex, insula, striatum, cingulate and
cerebellum among other regions (see Figure 22 and Table 13).
−1
0
1
2
−7 −6 −5 −4 −3
Baseline log of Discounting (k)
Fast − Fed mean log(K)
Trait Discounting Predicts Homeostatsis Changes
in Estimated Delay Discounting (log k)

77
Figure 22. Whole-brain activation map for the main effect of Hard choices greater than implicit baseline
across both days. Significant activation for this contrast was widespread so to enhanced interpretability of
activation clusters, thresholding for this map was set to p<0.001 in an attempt to better parse task-relevant
regions. There was significant activation in the bilateral visual cortex, striatum, cingulate and insula among
a few other regions. Thresholding using the corrected p-value output of FSL’s randomise permutation
algorithm that utilizes threshold-free cluster enhancement to search for ‘cluster-like’ behavior in voxels.

Cluster Voxels Max
(1-p)
Side Brain Region Max
X
Max
Y
Max
Z
1 40651 0.9998 L/R Visual Cortex/Insula/Cingulate/
Striatum/Thalamus/Hippocampus/
Cerebellum
20 24 -12
2 1595 0.9998 L Anterior Insula/Caudate/Putamen -32 20 4
3 44 0.9998 L Middle Insula -40 -6 8
Table 13. Significant whole brain cluster for the main effect of hard intertemporal choice > baseline
contrast averaged across the two experiment days. Due to widespread activation for this contrast,
thresholding was set to p<0.001. Even at this conservative threshold, there was widespread activation
across reward and decision-making regions.

Choice > Control Choice
No significant clusters for this contrast meet thresholding.
Smaller Sooner Choice > Larger Later Choice
No significant clusters for this contrast meet thresholding.
Magnitude Tracking
No significant clusters for this contrast meet thresholding.
Day Differences
Choice > Baseline

78
No significant clusters for this contrast meet thresholding.
Choice > Control Choice
No significant clusters for this contrast meet thresholding.
Smaller Sooner Choice > Larger Later Choice
No significant clusters for this contrast meet thresholding.
Magnitude Tracking
There was one cluster in the right medial Orbital Frontal Cortex that showed a significant
effect of homeostatic state (t(18)=5.82, p=0.048). Extraction of this cluster’s data
provided evidence that hunger states are associated with increased magnitude tracking in
this cluster when compared to the fed day (see Figures 23 and 24, as well as Table 14).
Figure 23. Whole-brain activation map for the main effect of experiment day during magnitude tracking
during ‘hard’ trials. When comparing hunger day, a single cluster the medial orbital frontal cortex met the
corrected p<0.05 whole-brain threshold. Thresholding using the corrected p-value output of FSL’s
randomise permutation algorithm that utilizes threshold-free cluster enhancement to search for ‘cluster-
like’ behavior in voxels.

Cluster Voxels Max
(1-p)
Side Brain Region Max X Max Y Max Z
1 1 0.952 R Orbital Frontal Cortex 20 24 -12
Table 14. Significant whole brain cluster for the main effect of hunger condition during magnitude tracking
for ‘hard’ trials. Only a single cluster met significance in the medial orbital frontal cortex with the
minimum p-value for this cluster reaching p=0.048 (t(18)=5.82).

79

Figure 24. Bar plot showing estimated magnitude tracking during the fasted and fed day. Visualization of
this effect illustrates increased magnitude tracking in the medial orbital frontal cortex during the fasted day,
with no apparent magnitude tracking during the fed day (t=5.82, p=0.048).

ROI Analysis
Main Effects
Hard Choices > Control Choices
Valuation ROIs
There is no significant main effect of hunger condition across Valuation spherical
ROIs (M[Difficulty] = 2.205, 95% HPD [-2.658, 6.98]).
Core ROIs
There was a significant main effect of choice difficulty across Core spherical
ROIs (M[Difficulty] = -16.99, 95% HPD [-24.37, -9.89]). Follow-up mixed models for
each ROI indicated that there was widespread activation across the Core ROIs when
comparing Hard to Easy choices (see Figure 25 and Table 15).

80
Figure 25. Core ROI activation in arbitrary units for the Hard vs. Easy contrast provided evidence of
significant activation in all tested ROIs but the medial Prefrontal Cortex through the fitting of a hierarchical
Bayesian model across all ROIs (M[Difficulty] = -16.99, 95% HPD [-24.37, -9.89]).

Core ROI Hard > Easy Follow-up t-tests
ROI Day Avg. z-score p-value
Left Temporal Parietal -4.24 p<0.001**
Right aiPFC 0.26 0.80
Right aiPFC 2 3.32 p=0.001**
Inferior Parietal -5.15 p<0.001**
Inferior Parietal 2 -5.27 p<0.001**
Medial PFC -1.41 P=0.16
Middle Temporal Gyrus -2.79 p=0.005**
Splenial PCC -4.17 p<0.001**
Table 15. Outputs of the follow-up linear mixed effect models across each of the Core ROIs across both
hunger conditions. Follow-up mixed model testing on the hierarchical Bayesian model showing a main
effect of choice difficulty provided evidence that nearly all core ROIs were involved in the tracking of
choice difficulty.

SS Choices > LL Choices
Valuation ROIs

81
There is no significant main effect of SS > LL choices across Valuation spherical
ROIs (M[SS] = -4.27, 95% HPD [-8.6, 0.04]).
Core ROIs
There is no significant main effect of SS > LL choices across Core spherical ROIs
(M[SS] = -0.44, 95% HPD [-5.16, 4.32]).
Hard Choice > Control Choice Magnitude Tracking
Valuation ROIs
There is no significant main effect of hard > control magnitude tracking across
Valuation spherical ROIs (M[Magnitude] = -0.61, 95% HPD [-5.66, 4.75]).
Core ROIs
There is no significant main effect of hard > control magnitude tracking across
core spherical ROIs (M[Magnitude] = -1.11, 95% HPD [-6.99, 4.63]).
Day Differences
Hard Choices > Control Choices
Valuation ROIs
There is no significant main effect of hunger condition across Valuation spherical
ROIs (M[Fed] = 2.205, 95% HPD [-2.658, 6.98]).
Core ROIs
There is no significant main effect of hunger condition across Core spherical
ROIs (M[Fed] = 0.056, 95% HPD [-5.062, 5.044]).
SS Choices > LL Choices
Valuation ROIs

82
There is no significant main effect of hunger condition across Valuation spherical
ROIs (M[Fed] = -1.379, 95% HPD [-5.141, 2.507]).
Core ROIs
There is a significant main effect of hunger condition when comparing SS and LL
choice across Core spherical ROIs (M[Fed] = -5.653, 95% HPD [-8.985, -2.085]).
Follow-up t-tests indicate increased fasting differential brain response to SS choices > LL
choices across Core ROIs, with only the right anterior inferior PFC ROIs reaching
significance (see Table 16 and Figure 26).
Figure 26. Fast – fed core ROI activation in arbitrary units for the contrast of smaller sooner choice > larger
later choice. Across all ROIs there appears to be slightly increased activity on the fasted day, with only one
of the two right anterior inferior Prefrontal Cortex ROIs reaching significance during follow-up testing
(t(18)=2.15, p=0.045).

83
Core ROI SS Choice > LL Choice Follow-up t-tests
ROI t-stat (Fast>Fed) p-value
Left Temporal Parietal 1.69 0.11
Right aiPFC 2.15 0.045*
Right aiPFC 2 1.93 0.0699
Inferior Parietal 1.50 0.15
Inferior Parietal 2 0.79 0.44
Medial PFC -0.64 0.53
Middle Temporal Gyrus 0.60 0.56
Splenial PCC 1.07 0.30
Table 16. Outputs of the follow-up paired sample t-tests across the core ROIs during the SS > LL contrast.
Results of the follow-up tests provided some evidence that only one of the two right aiPFC ROIs crossed
the threshold for a significant main effect of experiment day.

Hard Choice > Control Choice Magnitude Tracking
Valuation ROIs
There is a significant main effect of hunger condition when comparing hard
choice magnitude with control choice magnitude tracking across Valuation spherical
ROIs (M[Fed] = -10.704, 95% HPD [-14.906, -6.444]; see Figure 27). Follow-up t-tests
indicate increased fasting differential brain response to choice magnitude > control
magnitude across Valuation ROIs, with the midbrain and VST ROIs reaching
significance, and the Left Insula trending in the same direction (see Table 17).

84
Figure 27. Fast – fed valuation ROI activation in arbitrary units for the magnitude tracking contrast. Across
all ROIs there appears to be increased activity on the fasted day, with only the left Insula (t(18)=1.83,
p=0.08), ventral striatum (t(18)=3.21, p=0.005) and midbrain (t(18)=2.25, p=0.04) reaching significance
during follow-up testing.

Valuation ROIs Choice > Control Magnitude Follow-up t-tests
ROI t-stat (Fast>Fed) p-value
Left Insula 1.83 0.08
PCC 0.90 0.38
Right Insula 1.44 0.17
VST 3.21 0.005**
Lateral OFC 1.22 0.24
Medial PFC 1.50 0.15
Midbrain 2.25 0.04*
Table 17. Outputs of the follow-up paired sample t-tests across the core ROIs during the magnitude
tracking contrast. Results of the follow-up tests provided some evidence that the ventral striatum and
midbrain valuation ROIs crossed the threshold for a significant main effect of experiment day.

Core ROIs

85
There is a significant main effect of hunger condition when comparing hard
choice magnitudes and control choice magnitudes across Core spherical ROIs (M[Fed] =
-8.735, 95% HPD [-12.774, -4.539]; see Figure 28). Follow-up t-tests indicate increased
fasting differential brain response to choice magnitude relative to control magnitude
across Core ROIs, with the Left Temporal Parietal ROI reaching significance, and the
Splenial PCC and Middle Temporal Gyrus trending in the same direction (see Table 18).
Figure 28. Fast – fed core ROI activation in arbitrary units for the magnitude tracking contrast. Across all
ROIs there appears to be increased activity on the fasted day, with only the left Temporal Parietal
(t(18)=2.24, p=0.038), Posterior Cingulate (t(18)=1.98, p=0.06) and middle Temporal Gyrus (t(18)=1.96,
p=0.066) reaching significance or trending during follow-up testing.

Core ROIs Choice > Control Magnitude Follow-up t-tests
ROI t-stat (Fast>Fed) p-value
Left Temporal Parietal 2.24 0.038*

86
Right aiPFC 1.47 0.16
Right aiPFC 2 1.00 0.33
Inferior Parietal 0.62 0.55
Inferior Parietal 2 -0.03 0.98
Medial PFC 1.50 0.15
Middle Temporal Gyrus 1.98 0.06
Splenial PCC 1.96 0.066
Table 18. Outputs of follow-up paired samples t-tests for core ROIs comparing the fasted and fed days for
the magnitude tracking contrast. Results provide evidence of increased magnitude tracking in the left
Temporal Parietal, middle Temporal, and spenial PCC during the fasted day when compared to the fed day.

Discussion
The present study sought to investigate the behavioral and neural effects of
experimental manipulation of homeostatic state, specifically hunger status, during
intertemporal choice decision-making. Interestingly, when hungry subjects engaged in
“hard” intertemporal decision making, there was a significant main effect of homeostatic
state on reaction time, specifically with significantly increased response times during the
fed condition. This finding is in line with past work that used a standardized set of
intertemporal choices as opposed to individualized titration procedure used in this work
(Wierenga et al., 2015), providing evidence that this response time effect may be
significantly different across hunger conditions when all choice-sets hover around the
individuals indifference point. How or why there is either a speeding on the fasted day or
a slowing on the fed day was addressed through computational modeling of response
times using the Hierarchical Drift Diffusion Model.
There are a multitude of cognitive processes working as the decision-maker
reveals preferences to multi-attribute choice sets like the intertemporal choice sets used in
the current work. In an attempt to unpack the significant RT findings seen in this study,
Easy vs. Hard trials and Fast vs. Fed during only hard trials were fit to a variety of
different hierarchical drift diffusion models. Ratcliff, Smith, Brown, & McKoon, 2016

87
provided early evidence that task difficulty in simple perceptual designs (e.g. motion
discrimination tasks in which you choose whether the majority of dots are going left/right
or up/down) is related to steeper drift rates. The best fitting model for the effects of
choice difficulty on intertemporal choice in this dataset were in line with this finding (i.e.
the best fitting model did allow drift rate to vary by question difficulty), however,
experimental data provided evidence that both decision thresholds and non-decision times
also differ according to task difficulty during intertemporal choice. Taken together, the
best-fitting difficulty HDDM model was in line with prior research on task difficulty and
provides an indication of the validity of fitting intertemporal choice sets with the HDDM
due to prior works instead focusing on different sequential sampling models (e.g. Linear
Ballistic Accumulator Models in Rodriguez et al., 2015).
In the current dataset, hierarchical Bayesian estimations of the effect of disruption
of homeostasis through a 12 hour fast when compared to recently satiation provides
evidence that feeding was associated with a slowing of response times during ‘hard’
intertemporal choice sets. The best fitting HDDM model as determined via a
minimization of DIC score, and through comparisons of empirical data with simulated
datasets estimated via the best-fitting model posterior distributions provided evidence of
hunger-based modulation of the decision threshold and non-decision time parameters, a
and t respectively. Hunger-based modulation of the decision threshold parameter in the
HDDM, while holding drift rate constant across days, indicates that the change in
response time is not necessarily due to increased or decreased slope of the value
accumulation slope (information from the environment is not accumulated faster/slower
based on hunger status) but is likely due instead to the adjustment of decision thresholds.

88
This homeostasis dependent shift in decision threshold could in theory be responsible for
making fasted decisions both faster and noisier. By collapsing the decision boundaries
inwards when fasted, this will work to decrease reaction times and will increase the
probability that a spontaneous noise-based fluctuation will lead to an accumulation of
value that surpasses the decision threshold. This collapsing inwards of decision
thresholds when fasting is in line with the prior neurobiological work in humans and
animals in which hunger-based homeostatic imbalance significantly decreases reward-
thresholds in dopaminergic reward networks in the brain (e.g., Bruijnzeel et al., 2011).
This effect of decreased reward thresholds can be conceptualized as reflecting a decrease
in the necessary amount of reward or evidence needing to be accumulated before a
decision or action is triggered when the individual is fasted, which is in line with the
findings from the fitting of the computational HDDM model of intertemporal choice RTs
in the current work.
Interpretation of the hunger based changes in the non-decision time parameter is
not as straight forward as the decision-threshold findings just discussed. This hunger-
based change in non-decision time is likely not due to changing reward thresholds in the
brain, however, this finding may still be a simple behavioral byproduct of the
experimental manipulation of feeding. In addition to increasing self-reported satiety and
decreasing self-reported hunger, feeding (especially meals high in carbohydrates like the
manipulation used here) results in increased reports of sleepiness and even sleep arousals
and slow wave sleep expression while awake (St-Onge, Roberts, Shechter, & Choudhury,
2016). The non-decision time parameter represents the combination of both defined and
undefined cognitive processes that go in to the initiation of sampling from the

89
environment (i.e. perception, movement initiation and execution (Wiecki et al., 2013)), if
the decision-maker is in a state of increased sleepiness or decreased awareness this very
well may lead to an increased estimation of the non-decision time parameter as is seen
here. Unfortunately self-reported data on sleepiness was not collected in the current work,
but future works should seek to correlate individual differences in feeding-based
increases in non-decision times with feeding induced increases in sleepiness.
Despite the Fed day having a significantly longer RT on ‘hard’ intertemporal
choices than the Fasting day (see Chapter 2 for full details of significant main effect),
neither of the inferential tests on the posterior distributions of the parameters that were
allowed to vary based on hunger status met the pre-defined threshold of a 95% difference
in posterior distributions. The use of the HDDM in the current work as a computational
model to probe the effects of homeostatic state on of response times was conceptualized
post-hoc and can be considered as exploratory. The lack of a significant difference
between the posterior distributions of the estimated model parameters is likely due to
either not enough trials within each condition (uninformative priors makes more trials
very helpful in determining accurate posterior distributions), or not enough subjects to
see a significant main effect of condition on these parameters, or some combination of
the two. Future studies that intend to use similar computational modeling of response
times should seek to increase the within subject number of trials in each condition, and
should increase the overall number of subjects used to minimize across subject variance
and its negative impact on parameter inference for condition level posterior distributions.
Switching now to hunger-based alterations in delay discounting preferences, past
studies seem to converge on increased discounting and immediacy bias when subjects

90
complete intertemporal choice sets during food deprivation states (see Ashton, 2015;
Kuhn et al., 2014; X. T. Wang & Dvorak, 2010). However, there does exist in the recent
literature a dataset in which there was no effect of hunger status on discounting (Ridder et
al., 2014). This may imply that either the main effect of hunger status on discounting may
be subtle, or it may be due to some other factor such as individual differences across
subjects similar to the findings relating hunger and risk aversion (Levy et al., 2013). In
the current work, there does not appear to be a main effect of hunger status on
discounting when the all smaller sooner options are available today. There does however
appear to be a significant interaction between the offered choice set and hunger status on
choice, with follow-up tests indicating that this may specifically be related to the
relationship between trait-level discounting and hunger status with hunger states making
steep discounters steeper and shallow discounters shallower.
In Alcohol Myopia, alcohol narrows attention when consumed so that the
individual processes fewer cues less well (Josephs & Steele, 1990). More specifically,
behavior will tend to be more or less inhibited when drinking alcohol depending on
whether attention is paid to inhibiting or disinhibiting cues. In other words, alcohol
consumption can greatly reduce the influence of secondary cues or considerations,
making the individual relatively insensitive to all but the dominant cues. Likely due to
this model’s success and replicability, these attentional changes following alcohol
consumption have been extended to include anything that occupies cognition while the
individual is in a time of high inhibitory conflict be it a substance like alcohol or even
simply increases in non-emotional cognitive load (Ward & Mann, 2000). This attentional
myopia is manifested in eating behaviors through the increase in food consumption in

91
restrained eaters under cognitive load and the decrease in food consumption in
unrestrained eaters under the same conditions. This theory can be construed as a latent
underlying cognitive process affecting a variety of experimental paradigms.
The current work sought to manipulate the homeostatic condition of the
individual while putting individuals in a situation of high inhibitory conflict, via a “hard”
intertemporal choice task. When viewed through the lens of attentional myopia, the effect
of experimental manipulation of hunger status on intertemporal choice may result in
increased or decreased delay discounting behavior depending on whether increased
attention is paid to inhibitory (in this case $ amount) or disinhibitory (in this case
immediacy) cues. Work combining eye-tracking with intertemporal choice has shown
that an increased attention to delayed option attributes and cues is associated with
decreased delay discounting (Radu, Yi, Bickel, Gross, & McClure, 2011). The current
finding of steeper discounting amongst steep discounters and shallower discounting
amongst shallow discounters under deprivation states are in line with these lines of
research. Furthermore, these findings imply that there is likely an even greater attentional
shift amongst shallow and steep discounters under food deprivation states when
compared to states of satiation, something that future works should seek to confirm
perhaps through the use of eye-tracking. In addition, the current work specifically
encourages attention to amount cues which appears to be interacting with baseline
discounting in predicting the effects of hunger on delay discounting preferences. Future
works ought to experimentally manipulate both amount and delay in an attempt to
experimentally control what aspects of the multi-attribute intertemporal choice set are

92
most salient to investigate whether this changes the trait-level association with baseline
discounting rate.
To date, only Wierenga et al., 2015 has investigated the neurobiology of how
experimental manipulation of homeostatic state affects intertemporal choice processes.
Though this study largely focused on the differences between recovered anorexics and
control women, findings in the control group only center around an increase in activity in
reward-related brain regions on the hunger day during processing of immediate reward
(referred to as a Beta regressor in the Dual-systems literature, e.g. Samuel M. McClure et
al., 2004), and increased cognitive control activity across all choices (referred to as a
Delta regressor in the Dual-systems literature) when participants are fed. The current
work includes only questions that would be considered Beta in the dual systems literature
(due to the smaller sooner option always being available today) in an attempt to better
experimentally control the attributes of intertemporal choice sets that are changing during
the experiment, which makes the same Beta/Delta comparisons not possible in the current
work.
In line with past work, the current experimental manipulation provides further
evidence of increased recruitment of value-tracking regions of the medial OFC in
magnitude tracking during hunger. Narrowing in on the ROIs chosen based on a recent
meta-analysis, the current work also finds increased SS > LL differences in activity in
“Core” ROIs that are thought to be involved in cognitive control and prospective
thinking. Though without a significant main effect of this contrast across the two
experiment days, it would perhaps be wise to interpret this finding with some caution.
Directionality of this finding implies that there is increased recruitment of these

93
inhibitory control regions when the individual chooses the larger later option on the fed
day driven in large part by the pair of ROIs in the right anterior inferior PFC.
In the current work there was increased activity in both the valuation and core
ROIs during magnitude tracking on the fasted day driven primarily by the dopaminergic
VST and midbrain in the valuation ROIs and in temporal and cingulate regions in the
core ROIs. Along with the whole brain findings in the value-tracking medial OFC during
magnitude tracking, these findings imply that perhaps hunger states are resulting in
increased tracking of minor changes in reward magnitude. The post-hoc finding of a
relationship between individual differences in trait-level delay discounting and the effects
of homeostatic state on indifference points during intertemporal choice may hint that the
behavioral manifestation of a hunger-based disturbance in homeostasis may even be
related to individual differences in neural response to intertemporal choice, something
that should be investigated in future works.
Experimental manipulation of hunger appears to create potentially large
systematic changes in behavioral and neural response during an intertemporal choice
task. Across subjects, hunger states are associated with decreases in reaction time relative
to feeding and may be associated with decreased decision thresholds and increased non-
decision times. Additionally, hunger may be associated with increased recruitment of
core and valuation ROIs during magnitude tracking and decreased recruitment of core
ROIs during choice of the LL option. In addition to these main effects, there may even be
a role of individual differences in the behavioral effects of food-deprivation during
intertemporal choice with steep discounters becoming steeper, and shallow discounters
becoming shallower. Systematic recruiting of high and low discounters during the

94
manipulation of hunger in follow-up works should investigate if this behavioral finding
extends to neural response to intertemporal choice sets in a better powered sample.

95
Chapter 4: General Discussion
The current dissertation sought to expand upon the extensive behavioral and
computational modeling of the effects of homeostatic imbalance on delay discounting
already in the experimental literature. Prior works investigating intertemporal choice and
hunger in humans seem to converge on three major findings, and two less established
findings: increase in immediacy bias, increased convexity of the utility function, increase
in delay discounting, and in the lone study investigating RT and biology, decreased
response times and increased neural response in reward-related regions of the limbic
system respectively. The current work sought to specifically target and better explain the
response time and neurobiological effects of experimental manipulation of hunger on
intertemporal choice preferences.
In line with past findings, the current work finds decreased reaction times during
intertemporal choice sets on the fasted day when compared to when participants were
‘comfortably full’ replicating this finding across various trial difficulties and extending
this finding to exclusively ‘hard’ trials (Wierenga et al., 2015). In attempt to better
understand which cognitive processes are being modified by the hunger-based imbalance
in homeostasis, reaction time data for the ‘hard’ intertemporal choice data a Hierarchical
Drift Diffusion Model was fit to the choice and RT data (Wiecki et al., 2013). Since this
computational modeling was done post-hoc in an attempt to untangle the current
previously published RT findings of hunger on intertemporal choice, all possible model
combinations were tested to ensure the best fitting model was chosen. The best fitting
model provides some early evidence that hunger may be interfering with some
combination of decision thresholds (the point at which sampling has reached a decision

96
threshold in either direction) and non-decision times (e.g. motor preparation, sensation
etc.).
Through the use of standardized sets of intertemporal choices, or by having
subjects complete a series of convex budget sets, past experimental evidence provides
evidence of increased immediacy bias, utility function convexity, and delay discounting
when participants are hungry (Ashton, 2015; Kuhn et al., 2014; X. T. Wang & Dvorak,
2010). The current dissertation was specifically designed to follow up on the past
immediacy bias and delay discounting findings. Thanks to the use of pre-testing,
experimental control over ‘easy’ and ‘hard’ intertemporal choice sets was maintained for
the entirety of both experimental sessions. The current design was also specifically
designed to be sensitive to magnitude effects in particular due to all delays being fixed at
Today and in 1 Month across the intertemporal choice sets given to subjects. Results of
the analysis of choice data unsurprisingly did not lead to increased probability of
choosing the SS or LL option across trials due to adaptive algorithm used in the task (S.
Luo et al., 2009). However, there was a significant interaction between the hypothetical
indifference point for the delay discounting parameter for the offered choice set with
homeostatic state. Investigation of this effect provided evidence that hunger status is
associated with a shift in the inflection point of the logistic curve, not in the stochasticity
of choices. More specifically, individual differences in trait-level discounting were
related to hunger-based changes in indifference points with shallow discounters
becoming shallower when hungry, and steep discounters becoming steeper. Though the
role of individual differences is new to the delay discounting literature, evidence of the

97
role of Trait-level variables in determining hunger’s effect on behavioral economic
variables has a history in the risk aversion literature (Levy et al., 2013).
To date, only one study has investigated the neurobiology of hunger, satiety and
intertemporal choice sets. Wierenga et al., 2015 took a dual-systems approach to the
design and analysis of the intertemporal choice during fMRI by using a standardized set
of intertemporal choice sets. This research was extended in the current work via the use
of titration procedures and the simplifying of possible delays in order to further
investigate the neurobiology of homeostatic imbalance on intertemporal choice sets.
Findings from the current work indicate increased differential response in the Core
network of ROIs to Smaller Sooner > Larger Later, namely increased use of the Core
ROIs during the choice of Larger Later choice on the satiated Day. In addition, there was
a significant whole brain cluster in the value tracking medial OFC, and across both the
Core and Valuation ROIs when investigating value-tracking magnitude effects in the
brain. Specifically, on the hungry day across all ROIs (Core and Valuation) and the
whole brain medial OFC cluster there was increased neural tracking of reward
magnitudes. This hunger-based modulation of value-tracking for monetary rewards
across these regions is in line with the critical role of these same regions in hunger-based
fluctuations in the reinforcing value of foods (Rolls, 2016).
In line with our initial hypothesis, the behavioral and neurobiological effects of
hunger during intertemporal choice preferences highlighting decision-thresholds and
magnitude tracking provide some early indication that the underlying mechanism may be
related to reward sensitivity. In an attempt to better understand what hunger and satiety
are doing to reward sensitivity, the intertemporal variant of the Monetary Incentive Delay

98
task was also administered to individuals in the scanner. Behavioral results indicate that
there was a large effect of homeostatic state on reaction time, namely that only on the
fasted day did individuals increase their RT to rewarded cues. This increase in reward
reactivity to the simple RT portion of the intertemporal MID task on the fasted day
provides evidence that satiety may result in some form of a generalized decrease in
reward reactivity due to feeding’s restoration of homeostasis.
Examination of the neural data collected during the intertemporal MID, which
focused on the reward anticipation period, paints a similar picture. Though there were no
significant whole-brain findings when comparing homeostatic state, the limbic ROIs
examined provided evidence that there is increased neural response among these regions
to rewarded cues when compared to baseline, specifically on the fasted day when
compared to the fed day. Though these findings were significant as a main effect of day
across all limbic ROIs, this main effect appeared to be driven in large part by increased
response to rewarded cues in the bilateral caudate anatomical ROIs.
Though this finding in the bilateral caudate was across all reward cue types (as was
the response time finding), the specificity to the caudate may provide some evidence as to
the manner in which hunger is affecting reward sensitivity in a way that would affect
intertemporal choice preferences. The caudate’s involvement in reward and choice along
with food deprivation’s ability to increase the caudate’s hold over reward thresholds has
been in the biological literature for decades (Brady, Boren, Conrad, & Sidman, 1957).
More specifically, there is evidence that the dorsal caudate is responsible more
specifically in the neural tracking of marginally diminishing utility functions (Alex Pine
et al., 2009). Authors of this study imply that the dorsal striatum may be involved in

99
integrating subjective valuation systems inherent to time and magnitude thereby
providing an overall metric of value to guide choices. Since this finding was not specific
to either magnitude or delay, there is a chance this feeding-based increase in reward
reactivity in the bilateral caudate may be the mechanism for how hunger alters a variety
of behavioral economic preferences (e.g. risk aversion in Levy et al., 2013), consumption
of non-food items (e.g. Xu et al., 2015), though this is merely speculation.
The behavioral and neurobiological findings in the current work further my initial
hypothesis that hunger not only results in systematic changes in reward thresholds and
subsequent choices of food items (e.g. Wansink et al., 2012), but through it’s effects on
basic reward neurocircuitry can affect reward thresholds in general. In a fascinating early
work, hungry cats were seen to dramatically increase self-stimulation of the dorsal
striatum when compared to sated cats (Brady et al., 1957). In line with this finding, food
restriction in animals has also been shown to dramatically increase the central rewarding
effects of various drugs of abuse like amphetamine, PCP and nicotine (Vaca & Carr,
1998). Looking instead at human subjects, satiety decreases resting activity in reward
circuitry and increased activity in cognitive control circuitry (Li et al., 2012). The role of
visceral factors likely as a behavioral downstream byproduct of these biological changes
has become one of the major tenets of modern-day science (e.g. Bechara & Damasio,
2005; Loewenstein, 1996). It has only been more recently that these effects have been
generalized to all reward-based decision processes and choices affecting multiple aspects
of decision-making and risk taking (Levy et al., 2013), which has been termed ‘Spillover’
of hunger based changes in reward sensitivity aimed at altering food preferences and

100
instead affecting food-irrelevant choices (Fung et al., 2017). With recent interest being
turned more specifically towards the effects of hunger on delay discounting preferences.
In rats, hunger has been shown to affect impulsive decisions only when the choices
are on the order of minutes, not seconds (Eisenberger, Masterson, & Lowman, 1982).
This food-based intetemporal choice was later extended to humans, showing people will
make dynamically inconsistent choices over time dependent on current and future hunger
states (Read & van Leeuwen, 1998). Which was then extended to standard monetary
intertemporal choice sets in humans, with hunger increasing immediacy bias, discounting
and utility function convexity (e.g. Ashton, 2015; Kuhn et al., 2014), decreasing RT, and
increasing reward tracking (Wierenga et al., 2015). The current dissertation is an initial
attempt to piece together the biological literature hinting at a hunger-based alteration in
reward sensitivity and the behavioral literature showing these systematic shifts in delay
discounting preferences. Though his relationship between reward sensitivity and hunger-
based shifts in delay discounting (among other behavioral economic estimates of state-
varying utility functions) is very much still at a broad theoretical level, this will hopefully
guide future works in further investigating this fascinating link.

101
References:

Aarøe, L., & Petersen, M. B. (2013). Hunger Games : Fluctuations in Blood Glucose
Levels Influence Support for Social Welfare. Psychological Science, 24(12),
2550–2556. https://doi.org/10.1177/0956797613495244
Anderberg, R. H., Hansson, C., Fenander, M., Richard, J. E., Dickson, S. L., Nissbrandt,
H., … Skibicka, K. P. (2015). The Stomach-Derived Hormone Ghrelin Increases
Impulsive Behavior. Neuropsychopharmacology, 41(5), npp2015297.
https://doi.org/10.1038/npp.2015.297
Andersen, S., Harrison, G. W., Lau, M. I., & Rutstr\öm, E. E. (2008). Eliciting risk and
time preferences. Econometrica, 76(3), 583–618.
Araujo, I. E. De, Ferreira, J. G., Tellez, L. A., Ren, X., & Yeckel, C. W. (2012).
Physiology & Behavior The gut – brain dopamine axis : A regulatory system for
caloric intake. Physiology & Behavior, 106(3), 394–399.
https://doi.org/10.1016/j.physbeh.2012.02.026
Ardianto, C., Yonemochi, N., Yamamoto, S., Yang, L., Takenoya, F., Shioda, S., …
Kamei, J. (2016). Opioid System in the Lateral Hypothalamus egulate Feeding
Behavior Through Orexin and Gaba Neurons. NEUROSCIENCE, 320, 183–193.
https://doi.org/10.1016/j.neuroscience.2016.02.002
Ashton, L. (2015). Hunger Games : Does Hunger Affect Time Preferences? University of
Wisconsin-Madison, 1–57.
Avena, N. M., Rada, P. V, States, U., States, U., & Andes, D. L. (2015). Cholinergic
modulation of food and drug satiety and withdrawal. Physiology & Behavior,
106(3), 332–336. https://doi.org/10.1016/j.physbeh.2012.03.020.Cholinergic

102
Ballard, K., & Knutson, B. (2009). Dissociable neural representations of future reward
magnitude and delay during temporal discounting. NeuroImage, 45(1), 143–150.
https://doi.org/10.1016/j.neuroimage.2008.11.004
Bechara, A., & Damasio, A. (2005). The somatic marker hypothesis: A neural theory of
economic decisions. Games and Economic Behavior, 52, 336–372.
Bialkowski, J., Etebari, A., & Wisniewski, T. P. (2012). Fast profits : Investor sentiment
and stock returns during Ramadan. Journal of Banking and Finance, 36, 835–845.
https://doi.org/10.1016/j.jbankfin.2011.09.014
Bickel, W. K., Odum, A. L., & Madden, G. J. (1999). Impulsivity and cigarette smoking:
delay discounting in current, never, and ex-smokers. Psychopharmacology,
146(4), 447–454. https://doi.org/10.1007/PL00005490
Brady, J. V., Boren, J. J., Conrad, D., & Sidman, M. (1957). The effect of food and water
deprivation upon intracranial self-stimulation. Journal of Comparative and
Physiological Psychology, 50(2), 134.
Briers, B., Pandelaere, M., Dewitte, S., & Warlop, L. (2006). Hungry for Money: The
Desire for Caloric Resources Increases the Desire for Financial Resources and
Vice Versa. Psychological Science, 17(11), 939–943.
Bruijnzeel, A. W., Corrie, L. W., Rogers, J. A., & Yamada, H. (2011). Effects of insulin
and leptin in the ventral tegmental area and arcuate hypothalamic nucleus on food
intake and brain reward function in female rats. Behavioural Brain Research,
219(2), 254–264. https://doi.org/10.1016/j.bbr.2011.01.020
Cannon, W. B. (1929). The sympathetic division of the autonomic system in relation to
homeostasis. Archives of Neurology & Psychiatry, 22(2), 282–294.

103
Carter, R. M., Meyer, J. R., & Huettel, S. A. (2010). Functional Neuroimaging of
Intertemporal Choice Models: A Review. Journal of Neuroscience, Psychology,
and Economics, 3(1), 27–45. https://doi.org/10.1037/a0018046
Chib, V. S., Rangel, A., Shimojo, S., & Doherty, J. P. O. (2009). Evidence for a Common
Representation of Decision Values for Dissimilar Goods in Human Ventromedial
Prefrontal Cortex. The Journal of Neuroscience, 29(39), 12315–12320.
https://doi.org/10.1523/JNEUROSCI.2575-09.2009
Clewett, D., Luo, S., Hsu, E., Ainslie, G., Mather, M., & Monterosso, J. (2014). Increased
functional coupling between the left fronto-parietal network and anterior insula
predicts steeper delay discounting in smokers. Human Brain Mapping, 35(8),
3774–3787.
Damasio, A., & Damasio, H. (2016). Exploring the concept of homeostasis and
considering its implications for economics. Journal of Economic Behavior &
Organization, 126, 125–129.
Danziger, S., Levav, J., & Avnaim-Pesso, L. (2011). Extraneous factors in judicial
decisions. Proceedings of the National Academy of Sciences, 108(17), 6889–
6892. https://doi.org/10.1073/pnas.1018033108
Deichmann, R., Gottfried, J. A., Hutton, C., & Turner, R. (2003). Optimized EPI for
fMRI studies of the orbitofrontal cortex. NeuroImage, 19(2), 430–441.
https://doi.org/10.1016/S1053-8119(03)00073-9
Doyle, J. (2013). Survey of time preference, delay discounting models. Judgment and
Decision Making, 8(2), 116–135.

104
Duraffourd, C., Vadder, F. De, Goncalves, D., Delaere, F., Penhoat, A., Brusset, B., …
Stefanutti, A. (2012). Mu-Opioid Receptors and Dietary Protein Stimulate a Gut-
Brain Neural Circuitry Limiting Food Intake. Cell, 150, 377–388.
https://doi.org/10.1016/j.cell.2012.05.039
Eisenberger, R., Masterson, F. A., & Lowman, K. (1982). Effects of previous delay of
reward, generalized effort, and deprivation on impulsiveness. Learning and
Motivation, 13(3), 378–389. https://doi.org/10.1016/0023-9690(82)90016-9
Eklund, A., Nichols, T. E., & Knutsson, H. (2016). Cluster failure: Why fMRI inferences
for spatial extent have inflated false-positive rates. Proceedings of the National
Academy of Sciences, 113(28), 7900–7905.
https://doi.org/10.1073/pnas.1602413113
Figner, B., Knoch, D., Johnson, E. J., Krosch, A. R., Lisanby, S. H., Fehr, E., & Weber,
E. U. (2010). Lateral prefrontal cortex and self-control in intertemporal choice.
Nat Neurosci, 13(5), 538–539. https://doi.org/10.1038/nn.2516
Forstmann, B. U., Ratcliff, R., & Wagenmakers, E.-J. (2016). Sequential Sampling
Models in Cognitive Neuroscience: Advantages, Applications, and Extensions.
Annual Review of Psychology, 67(1), 641–666. https://doi.org/10.1146/annurev-
psych-122414-033645
Fung, B. J., Bode, S., & Murawski, C. (2017). High monetary reward rates and caloric
rewards decrease temporal persistence. Proc. R. Soc. B, 284(1849), 20162759.
https://doi.org/10.1098/rspb.2016.2759
Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevel/hierarchical
models. New York: Cambridge University Press.

105
Goldstone, A. P., Prechtl de Hernandez, C. G., Beaver, J. D., Muhammed, K., Croese, C.,
Bell, G., … Bell, J. D. (2009). Fasting biases brain reward systems towards high-
calorie foods. European Journal of Neuroscience, 30(8), 1625–1635.
https://doi.org/10.1111/j.1460-9568.2009.06949.x
Gorgolewski, K., Burns, C. D., Madison, C., Clark, D., Halchenko, Y. O., Waskom, M.
L., & Ghosh, S. S. (2011). Nipype: A Flexible, Lightweight and Extensible
Neuroimaging Data Processing Framework in Python. Frontiers in
Neuroinformatics, 5. https://doi.org/10.3389/fninf.2011.00013
Hare, T. A., Camerer, C. F., & Rangel, A. (2009). Self-Control in Decision-Making
Involves Modulation of the vmPFC Valuation System. Science, 324(5927), 646–
648. https://doi.org/10.1126/science.1168450
Hare, T. A., Hakimi, S., & Rangel, A. (2014). Activity in dlPFC and its effective
connectivity to vmPFC are associated with temporal discounting. Frontiers in
Neuroscience, 8:50. https://doi.org/DOI=10.3389/fnins.2014.00050
Havel, P. J. (2001). Information to the Brain : Short-Term and Long-Term Regulation of
Food Intake and. Experimental Biology and Medicine, 226, 963–977.
Ho, M. Y., Wogar, M. A., Bradshaw, C. M., & Szabadi, E. (1997). Choice between
Delayed Reinforcers: Interaction between Delay and Deprivation Level. The
Quarterly Journal of Experimental Psychology: Section B, 50(3), 193–202.
Huston-Lyons, D., & Kornetsky, C. (1992). Effects of nicotine on the threshold for
rewarding brain stimulation in rats. Pharmacology Biochemistry and Behavior,
41(4), 755–759.

106
Jenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002). Improved Optimization for
the Robust and Accurate Linear Registration and Motion Correction of Brain
Images. NeuroImage, 17(2), 825–841. https://doi.org/10.1006/nimg.2002.1132
Jenkinson, M., & Smith, S. (2001). A global optimisation method for robust affine
registration of brain images. Medical Image Analysis, 5(2), 143–156.
https://doi.org/10.1016/S1361-8415(01)00036-6
Johnson, D. J., Hopwood, C. J., Cesario, J., & Pleskac, T. J. (2017). Advancing Research
on Cognitive Processes in Social and Personality Psychology: A Hierarchical
Drift Diffusion Model Primer. Social Psychological and Personality Science,
8(4), 413–423. https://doi.org/10.1177/1948550617703174
Josephs, R., & Steele, C. (1990). The Two Faces of Alcohol Myopia: Attentional
Mediation of Psychological Stress (Vol. 99). https://doi.org/10.1037//0021-
843X.99.2.115
Kable, J. W., & Glimcher, P. W. (2007). The neural correlates of subjective value during
intertemporal choice. Nature Neuroscience, 10(12), 1625–1633.
Kahneman, D., & Tversky, A. (1984). Choices, values, and frames. American
Psychologist, 39(4), 341–350.
Kirby, K. N. (2009). One-year temporal stability of delay-discount rates. Psychonomic
Bulletin & Review, 16(3), 457–462. https://doi.org/10.3758/PBR.16.3.457
Kirby, K. N., Petry, N. M., & Bickel, W. K. (1999). Heroin Addicts Have Higher
Discount Rates for Delayed Rewards Than Non-Drug-Using Controls. Journal of
Experimental Psychology: General, 128(1), 78–87.

107
Kirby, K., Petry, N., & Bickel, W. (1999). Heroin Addicts have Higher Discount Rates
for Dealyed Rewards Than Non-Drug-Using Controls. Journal of Experimental
Psychology: General, 128(1), 78–87.
Knutson, B., Adams, C. M., Fong, G. W., & Hommer, D. (2001). Anticipation of
Increasing Monetary Reward Selectively Recruits Nucleus Accumbens. The
Journal of Neuroscience, 21, 1–5.
Knutson, B., Fong, G. W., Bennett, S. M., Adams, C. M., & Hommer, D. (2003). A
region of mesial prefrontal cortex tracks monetarily rewarding outcomes:
characterization with rapid event-related fMRI. Neuroimage, 18(2), 263–272.
Knutson, B., & Heinz, A. (2015). Commentary Probing Psychiatric Symptoms with the
Monetary Incentive Delay Task. Biological Psychiatry, 77(5), 418–420.
https://doi.org/10.1016/j.biopsych.2014.12.022
Knutson, B., Taylor, J., Kaufman, M., Peterson, R., & Glover, G. (2005). Distributed
Neural Representation of Expected Value. The Journal of Neuroscience, 25(19),
4806–4812.
Knutson, B., Westdorp, A., Kaiser, E., & Hommer, D. (2000). FMRI visualization of
brain activity during a monetary incentive delay task. Neuroimage, 12(1), 20–27.
Kuhn, M. A., Kuhn, P., & Villeval, M. C. (2014). Self Control and Intertemporal Choice :
Evidence from Glucose and Depletion Interventions. Working Paper.
Lempert, K., Glimcher, P., & Phelps, E. (2015). Emotional Arousal and Discount Rate in
Intertemporal Choice Are Referrence Dependent. Journal of Experimental
Psychology: General, Advance online
puplication.(http://dx.doi.org/10.1037/xge0000047).

108
Levy, D. J., & Glimcher, P. W. (2011). Comparing Apples and Oranges : Using Reward-
Specific and Reward-General Subjective Value Representation in the Brain. The
Journal of Neuroscience, 31(41), 14693–14707.
https://doi.org/10.1523/JNEUROSCI.2218-11.2011
Levy, D. J., Thavikulwat, A. C., & Glimcher, P. W. (2013). State Dependent Valuation :
The Effect of Deprivation on Risk Preferences. PLOS One, 8(1), e53978.
https://doi.org/10.1371/journal.pone.0053978
Li, J., An, R., Zhang, Y., Li, X., & Wang, S. (2012). Correlations of macronutrient-
induced functional magnetic resonance imaging signal changes in human brain
and gut hormone responses. The American Journal of Clinical Nutrition, 96, 275–
282. https://doi.org/10.3945/ajcn.112.037440.1
Loewenstein, G. (1996). Out of Control : Visceral Influences on Behavior.
Organizational Behavior and Human Decision Processes, 65(3), 272–292.
Luo, S., Ainslie, G., Giragosian, L., & Monterosso, J. R. (2009). Behavioral and Neural
Evidence of Incentive Bias for Immediate Rewards Relative to Preference-
Matched Delayed Rewards. Journal of Neuroscience, 29(47), 14820–14827.
https://doi.org/10.1523/JNEUROSCI.4261-09.2009
Luo, S., Ainslie, G., Giragosian, L., & Monterosso, J. R. (2009). Behavioral and Neural
Evidence of Incentive Bias for Immediate Rewards Relative to Preference-
Matched Delayed Rewards, 29(47), 14820–14827.
https://doi.org/10.1523/JNEUROSCI.4261-09.2009

109
Luo, S., Ainslie, G., Giragosian, L., & Monterosso, J. R. (2010). Striatal hyposensitivity
to delayed rewards among cigarette smokers. Drug and Alcohol Dependence.
https://doi.org/10.1016/j.drugalcdep.2010.11.012
Luo, S., Ainslie, G., Giragosian, L., & Monterosso, J. R. (2011). Striatal hyposensitivity
to delayed rewards among cigarette smokers. Drug and Alcohol Dependence,
116(1), 18–23.
Luo, S., Ainslie, G., & Monterosso, J. (2014). The behavioral and neural effect of
emotional primes on intertemporal decisions. Social Cognitive and Affective
Neuroscience, 9, 283–291. https://doi.org/10.1093/scan/nss132
Mayack, C., & Naug, D. (2015). Starving honeybees lose self-control. Biology Letters,
11(20140820).
McClure, S. M., Ericson, K. M., Laibson, D. I., Loewenstein, G., & Cohen, J. D. (2007).
Time discounting for primary rewards. The Journal of Neuroscience, 27(21),
5796–5804.
McClure, S. M., Laibson, D. I., Loewenstein, G., & Cohen, J. D. (2004). Separate Neural
Systems Value Immediate and Delayed Monetary Rewards. Science, 306(5695),
503–507. https://doi.org/10.1126/science.1100907
Mccoy, A. N., & Platt, M. L. (2005). Risk-sensitive neurons in macaque posterior
cingulate cortex. Nature Neuroscience, 8(9), 1220–1227.
https://doi.org/10.1038/nn1523
Mehta, S., Melhorn, S. J., Smeraglio, A., Tyagi, V., Grabowski, T., Schwartz, M. W., &
Schur, E. A. (2012). Regional brain response to visual food cues is a marker of

110
satiety that predicts food choice. The American Journal of Clinical Nutrition, 96,
989–999. https://doi.org/10.3945/ajcn.112.042341.INTRODUCTION
Monterosso, J., & Luo, S. (2010). An Argument Against Dual Valuation System
Competition: Cognitive Capacities Supporting Future Orientation Mediate Rather
Than Compete with Visceral Motivations. Journal of Neuroscience, Psychology,
and Economics, 3(1), 1–14.
Mukherjee, S., & Heithaus, M. R. (2013). Dangerous prey and daring predators : A
review. Biological Reviews, 88, 550–563. https://doi.org/10.1111/brv.12014
Peters, J., & Bu, C. (2011). The neural mechanisms of inter-temporal decision-making:
understanding variability. Trends in Cognitive Sciences, 15(5), 227–239.
https://doi.org/10.1016/j.tics.2011.03.002
Petersen, M. B., Aarøe, L., Jensen, N. H., & Curry, O. (2014). Social Welfare and the
Psychology of Food Sharing : Short-Term Hunger Increases Support for Social
Welfare. Political Psychology, 35(6), 757–773.
https://doi.org/10.1111/pops.12062
Pine, A., Seymour, B., Roiser, J. P., Bossaerts, P., Friston, K. J., Curran, H. V., & Dolan,
R. J. (2009). Encoding of Marginal Utility across Time in the Human Brain. J.
Neurosci., 29(30), 9575–9581. https://doi.org/10.1523/JNEUROSCI.1126-
09.2009
Pine, A., Shiner, T., Seymour, B., & Dolan, R. J. (2010). Dopamine, Time, and
Impulsivity in Humans. Journal of Neuroscience, 30(26), 8888.

111
Plassmann, H., O’Doherty, J., & Rangel, A. (2007). Orbitofrontal Cortex Encodes
Willingness to Pay in Everyday Economic Transactions. The Journal of
Neuroscience, 27(37), 9984–9988.
Pomorska, D. K., & Zubrzycka, M. (2016). Opioid and Cannabinoid System in Food
Intake. Current Pharmaceutical Design, 22, 1361–1370.
Radu, P. T., Yi, R., Bickel, W. K., Gross, J. J., & McClure, S. M. (2011). A mechanism
for reducing delay discounting by altering temporal attention. Journal of the
Experimental Analysis of Behavior, 96(3), 363–385.
Rangel, A., Camerer, C., & Montague, R. (2008). A framework for studying the
neurobiology of value-based decision making. Nature Reviews Neuroscience, 9,
545–556.
Ratcliff, R., Smith, P. L., Brown, S. D., & McKoon, G. (2016). Diffusion Decision
Model: Current Issues and History. Trends in Cognitive Sciences, 20(4), 260–281.
https://doi.org/10.1016/j.tics.2016.01.007
Raynor, H. A., & Epstein, L. H. (2003). The relative-reinforcing value of food under
differing levels of food deprivation and restriction. Appetite, 40(1), 15–24.
https://doi.org/10.1016/S0195-6663(02)00161-7
Read, D., & van Leeuwen, B. (1998). Predicting Hunger: The Effects of Appetite and
Delay on Choice. Organizational Behavior and Human Decision Processes,
76(2), 189–205. https://doi.org/10.1006/obhd.1998.2803
Ridder, D. De, Kroese, F., Adriaanse, M., & Evers, C. (2014). Always Gamble on an
Empty Stomach : Hunger Is Associated with Advantageous Decision Making.

112
PLOS One, 9(10), e111081. doi:10.1371/journal.pone.0111081.
https://doi.org/10.1371/journal.pone.0111081
Rodriguez, C. A., Turner, B. M., Van Zandt, T., & McClure, S. M. (2015). The neural
basis of value accumulation in intertemporal choice. European Journal of
Neuroscience, 42(5), 2179–2189. https://doi.org/10.1111/ejn.12997
Rolls, E. T. (2016). Reward Systems in the Brain and Nutrition. Annual Review of
Nutrition, 36(1), 435–470. https://doi.org/10.1146/annurev-nutr-071715-050725
Schwartz, M. W., Peskind, E., Raskind, M., Boyko, E. J., & Porte Jr., D. (1996).
Cerebrospinal fluid leptin levels: Relationship to plasma levels and to adiposity in
humans. Nature Medicine, 2(5), 589–593.
Seyyed, F. J., Abraham, A., & Al-hajji, M. (2005). Seasonality in stock returns and
volatility : The Ramadan effect. Research in International Business and Finance,
19, 374–383. https://doi.org/10.1016/j.ribaf.2004.12.010
Shah, B. P., Vong, L., Olson, D. P., Koda, S., Krashes, M. J., & Ye, C. (2014). MC4R-
expressing glutamatergic neurons in the paraventricular hypothalamus regulate
feeding and are synaptically connected to the parabrachial nucleus. PNAS,
111(36), 13193–13198. https://doi.org/10.1073/pnas.1407843111
Simon, J. J., Skunde, M., Wu, M., Schnell, K., Herpertz, S. C., Bendszus, M., …
Friederich, H. (2015). Neural dissociation of food- and money-related reward
processing using an abstract incentive delay task, 2.
https://doi.org/10.1093/scan/nsu162

113
Skrynka, J., & Vincent, B. T. (2017). Subjective hunger, not blood glucose, influences
domain general time preference. PsyArXiv.
https://doi.org/10.17605/OSF.IO/QGP54
St-Onge, M.-P., Roberts, A., Shechter, A., & Choudhury, A. R. (2016). Fiber and
Saturated Fat Are Associated with Sleep Arousals and Slow Wave Sleep. Journal
of Clinical Sleep Medicine : JCSM : Official Publication of the American
Academy of Sleep Medicine, 12(1), 19–24. https://doi.org/10.5664/jcsm.5384
Symmonds, M., Emmanuel, J. J., Drew, M. E., Batterham, R. L., & Dolan, R. J. (2010).
Metabolic State Alters Economic Decision Making under Risk in Humans. PLOS
One, 5(6), 1–7. https://doi.org/10.1371/journal.pone.0011090
Tal, A., & Wansink, B. (2013). Fattening Fasting: Hungry Grocery Shoppers Buy More
Calories, Not More Food. JAMA Internal Medicine, 173(12), 1146–1148.
Tataranni, P. A., Gautier, J.-F., Chen, K., Uecker, A., Bandy, D., Salbe, A. D., …
Ravussin, E. (1999). Neuroanatomical correlates of hunger and satiation in
humans using positron emission tomography. PNAS, 96, 4569–4574.
Tversky, A., & Kahneman, D. (1986). Rational Choice and the Framing of Decisions.
The Journal of Business, 59(4), S251–S278.
Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative
representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297–323.
Uher, R., Treasure, J., Heining, M., Brammer, M. J., & Campbell, I. C. (2006). Cerebral
processing of food-related stimuli : Effects of fasting and gender. Behavioural
Brain Research, 169, 111–119. https://doi.org/10.1016/j.bbr.2005.12.008

114
Vaca, S. C. de, & Carr, K. D. (1998). Food Restriction Enhances the Central Rewarding
Effect of Abused Drugs. Journal of Neuroscience, 18(18), 7502–7510.
Voigt, J., & Fink, H. (2015). Serotonin controlling feeding and satiety. Behavioural Brain
Research, 277, 14–31. https://doi.org/10.1016/j.bbr.2014.08.065
Voskuilen, C., Ratcliff, R., & Smith, P. L. (2016). Comparing fixed and collapsing
boundary versions of the diffusion model. Journal of Mathematical Psychology,
73(Supplement C), 59–79. https://doi.org/10.1016/j.jmp.2016.04.008
Wang, Q., Luo, S., Monterosso, J., Zhang, J., Fang, X., Dong, Q., & Xue, G. (2014).
Distributed Value Representation in the Medial Prefrontal Cortex during
Intertemporal Choices. The Journal of Neuroscience, 34(22), 7522–7530.
Wang, X. T., & Dvorak, R. D. (2010). Sweet Future: Fluctuating Blood Glucose Levels
Affect Future Discounting. Psychological Science.
https://doi.org/10.1177/0956797609358096
Wansink, B., Tal, A., & Shimizu, M. (2012). First Foods Most: After 18-Hour Fast,
People Drawn to Starches First and Vegetables Last. Archives of Internal
Medicine, 172(12), 961–963.
Ward, A., & Mann, T. (2000). Don’t Mind If I Do: Disinhibited Eating Under Cognitive
Load. Journal of Personality and Social Psychology, 78(4), 753–763.
Weller, R., Cook, E., Avsar, K. B., & Cox, J. E. (2008). Obese women show greater
delay discounting than healthy-weight women. Appetite, 51, 563–569.
https://doi.org/10.1016/j.appet.2008.04.010

115
Wiecki, T. V., Sofer, I., & Frank, M. J. (2013). HDDM: Hierarchical Bayesian estimation
of the Drift-Diffusion Model in Python. Frontiers in Neuroinformatics, 7.
https://doi.org/10.3389/fninf.2013.00014
Wierenga, C. E., Bischoff-grethe, A., Melrose, A. J., Irvine, Z., Torres, L., Bailer, U. F.,
… Kaye, W. H. (2015). Archival Report Hunger Does Not Motivate Reward in
Women Remitted from Anorexia Nervosa. Biological Psychiatry, 77(7), 642–652.
https://doi.org/10.1016/j.biopsych.2014.09.024
Winkler, A. M., Ridgway, G. R., Webster, M. A., Smith, S. M., & Nichols, T. E. (2014).
Permutation inference for the general linear model. NeuroImage, 92(Supplement
C), 381–397. https://doi.org/10.1016/j.neuroimage.2014.01.060
Xu, A. J., Schwarz, N., & Wyer, R. S. (2015). Hunger promotes acquisition of nonfood
objects. PNAS, 112(9), 2688–2692. https://doi.org/10.1073/pnas.1417712112

116
Appendices:

Appendix 1: fMRI Preprocessing Workflow

117
Appendix 2: MCMC trace plots

Appendix 2.1.1 Hunger Ratings MCMC

118
Appendix 2.1.2 Fullness Ratings MCMC

119
Appendix 2.2.1 Hard vs. Easy ITC RT MCMC

120
Appendix 2.2.2 Fast vs. Fed ITC RT MCMC

121
Appendix 2.3.1 ITC HDDM GAT Model MCMC

122

123

124
Appendix 2.3.2 HDDM Posterior Distributions of Simulated Dataset for GAT Model

125

126
Appendix 2.4 MID RT MCMC

127

128
Appendix 2.5 MID RT Expanded MCMC

129

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Sequential decisions on time preference: evidence for non-independence

PDF

Behavioral and neural evidence of incentive bias for immediate rewards relative to preference-matched delayed rewards

PDF

Behabioral and neural evidence of state-like variance in intertemporal decisions

PDF

Visual and audio priming of emotional stimuli and their relationship to intertemporal preference shifts

PDF

Value-based decision-making in complex choice: brain regions involved and implications of age

PDF

Validation of a neuroimaging task to investigate decisions involving visceral immediate rewards

PDF

Reward immediacy and subjective stress modulate anticipation of primary and secondary rewards in temporarily-abstinent cigarette smokers

PDF

The acute impact of glucose and sucralose on food decisions and brain responses to visual food cues

PDF

Brain and behavior correlates of intrinsic motivation and skill learning

PDF

Reward substitution: how consumers can be incentivized to choose smaller food portions

PDF

The evolution of decision-making quality over the life cycle: evidence from behavioral and neuroeconomic experiments with different age groups

PDF

Testosterone’s effect on reward, cognition, and myelination: rat brain and behavior

PDF

Heart, brain, and breath: studies on the neuromodulation of interoceptive systems

PDF

Insula activity during safe-sex decision-making in sexually risky men suggests negative urgency and fear of rejection drives risky sexual behavior

PDF

Effect of mindfulness training on attention, emotion-regulation and risk-taking in adolescence

PDF

Biological and behavioral correlates of emotional flexibility and associations with exposure to family aggression

PDF

Investigating the effects of Pavlovian cues on social behavior

PDF

Mechanisms of stress effects on learning and decision making in younger and older adults

PDF

Noradrenergic mechanisms of arousal-enhanced memory selectivity

PDF

The effect of present moment awareness and value intervention of ACT on impulsive decision-making and impulsive disinhibition

Asset Metadata

Creator Melrose, Andrew James (author)

Core Title Homeostatic imbalance and monetary delay discounting: effects of feeding on RT, choice, and brain response

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 04/04/2018

Defense Date 12/14/2017

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

Tag cognitive neuroscience,cognitive science,fMRI,Hunger,neuroeconomics,OAI-PMH Harvest,reward

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Monterosso, John Robert (committee chair), Bechara, Antoine (committee member), Brocas, Isabelle (committee member), John, Richard (committee member), Page, Kathleen Alanna (committee member)

Creator Email amelrose@usc.edu,andrewjamesmelrose@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-490367

Unique identifier UC11267178

Identifier etd-MelroseAnd-6143.pdf (filename),usctheses-c40-490367 (legacy record id)

Legacy Identifier etd-MelroseAnd-6143.pdf

Dmrecord 490367

Document Type Dissertation

Rights Melrose, Andrew James

Type texts

Source 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 a...

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