Content uploaded by Richard Huskey
Author content
All content in this area was uploaded by Richard Huskey on Jun 13, 2023
Content may be subject to copyright.
Computationally modeling mood management theory:
a drift-diffusion model of people’s preferential choice
for valence and arousal in media
Xuanjun Gong
1,2
, Richard Huskey
1,3,
*, Allison Eden
4
Ezgi Ulusoy
4
1
Department of Communication, Cognitive Communication Science Lab, University of California Davis, CA, USA
2
Department of Statistics, University of California Davis, CA, USA
3
Center for Mind and Brain, University of California Davis, CA, USA
4
Department of Communication, Michigan State University, MI, USA
*Corresponding author: Richard Huskey; E-mail: rwhuskey@ucdavis.edu
Abstract
Mood management theory (MMT) hypothesizes that people select entertainment content to maintain affective homeostasis. However, this
hypothesis lacks a formal quantification of each affective attributes’ separate impact on an individual’s media content selection, as well as an
integrated cognitive mechanism explaining media selection. Here we present a computational decision-making model that mathematically
formalizes this affective media decision-making process. We empirically tested this formalization with the drift-diffusion model using three
decision-making experiments. Contrary to MMT, all three studies showed that people prefer negatively valenced and high-arousal media content
and that prevailing mood does not shape media selection as predicted by MMT. We also discovered that people are less cautious when choices
have larger valence differences. Our results support the proposed mathematical formalization of affective attributes’ influence on media
selection, challenge core predictions drawn from MMT, and introduce a new mechanism (response caution) for media selection.
Keywords: mood management theory, drift-diffusion model, computational modeling, open science
Why and how we select specific media has been a central ques-
tion for media researchers since the 1940s. One of the prevailing
theories explaining entertainment media selection is mood man-
agement theory (MMT; Zillmann, 1988).
1
MMT suggests that
people’s selective exposure to media content is determined by
the media users’ prevailing mood and the affective properties of
media content, such that people choose media content that will
help them maintain a moderate level of arousal (avoiding hypo-
or hyper-arousal) and will change their prevailing mood state to-
wards a positive direction. However, MMT research has mainly
focused on behavioral tests of media choice or the effects of
choice on mood (Carpentier, 2020;Reinecke, 2016); and
empirical MMT research shows mixed support for key proposi-
tions of the theory (e.g., Knobloch-Westerwick, 2014;
Strizhakova & Krcmar, 2007). Thus, despite sustained effort, it
remains unclear how users’ affective states influence their selec-
tion of specific media content. In fact, in our review of the litera-
ture, MMT only meets a few of the established criteria for
evaluating a theory in communication (DeAndrea and Holbert,
2017). It is parsimonious, heuristically generative, and provides
a clear organizing scheme in terms of selective exposure to media
along specific dimensions. However, to date, the extent to which
MMT offers explanatory power, falsifiability, and internal con-
sistency is not well reflected in the literature.
We propose that computational modeling provides clarity
into media selection processes and demonstrate how this ap-
proach can be applied to test hypotheses derived from MMT
by making sense of observed behavioral media choice and re-
sponse time (RT) data with precise mathematical models
(Wilson & Collins, 2019). In three studies, we apply a
computational decision-making model, the drift-diffusion
model (DDM; Ratcliff and McKoon, 2008), to a two-choice
(dichotomous) media selection task in order to investigate
whether and how people’s entertainment media choices are
influenced by their prevailing mood and affective attributes of
media. When using the DDM, people’s value-based decision
making, such as choosing a movie (e.g., comedy, drama), is
assumed to be a continuous noisy preferential evidence accu-
mulation process that drifts toward one of two possible deci-
sion boundaries (e.g., comedy or drama). Said differently, the
DDM accounts for these observable media selection outcomes
using choice and RT data. The DDM helps us better under-
stand how affect influences people’s media selection and con-
tributes to theory by providing empirical evidence for or
against the person- and content-specific mechanisms that in-
fluence media selection. Our study shows how to utilize such
models to formalize and test a verbal theory, and demon-
strates the theoretical value (DeAndrea & Holbert, 2017)of
the approach by aiding in falsification, identification of
boundary conditions, and identifying new mechanisms in
existing communication theories.
Mood management theory
MMT, or the theory of affect-dependent stimulus arrange-
ment (Zillmann, 1988), suggests that individuals are moti-
vated to terminate noxious affective states by arranging their
media environments in a way to “perpetuate and increase the
intensity of gratifying, pleasurable excitable states” (p. 158).
These arrangements can be grouped to form four primary hy-
potheses: (a) persons in aversive states will prefer hedonically
Received: 28 August 2022. Revised: 3 May 2023. Accepted: 3 May 2023
V
CThe Author(s) 2023. Published by Oxford University Press on behalf of
International Communication Association. All rights reserved. For permissions, please email: journals.permissions@oup.com
Journal of Communication,2023, 00, 1–18
https://doi.org/10.1093/joc/jqad020
Original Article
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
positive stimuli (valence hypothesis), (b) persons in states of
extreme over- or under stimulation will act to return to a
baseline (excitatory homeostasis), (c) persons in a noxious
mood state will display a preference for the most absorbing
types of stimulation to intervene in a prevailing mood state
(intervention potential), and (d) persons will select stimuli
with minimal behavioral affinity with their experiential state
(semantic affinity). DDM is particularly well suited for
addressing MMT’s first two hypotheses.
2
Therefore, in what
follows, we explicate linkages between DDM and MMT with
regard to MMT’s valence and excitatory homeostasis
hypotheses.
The drift-diffusion model of decision making
The DDM (Ratcliff and McKoon, 2008) explains people’s
preferential choice and response time in speeded two-choice
decision tasks. DDM suggests that a person’s decision is the
result of an evidence-accumulating process, where decision
makers collect preferential evidence in favor of one option or
another alternative, as a function of weighted attributes of the
options (Figure 1A). The options in a two-choice decision
task are represented as an upper decision boundary (A; the
correct or high subjective value option) and a lower decision
boundary (0; the error or low subjective value option). When
the accumulation of evidence reaches either one of the choice
boundaries, the decision-making process is complete. The
drift process begins at a starting point (Z) in between the two
choice boundaries which represents predecision bias. After ac-
counting for non-decision time (T), the decision maker accu-
mulates evidence or information that leads to a noisy drift
process (v) toward the preferred choice boundary, following a
signal detection model.
Computational parameters in the DDM have unique con-
ceptual operationalizations. Specifically, non-decision time
(T), accounts for perceptual processing and executing the de-
cision; decision boundaries (A, 0), account for decision cau-
tiousness, with wider boundaries representing more cautious
decision making; decision starting point (Z), accounts for bi-
ased preferences before encountering the decision options;
and decision drift rate (v), accounts for the rate of evidence
accumulation or the subjective value difference between
options. The decision drift rate sign indicates choice
Figure 1 (A) The DDM. Choices in a decision task are represented as upper (A) and lower (0) boundaries. The drift process starts at a biased point (Z). Drift
rate (v) follows a noisy drift toward the preferred choice. Non-decision time (T) is also estimated. The upper and lower panels show RT distributions. (B)
Experimental procedure. For all studies, participants were shown movie texts and asked to rate the valence and arousal of the movie texts (training
phase). In studies two and three only, participants were randomly assigned into one of four experimental mood induction (image viewing) conditions
which included an induction check (prevailing mood rating was measured using the SAM). Lastly, for all studies, participants completed the decision
tasks.
2A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
preference. Positive signs indicate a preference for the high-
value (or correct) choice, negative signs indicate a preference
for the low-value (or error) choice.
The DDM has been widely applied (Ratcliff et al., 2016)
with emerging research focused on value-based decision mak-
ing. Value-based decisions are when people make decisions
based on a comparison of the subjective values of the choices,
such as deciding what to eat (Krajbich et al., 2015), purchase
(Krajbich et al., 2012), or who to socialize with (Krajbich
et al., 2015).
During value-based decision making, decision makers con-
struct object representations featuring attributes of each
choice (Rangel et al., 2008). People then evaluate the subjec-
tive value for each attribute of the representations (Krajbich
et al., 2010;Milosavljevic et al., 2010). Subjective value refers
to the positive or negative value an individual assigns to an at-
tribute, which is then weighted by how important that attrib-
ute is to the individual. The integrated subjective values for
each choice are then compared. People’s value-based decision
making is mainly driven by subjective value differences be-
tween the choices, regardless of each choice’s absolute value
(Tajima et al., 2016). Therefore, the decision process takes
longer for choices with small subjective value differences com-
pared to choices with large subjective value differences. In
other words, choosing between two apples should take longer
than choosing between an apple and an orange.
Combining the DDM and MMT
Given that MMT hypothesizes that affective attributes of me-
dia content influence the expected subjective value of a media
choice, media selection in the service of mood management
can be considered a value-based decision-making task.
Treating media selection as a value-based decision-making
task is in line with literature suggesting that affect is informa-
tion integrated into subjective value calculations for decision
making (Greifeneder et al., 2011; Hartley et al., 2018;
Roberts & Hutcherson, 2019;Schwarz, 2012;Shulman &
Bullock, 2019). This perspective is also consistent with media
scholars such as Knobloch-Westerwick (2014), who suggest
that mood management via media can be fit to an
expectancy-value framework, meaning that a linear relation-
ship between affective attributes of media content and
expected subjective value may ultimately determine media se-
lection. Thus, we propose that media selection can be mod-
eled as a multi-attribute value-based drift-diffusion process.
What does this mean? When given a two-choice decision
task (choosing content A or content B), the direction and
speed of evidence accumulation (which is directly related to
drift rate) will be based on the subjective value difference be-
tween the media choices. Within a two-choice decision task,
the affective attributes of choice A and B vary, as does the im-
portance of these affective attributes for each individual.
Jointly, and as discussed above, the subjective value difference
between these attributes contributes to the choice. Two
choices that have very similar attributes will have a small sub-
jective value difference whereas two choices that have very
dissimilar attributes will have a large subjective value differ-
ence. When subjective value differences are small, drift rate is
low and selection approximates chance such that (a) both
choices are equally likely to be selected and (b) RTs for such a
choice would be slow. When the subjective value difference
between two choices is high, (a) drift rate is high and (b) a
dominant selection preference for the higher-value choice
emerges along with (c) fast RTs. Moreover, when option A
has a higher subjective value than option B, the subjective
value difference (A-B) will be positive and the drift rate will
be positive, thus option A will be preferred over option B. If
option B has a higher subjective value, the subjective value
difference will be negative and the drift rate will be negative,
thus option B will be preferred over option A.
Using the MMT framework, the subjective value of enter-
tainment media should be determined by the affective attrib-
utes of the media content and each individual’s mood and
content preferences, which function as weights. For a person
who is in a negative mood, positive content will have a higher
potential gain, which will increase the likelihood of selecting
positive content. Whereas, for a person who is in a positive
mood, the potential gain from positive content might not be
as strong as the person in a negative mood. Hence, following
Roberts and Hutcherson (2019), we propose that affective
attributes of media content contribute to the expected subjec-
tive value of a choice in a weighted linear way:
SVchoice ¼XaWaAttributeaEquation 1
In Equation 1,SV represents the expected subjective value
of a choice (media content), Attribute represents an affective
attribute of the media content, and Wrepresents the attribute
weight. Consistent with MMT, we focus on valence and
arousal as affective attributes of media content, and propose
that:
SVchoice ¼Warousal Arousal þWvalence Valence
Equation 2
To test this multi-attribute value-based drift diffusion
model (Busemeyer et al., 2019), we need a series of two-
choice decision tasks that carefully differentiate the effects
from either one of the affective attributes (i.e., valence,
arousal) or both. We expect that drift rate (v) is a function of
the value difference between two choices:
v¼SVchoiceA SVchoiceB Equation 3
Accordingly, we developed a choice set that included four
categories of movie stimuli (arousal high/low xvalence posi-
tive/negative). Combining the movie stimuli from these four
categories gives 10 possible decision types, which are reduc-
ible to five mutually exclusive and completely exhaustive
types of decision tasks (Supplementary Section 1). We derive
falsifiable hypotheses from MMT and test them by comparing
the DDM parameters estimated from these five decision tasks.
In the following sections, we report the rationale, hypotheses,
and results for three studies that use the DDM to test core pre-
dictions derived from MMT.
Open science practices
Following calls for open science practices in communication
(Dienlin et al., 2021;Lewis, 2020), our hypotheses were pre-
registered (https://osf.io/tb3ca/). The code for stimulus genera-
tion, the stimuli themselves, the PsychoJS code required to
run each experiment, the raw data, and the Python and R
code for data analysis is posted on GitHub (https://github.
com/cogcommscience-lab/movie_selection).
Journal of Communication (2023) 3
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Study one
A core prediction of MMT is that people have a preference
for positively valenced media content. Therefore, we expect a
positively signed drift rate for valence such that: (H1a) The
group-level drift rate in a decision task with only a valence
difference (1V0A) is positive (above 0), and (H1b) the group-
level drift rate in a decision task with only a valence difference
(1V0A) is higher than the decision task with no valence and
no arousal difference (0V0A). Moreover, based on MMT, we
expect that drift rate, determined by movie valence, is also
shaped by people’s current mood valence, meaning that indi-
viduals in a negatively valenced mood will have a higher drift
rate (preference) toward positively valenced movie options.
Therefore, (H2) the individual-level drift rate in a decision
task with only a valence difference (1V0A) is negatively corre-
lated with people’s prevailing mood valence.
The excitatory homeostasis prediction of MMT states that
media preferences are determined by prevailing arousal, mean-
ing that under-stimulated individuals would have a positively
signed drift rate (preference) toward a high-arousal movie op-
tion, while over-stimulated individuals would have a negatively
signed drift rate (avoidance) toward a high-arousal movie op-
tion. Accordingly, we expect that (H3) the individual-level drift
rate in a decision task with only an arousal difference (0V1A) is
negatively correlated with people’s prevailing excitatory state.
Additionally, we ask exploratory questions to determine if
people have a basic preferential tendency toward high or low-
arousal movies: (RQ1a) Is the group-level drift rate in a deci-
sion task with only an arousal difference (0V1A) negative
(preference for low-arousal), positive (preference for high
arousal), or not distinct from 0 (no clear preference) and
(RQ1b) is the group-level drift rate in a decision task with
only an arousal difference (0V1A) negative (preference for
low-arousal), positive (preference for high arousal), or not
distinct from the decision task with no valence and no arousal
difference (0V0A)?
Moreover, based on the proposed additive linear model
(Equation 2) combining both arousal and valence attributes
as integral parts of subjective value in determining movie
choice, we expect an additive effect of arousal and valence on
drift rate in decision tasks that differ on both arousal and va-
lence. This hypothesis is based on the logic described above
demonstrating that, as the subjective value difference
(Equation 3) between two choices increases, drift rates be-
come faster (high-drift rate). Therefore, (H4a) the decision
task with both valence and arousal differences in the same di-
rection (1V1A) will have a higher group-level drift rate com-
pared with drift rate in decision tasks involving only either
valence or arousal (1V0A or 0V1A).
Similarly, when valence and arousal are integrated in an op-
posite direction (similar to the value of valence minus the
value of arousal), the integrated value becomes smaller than
the value of valence or arousal alone. Therefore, (H4b) the de-
cision task with both valence and arousal differences in the
opposite direction (1V-1A) will have a lower group-level drift
rate compared with drift rate in decision tasks involving only
either valence or arousal (1V0A or 0V1A).
Method
Stimulus generation
Single-sentence movie summaries were generated using natural-
language processing techniques to shorten paragraph-length plot
descriptions from 42,306 films (Bamman et al., 2013). A
dictionary-based approach was then used to label each movie
summary on two dimensions, arousal and valence (Bradley &
Lang 1999;Warriner et al. 2013). Following best-practices for
using dictionary-based approaches (Song et al., 2020), the
arousal and valence labels were then cross validated using self-
assessment manikin (SAM; Bradley & Lang, 1994) ratings from
human annotators (n¼164). This resulted in a final dataset of
56 stimuli that systematically varied on arousal (high/low) and
valence (positive/negative; Supplementary Section 1).
Decision-making task
This experiment was conducted on https://pavlovia.org/.At
the beginning of the experiment, participants self-reported
their prevailing mood state using the SAM (Bradley & Lang,
1994), which captures feelings of valence and arousal. Then
participants rated eight movie summaries using the SAM
(Figure 1B) as a training block in order to familiarize partici-
pants with each of the summaries. Next, in a testing block,
participants were presented with a series of two-choice deci-
sion tasks consisting of the movie summaries shown in the
previous training block (20 trials, four trials for each decision
type). The decision trials were randomly and independently
generated for each participant. Participants chose their pre-
ferred option as quickly as possible. Choice and RT were
recorded (Supplementary Section 2). This training-then-
testing procedure was repeated for a total of seven training
and testing blocks. Lastly, participants provided demographic
and media preference information.
Analysis
To test our hypothesis and questions, we applied a hierarchi-
cal Bayesian drift-diffusion model (HDDM; Gong & Huskey,
in press; Wiecki et al., 2013) to the choice and RT data
(Supplementary Sections 3 and 4). To aid in interpretation,
we fixed the upper boundary to be the choice with higher
arousal or valence, and lower boundary to be the choice with
lower arousal or valence (as determined in the stimulus gener-
ation pretest). This means that a positively signed drift rate
indicates a preference for high arousal or positively valenced
movie choices, whereas a negatively signed drift rate indicates
a preference for low arousal or negatively valenced movie
choices. HDDM generates a posterior probability distribution
for both group- and individual-level drift rate parameters for
the decision process. We applied three common procedures to
deal with contamination RTs (Supplementary Section 5).
Inferential testing was conducted on the posterior probability
distributions using Bayesian logic.
Sample
Participants were n¼140 undergraduates from the University
of California Davis. Sample characteristics are reported in
Supplementary Section 6 and preregistered exclusion criteria
are discussed in Supplementary Section 7. This sample size
was based on a power analysis of a regression model (p<
.05; 90% power; d¼0.1) conducted using GPower (Faul
et al., 2007). This estimated sample (n¼140) was then used
in a subsequent and more precise power analysis which simu-
lated 100 datasets, each with 140 subjects, with varying drift
rates. A HDDM was estimated for each dataset. Results indi-
cate that a sample size of n¼140 provides very high power
(94% power; d¼0.1). Complete code for conducting the
4A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
simulations and calculating power can be found on the proj-
ect’s GitHub repository.
Results
Manipulation checks show that our movie summaries system-
atically varied along arousal and valence dimensions as
expected (Supplementary Section 8). We report the mean and
95% credible interval for each posterior probability distribu-
tion. Bayesian inference testing is done by comparing one pos-
terior probability distribution against another (as specified by
our hypotheses)—resulting in a distribution of differences—
which is compared against zero (Kruschke, 2013). For hy-
potheses where a comparison between posterior probability
distributions is not specified, we test a single posterior proba-
bility distribution against zero. The percentage of posterior
greater than 0 and less than 0 is reported. For example,
100.0% <0<0.0% indicates that 100.0% of posteriors are
smaller than 0 and 0.0% of posteriors are larger than 0.
Results where 97.5% of the distribution is >0 (or <0) are
considered credible.
3
Valence hypotheses
For tasks that differed only in the valence of the stimuli
(1V0A), we expected a positively signed drift rate that is credi-
bly different from zero (H1a) and higher drift rate compared
to tasks with no valence or arousal difference (0V0A; H1b).
These hypotheses were not supported (Figure 2A, left).
Instead, the drift rate for higher valence movies (M¼0.162,
95% CI [0.212, 0.111]) is negative, credibly different
from zero (100.0% <0<0.0%), and credibly lower (100.0%
<0<0.0%) than the 0V0A drift rate (M¼0.004, 95% CI
[0.053, 0.045]). This means that participants have a prefer-
ence for negatively valenced movies. The 0V0A drift rate was
centered about zero, which indicates that participants’
responses to these tasks were essentially at chance; that is,
participants were equally likely to choose either option during
0V0A tasks in which the two options had no difference in va-
lence and arousal (as would be expected within a value-based
decision-making framework).
We also expected that, for tasks differing only in valence
(1V0A), we would see a negative correlation between prevail-
ing valence mood state and drift rate (H2). We found that the
prevailing valence mood state has no credible relationship
with drift rate (M¼0.010, 95% CI [0.041, 0.060], 35.7%
<0<64.3%). Thus, H2 did not have credible evidence
(Figure 2A, right).
Arousal hypotheses
H3 specified that the drift rate in tasks with only arousal dif-
ferences (0V1A) will be negatively correlated with people’s
prevailing mood arousal. There was no credible evidence for
this hypothesis (Figure 2A, right; M¼0.004, 95% CI
[0.023, 0.031], 39.2% <0<60.8%). RQ1a asked if partic-
ipants had a preference for high-arousal movies in tasks with
only arousal differences (0V1A). Our results indicate that the
drift rate in tasks with only arousal differences (0V1A;
M¼0.014, 95% CI [0.034, 0.064]) is not credibly different
from zero (28.9% <0<71.1%). Finally, RQ1b asked if par-
ticipants had a stronger preference for tasks that differed in
arousal only (0V1A) relative to tasks that did not differ on ei-
ther attribute (0V0A). We found no credible difference
(30.3% <0<69.7%) between the drift rates. Together, these
results show that participants have no clear preference for
arousal attributes of movies in the decision task.
Multi-attribute (valence and arousal) hypotheses
H4a and H4b specified an additive influence of valence and
arousal on drift rates. Thus, we compared drift rate of tasks
differing in both arousal and valence (1V1A & 1V-1A) to
drift rate of tasks differing in only arousal or valence (0V1A,
1V0A). Our results showed that the posterior probability dis-
tributions for drift rates are signed differently (0V1A is posi-
tive; 1V0A, 1V1A & 1V-1A are negative). Therefore, we
tested H4a and H4b by evaluating the magnitude difference
between the drift rates by converting the estimated drift rates
of 1V0A, 1V1A, and 1V-1A to be positive before the compar-
ison. For H4a, the drift rate magnitude in tasks with both va-
lence and arousal differences (1V1A) is credibly higher than
drift rate magnitude in task with only arousal difference
(0V1A; 0.0% <0<100.0%), but not credibly higher than in
tasks with only valence differences (1V0A; 22.8% <
0<77.2%). Similarly, for H4b, the results indicate that drift
rate magnitude in tasks with both valence and arousal differ-
ences in the opposite direction (1V-1A) is credibly higher than
drift rate magnitude with only arousal differences (0V1A;
0.0% <0<100.0%), but not credibly higher than in tasks
with only valence differences (1V0A; 42.6% <0<57.4%).
Thus, both H4a and H4b were partially credible.
Finally, we note that the value of drift rate posterior distri-
bution is 1V-1A <1V0A <1V1A <0V0A <0V1A
(Figure 2A, left). This order partially supports our hypothesis
that the effect of valence and arousal on drift rates is additive,
although primarily negatively signed.
Exploratory analyses
These findings provide no evidence for our hypotheses drawn
from MMT. In fact, several results directly, and credibly, con-
tradict core MMT valence and arousal predictions. Were we
using solely experimental data, we would at this point be at a
loss. However, using a computational model allows us to
probe these contradictions. First, we broadened our analytical
approach to a general linear model framework where we can
estimate DDM model parameters by treating attributes
(arousal, valence) as factors in a regression model. Second,
the DDM has three additional parameters in addition to drift
rate: non-decision time, decision boundary, and decision bias.
Our exploratory analyses investigated these parameters in a
full DDM model.
In order to estimate the effects of valence and arousal on
the full DDM, we fit a regression model on the decision tasks
(excluding the 1V-1A decision task) with valence and arousal
as factors in a 2 (valence high/low) x2 (arousal positive/nega-
tive) design. Factors were dummy-coded based on the stimu-
lus generation pretest (low/negative ¼0, high/positive ¼1).
Importantly, DDM is always comparing choices (i.e., choiceA
vs. choiceB). Therefore, these factors were then subtracted
from eachother such that 0 ¼no difference between choices
and 1 ¼difference between choices). Accordingly, the param-
eter estimate represents the magnitude and directionality of
the higher factor-level (i.e., difference between choices) rela-
tive to the lower factor-level (i.e., no difference between
choices). We then tested whether the estimated posterior dis-
tributions for the main effects (for both valence and arousal)
and the interaction effect were different from zero for each
Journal of Communication (2023) 5
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Figure 2 (A) Confirmatory analysis, study 1. The left plot shows the posterior probability distributions for drift rates in decision tasks with distinct valence/
arousal differences. The right plot shows the posterior distributions for the effects of mood valence and mood arousal on drift rate. (B) Exploratory
analysis, study 1. Posterior probability distribution of the effects of valence, arousal and their interaction on drift rate (upper left), decision boundary (upper
right), non-decision time (bottom left) and decision bias (bottom right) in the full DDM model.
6A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
DDM model parameter (drift rate, decision boundary, deci-
sion bias, and non-decision time).
The results (Figure 2B) suggest that (a) valence has a credi-
ble negative main effect on drift rate (M¼-0.144, 95% CI
[0.183, 0.102], 100.0% <0<0.0%) and decision bound-
ary (M¼0.084, 95% CI [0.158, 0.015], 98.8% <
0<1.2%), but no credible effect on non-decision time
(M¼0.001, 95% CI [0.025, 0.026], 53.2% <0<46.8%)
or decision bias (M¼0.010, 95% CI [0.028, 0.008],
85.4% <0<14.6%), (b) arousal has a credible negative
main effect on decision boundary (M¼0.087, 95% CI
[0.159, 0.020], 99.8% <0<0.2%), but no credible effect
on drift rate (M¼0.037, 95%CI [0.002, 0.076], 3.4% <
0<96.6%), non-decision time (M¼0.010, 95% CI [0.014,
0.035], 21.7% <0<78.3%), or decision bias (M¼0.017,
95% CI [0.034, 0.000], 96.6% <0<3.4%), (c) the interac-
tion between arousal and valence does not have credible effect
on drift rate (M¼
0.031, 95% CI [0.088, 0.022], 87.2%
<0<12.8%), decision boundary (M¼0.046, 95%CI
[0.042, 0.142], 16.5% <0<83.5%), decision bias
(M¼0.020, 95% CI [0.005, 0.045], 5.2% <0<94.8%) or
non-decision time (M¼0.008, 95% CI [0.027, 0.039],
31.7.0% <0<68.3%). In sum, valence has a credible main
effect on drift rate, and both arousal and valence have credi-
ble main effects on boundary.
Discussion
Our preregistered results for study one largely do not support
our main hypotheses. First, we failed to find evidence for the
MMT hypothesis that people have a preference toward posi-
tively valenced movies. Instead, we found credible evidence
that people prefer negatively valenced movies. This finding is
also confirmed in our exploratory analysis. Second, we did
not detect a preference for arousal. Third, we found a credible
drift rate difference when valence and arousal effects are addi-
tive (1V1A & 1V-1A) compared to tasks with only arousal
differences (0V1A), but not to tasks with only valence differ-
ences (1V0A). We also found the order of the drift rate poste-
rior is largely consistent with the additive model. Finally, we
failed to find the effect of people’s prevailing mood on peo-
ple’s preference for movies with different valence and arousal.
In sum, these findings, particularly related to valence, chal-
lenge hypotheses drawn from MMT. However, MMT hy-
potheses are concerned with the effect of existing mood on
media selection, and unlike many MMT studies, we did not
include a mood induction in study one. As such, we con-
ducted a follow-up study that experimentally manipulated
mood state.
Study two
Using the MMT logic explicated above, we expected that in-
duced mood (valence and arousal) would interact with peo-
ple’s preference for the corresponding affective attributes
(valence and arousal) of movie choices when people are evalu-
ating the subjective value of movies. We hypothesized an in-
teraction effect for mood valence such that: (H1a) individuals
in the negative-mood valence condition will have a positively
signed drift rate toward positive-valence movies (upper
boundary) and (H1b) individuals in the positive-mood va-
lence condition will have a negatively signed drift rate toward
negative-valence movies (lower boundary).
Similarly, we also hypothesized an interaction effect for
arousal such that: (H2a) participants in the low-mood arousal
condition will have a positively signed drift rate towards high-
arousal movies (upper boundary) and (H2b) participants in
the high-mood arousal condition will have a negatively signed
drift rate towards low-arousal movies (lower boundary). We
also asked if: (RQ1) The MovieValence MoodValence or
MovieArousal MoodArousal interaction more strongly
influences drift rate.
Finally, in study one, for arousal and valence, we saw nega-
tively signed parameter estimates for decision boundary.
What does this mean? Decision boundary is often interpreted
as a measure of response caution (Ratcliff et al., 2016;
Roberts & Hutcherson, 2019). Wider decision boundaries are
associated with lower (slower) drift rates, more cautious deci-
sions, and in instances where there is an objectively correct
decision, decreased error rates. By comparison, narrower de-
cision boundaries are associated with higher (faster) drift
rates, less cautious decisions, and increased error rates. That
the main effects for arousal and valence are credible and nega-
tively signed tells us that, when the subjective value difference
between two choices is high, people become faster and less
cautious in their decision making. To our knowledge, there is
no theory of media selection that accounts for this finding.
Therefore, in study two, we sought to replicate the decision
boundary effects observed in study one, testing if high subjec-
tive value differences between media choices could make peo-
ple select specific movies faster. Specifically, we expected that
both (H3a) movie valence and (H3b) movie arousal will have
a negative main effect on decision boundary. We also asked if
participants’ mood states influences decision boundary. Does
(RQ2) mood valence and (RQ3) the interaction between
mood valence and movie valence have a credible effect on de-
cision boundary? Similarly, we will also test if: (RQ4) mood
arousal and (RQ5) the interaction between MoodArousal and
MovieArousal have a credible effect on decision boundary.
Method
For study two, the decision tasks remained identical to study
one, however, in addition to the decision tasks, we added a
mood induction procedure (from Kuijsters et al., 2016; see
Supplementary Section 9 for description) after presenting par-
ticipants with the movie summaries in the training phase, but
before participants made preferential choices in the decision
tasks. Participants were randomly assigned to one of four
mood induction groups: positive valence high arousal, posi-
tive valence low arousal, negative valence high arousal, and
negative valence low arousal. Within each group, participants
were instructed to view randomized and counter-balanced
stimuli presentations of images for the corresponding mood
induction. The stimulus manipulation (Supplementary Section
8) as well as the mood induction were successful
(Supplementary Section 10). For each group, the mood induc-
tion procedure was repeated before each experimental block.
In other words, the procedure for each experimental block
(n¼7) was as follows: movie summary text presentation
(training) !mood induction !SAM mood rating (induction
check) !decision tasks (testing; Figure 1B). Thus, the experi-
ment was a 2 (mood induction arousal high/low) x2 (mood
induction valence positive/negative) x2 (movie summary
arousal high/low) x2 (movie summary valence positive/nega-
tive) factorial design.
Journal of Communication (2023) 7
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Analysis
We constructed a mixed model to estimate parameters in the
decision task as follows:
v;a;T;Z¼b0jsubject þb1MoodValence þb2MovieValence
þb3MoodValence MovieValence þb4MoodArousal
þb5MovieArousal þb6MoodArousal MovieArousalþe
Equation 4
Equation 4 estimates drift rate (v), decision boundary (a),
non-decision time (T), and bias (Z). The terms encode main
(MovieValence, MoodValence, MovieArousal, MovieValence)
and interaction (MoodValence * MovieValence and
MoodArousal * MovieArousal) effects. Preregistered model
comparisons are reported in Supplementary Section 11.
To test the hypotheses and research questions, the estimated
posterior distributions (v, a,T,andZ) for each decision type
and each experimental condition were analyzed by using the
HDDMRegressor function to fit a mixed model (with movie
arousal and movie valence as within-subjects variables and
mood valence and mood arousal as between-subjects variables).
In this model, we specified random intercepts for subjects and
fixed effects for the regressors. As in study one, each factor was
dummy coded (low/negative ¼0, high/positive ¼1) and differen-
ces (0 ¼no difference between choices, 1 ¼difference between
choices) were calculated. Therefore, the results reported for
study two can be interpreted in the same way as study one.
Sample
Study two was also conducted on https://pavlovia.org/ and
participants were undergraduate students. In total, n¼127
participants were sampled from the University of California,
Davis (see Supplementary Section 6 for participant character-
istics and Supplementary Section 12 for exclusion criteria).
Power simulations show that the probability that a HDDM
model can detect an effect asymptotes when participants per
condition reaches n>25 (Wiecki et al., 2013).
Results
Results are reported and inference testing is conducted
consistent with study one.
Drift-rate hypotheses
We hypothesized an interaction effect such that people will
have a positive drift rate toward positively valenced movies in
the negatively valenced mood condition (H1a), and will have
a negative drift rate toward negatively valenced movies in the
positively valenced mood condition (H1b). Drift rates toward
movie valence were negative in both the negatively valenced
mood condition (M¼0.124, 95% CI [0.168, 0.081],
100.0% <0<0.0%) and in the positively valenced mood
condition (M¼0.058, 95% CI [0.103, 0.014], 99.6% <
0<0.4%). Thus, H1b is supported, but H1a is not, as people
show a preference for negatively valenced movies regardless
of their mood valence. Additionally, we found that people in
a negative mood state had a lower drift rate than people in a
positive mood state (98.3% <0<1.7%).
We can further explore this result by examining the main
and interaction effects for movie and mood valence (for cell
means, see Supplementary Section 13). The results of the re-
gression model (Figure 3A) show that movie valence has a
negative main effect on drift rate (M¼0.129, 95% CI
[0.172, 0.086], 100.0% <0<0.0%), and the interaction
between mood valence and movie valence is credible and posi-
tive (M¼0.072, 95% CI [0.012, 0.134], 0.8% <0<99.2%).
This interaction means that, contrary to what we expected,
participants in a negative mood have a stronger preference for
negatively valenced movies (Figure 3B).
Similarly, we expected (H2a) that people will have a positive
drift rate toward high-arousal movies in low mood arousal con-
dition, and (H2b) will have a negative drift rate toward low-
arousal movies in the high-mood arousal condition. We show
(Figure 3A) that participants in the low-mood arousal condi-
tion had a credible and positive drift rate toward high-arousal
movies (M¼0.048, 95%CI [0.005, 0.090], 1.4% <
0<98.6%). Participants in high-mood arousal condition had
also a credible positive drift rate toward high-arousal movies
(M¼0.057, 95% CI [0.012, 0.101], 0.7% <0<99.3%).
There is no credible difference between the effects at different
levels of mood arousal (61.4% <0<38.6%).
We also examined the hypotheses by investigating the main
and interaction effects for movie and mood arousal (for cell
means, see Supplementary Section 13). Movie arousal has a
positive main effect on drift rate (M¼0.046, 95% CI [0.006,
0.088], 1.4% <0<98.6%). However, the interaction be-
tween mood arousal and movie arousal was not credible
(M¼0.011, 95% CI [0.048, 0.071], 35.8% <0<64.2%).
Instead, people prefer high-arousal movies regardless of their
mood arousal, supporting H2a but not H2b (Figure 3C).
RQ1 compared the effect size of the interaction between
mood valence and movie valence with the effect size of the in-
teraction between mood arousal and movie arousal. The dif-
ference between the two interaction effects was not credible
(8.1% <0<91.9%).
Decision boundary hypotheses
H3a and H3b specified that movie valence and movie arousal
would have a negative main effect on decision boundary.
Results (Figure 3A) show that only movie valence has a negative
effect on decision boundary (M¼0.170, 95% CI [0.242,
0.103], 100.0% <0<0.0%). Movie arousal did not have a
credible main effect on decision boundary (M¼0.052, 95%
CI [0.132, 0.030], 90.0% <0<10.0%). Thus, H3a is
supported, but H3b is not.
Finally, we asked if mood valence (RQ2), the interaction
between mood valence and movie valence (RQ3), mood
arousal (RQ4), and the interaction between mood arousal
and movie arousal (RQ5) had credible effects on decision
boundary. Results show that mood valence (M¼0.027,
95% CI [0.239, 0.177], 60.4% <0<39.6%), mood
arousal (M¼0.117, 95% CI [0.355, 0.105], 85.6% <
0<14.4%), the interaction between mood valence and movie
valence (M¼0.064, 95% CI [0.033, 0.165], 10.4% <
0<89.6%), and the interaction between mood arousal and
movie arousal (M¼0.003, 95% CI [0.106, 0.109], 47.3%
<0<52.7%) do not credibly influence decision boundary.
Exploratory analyses
Mood arousal has a credible negative effect on non-decision
time (M¼-0.264, 95% CI [0.422, 0.119], 100.0% <
0<0.0%). This means that participants in higher arousal
mood take less time (0.26 s) on reading the decision task
options and/or executing the decision (e.g., button press).
Other effects for non-decision time and decision-bias were not
credible.
8A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Figure 3 (A) Confirmatory analysis, study 2. Posterior probability distributions for the effects of movie valence/arousal, mood valence/arousal, and their
interactions on drift rate (upper left), decision boundary (upper right), decision bias (bottom left), and non-decision time (bottom right). (B) Interaction effect
between mood valence and movie valence. (C) Interaction effect between mood arousal and movie arousal.
Journal of Communication (2023) 9
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Discussion
In study two, we tested hypotheses drawn from MMT by in-
ducing specific mood states prior to a media decision-making
task. We successfully induced positive or negative mood states
of high or low arousal, and fit a HDDM regression model on
people’s RT and choice data.
Study two replicated study one in that people showed a
preference toward negatively valenced content. Moreover,
mood valence and movie valence had a credible interaction ef-
fect on drift rate. Contrary to the valence hypothesis of
MMT, participants in our study showed a stronger preference
toward negatively valenced movies when in a negatively
valenced mood. On the other hand, movie arousal had a posi-
tive effect on drift rate, meaning participants showed a prefer-
ence toward high-arousal movies when they are in a low-
arousal state—which supports MMT’s excitatory potential
prediction. Contrary to MMT, participants also showed a
trend towards high-arousal movies even when they were in a
high-arousal state.
Lastly, we replicated the effect of movie valence on the deci-
sion boundary from study one, finding that higher subjective
value differences between choices makes people less cautious
in their decisions.
Study three
Studies one and two were conducted using a college student
sample during the Covid-19 pandemic (2020 and 2021, re-
spectively). In order to increase the generalizability of our
findings, we conducted a third study that directly replicated
study two in a more representative sample of American
adults. Study three was conducted in April 2022, after the in-
troduction and widespread availability of the Covid-19 vac-
cine (2021) and lifting of many Covid-19 related restrictions.
Method
Analysis
The hypotheses, experimental design, data cleaning, model,
and data analysis procedure were identical to study two. The
stimulus manipulation (Supplementary Section 8) and mood
induction were successful (Supplementary Section 10).
Sample
N¼301 participants that were approximately representative
of the United States population (in terms of age, gender, and
ethnicity) were recruited from Prolific Academic (see
Supplementary Section 6 for participant characteristics and
Supplementary Section 14 for exclusion criteria). Participants
were compensated $7.92 for their time (M¼59.324 minutes,
S.D. ¼24.602). The task was hosted on https://pavlovia.org/.
Results
In study three, we expected that the previously specified
HDDM regression model (Equation 4) would replicate study
two’s findings.
Drift-rate hypotheses
H1a specified that participants in the negative mood valence
condition would have a positively signed drift rate whereas
H1b specified that participants in the positive mood valence
condition would have a negatively signed drift rate. Neither
H1a (M¼0.023, 95% CI [0.052, 0.006], 94.0% <
0<6.0%) or H1b (M¼0.008, 95% CI [0.037, 0.019],
72.1% <0<27.9%) were supported. As in study two, we
also interrogated the main and interaction effects for movie
and mood valence (for study three cell means, see
Supplementary Section 15). The regression model (Figure 4A)
shows that the negative main effect of movie valence on drift
rate (M¼0.024, 95% CI [0.052, 0.004], 95.1% <
0<4.9%) is not credible. The mood valence and movie va-
lence interaction are also not credible (M¼0.015, 95% CI
[0.025, 0.056], 24.2% <0<75.8%). In sum, H1a and
H1b were not supported, and they did not replicate the results
from study two.
H2a specified that participants in the low mood arousal
condition will have a positively signed drift rate, whereas H2b
specified that participants in the high mood arousal condition
will have a negatively signed drift rate. Neither H2a
(M¼0.043, 95% CI [0.072, 0.014], 53.8% <
0<46.2%) or H2b (M¼0.009, 95% CI [0.020, 0.039],
2.4% <0<97.6%) were supported. Examination of the
main and interaction effects for mood and movie arousal
(Figure 4A; cell means, Supplementary Section 15) showed
that movie arousal (M¼0.002, 95% CI [0.030, 0.025],
56.0% <0<44.0%) and the interaction between mood
arousal and movie arousal (M¼0.030, 95% CI [0.009,
0.072], 7.0% <0<93.0%) did not have a credible effect on
drift rate. Here again, H2a and H2b were not supported and
they did not replicate study two.
RQ1 compared the effect size of the interaction between
mood valence and movie valence with the effect size of the in-
teraction between mood arousal and movie arousal. As in
study two, the difference between the two interaction effects
was not credible (M¼0.016, 95% CI [0.040, 0.076],
29.6% <0<70.4%).
Decision boundary hypotheses
It was expected that movie valence (H3a) and movie arousal
(H3b) would have a negative main effect on decision bound-
ary (Figure 4A). Movie valence showed a credible main effect
(M¼0.121, 95% CI [0.169, 0.071], 100.0% <
0<0.0%), but not movie arousal (M ¼0.000, 95% CI
[0.026, 0.027], 56.0% <0<44.0%). Thus, H3a was sup-
ported and replicated the result observed in study two. H3b
was not supported in either study two or three.
Turning now to our research questions, we see that mood
valence (RQ2; M¼0.015, 95% CI [0.121, 0.099], 63.3%
<0<36.7%), the interaction between mood valence and
movie valence (RQ3; M¼0.013, 95% CI [0.057, 0.081],
33.5% <0<66.5%), mood arousal (RQ4; M¼0.017,
95% CI [0.147, 0.127], 66.2% <0<33.8%), and the inter-
action between mood arousal and movie arousal (RQ5;
M¼0.018, 95% CI [0.054, 0.083], 30.6% <0<69.4%)
had non-credible effects on decision boundary. None of the
decision boundary RQs were credible in study three. The
same was true for study two.
Exploratory analysis
These confirmatory drift rate results do not replicate study
two. However, we do see a credible replication of the negative
main effect of movie valence on decision boundary. What
explains these mixed replication results? One important dis-
tinction is that studies one and two were undergraduate stu-
dents whereas participants in study three were approximately
representative of the United States population and therefore
10 A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Figure 4 Study three posterior probability distributions for the effects of movie valence/arousal, mood valence/arousal, and their interactions on drift rate
(upper left), decision boundary (upper right), decision bias (bottom left), and non-decision time (bottom right) for the (A) confirmatory, and (B) exploratory
analysis.
Journal of Communication (2023) 11
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
comprised a wider range of ages. It could be that people’s me-
dia preferences and mood management behaviors might differ
by age. MMT does not specify age as a moderator. However,
previous research has shown that media selection varies by
age (Mares et al., 2008), and that older adults tend to prefer
positively valenced media, especially for mood management
(Mares et al., 2008;Shifriss et al., 2015).
Our results, particularly the negative main effect of movie
valence on drift rate, are in the same direction as study two,
and nearly meet our strict credibility threshold. If older adults
do prefer positively valenced media whereas younger adults
prefer negatively valenced media, then this preference could
pull the negative main effect closer to zero, thereby washing
out the main effect in the presence of a credible interaction
effect. To test this possible explanation, we included the par-
ticipant’s age in the HDDM regression model as a moderator
in the media selection process, as shown in Equation 5:
v;a;T;Z¼b0jsubject þb1MovieValence þb2MovieArousal
þb3MoodValence þb4MoodArousal
þb5Ageþb6MoodValence MovieValence
þb7MoodArousal MovieArousal
þb8AgeMovieValence þb9Age
MovieArousal þe
Equation 5
When modeled this way, and consistent with study two, we
see a credible negative main effect of movie valence
Figure 5 (A) Scatterplot showing the relationship between age and drift rate for movie valence. (B) Scatterplot showing the relationship between age and
drift rate for movie arousal. (C) Scatterplot showing the relationship between age and decision boundary for movie valence. (D) Scatterplot showing the
relationship between age and decision boundary for movie arousal.
12 A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
(M¼0.400, 95% CI [0.470, 0.333], 100.0% <
0<0.0%) on drift rate. We also see a credible positive inter-
action effect between age and movie valence (M¼0.008,
95% CI [0.007, 0.009], 0.0% <0<100.0%) on drift rate
(Figure 4B). This means that younger participants have a pref-
erence for negatively valenced movies whereas older partici-
pants have a preference for positively valenced movies
(Figure 5A).
Also consistent with study two, movie arousal had a credible
positive main effect (M¼0.150, 95% CI [0.083, 0.208], 0.0%
<0<100.0%) on drift rate, Moreover, the interaction between
age and movie arousal had a credible negative effect
(M¼0.003, 95% CI [0.004, 0.002], 100.0% <
0<0.0%) on drift rate (Figure 4B). This shows that younger
participants prefer high arousal movies and older participants
prefer low-arousal movies (Figure 5B).
As in study two and the confirmatory model in study three,
we still observed a credible negative main effect of movie va-
lence (M ¼0.119, 95% CI [0.167, 0.068], 100.0% <
0<0.0%) on decision boundary (Figure 4B). We also found a
credible positive main effect of age (M¼0.009, 95% CI
[0.004, 0.012], 0.0% <0<100.0%) and a negative interac-
tion effect between age and both of movie valence
(M¼0.002, 95% CI [0.004, 0.000], 99.1% <
0<0.9%) and movie arousal (M¼0.002, 95% CI [0.004,
0.000], 99.4% <0<0.6%) on decision boundary. This
shows that age influences people’s media decision cautious-
ness such that older participants are generally more cautious
in media selection, but will become less cautious when there is
a difference in either valence (Figure 5C) or arousal
(Figure 5D) among the media options.
Both movie arousal (M¼0.046, 95% CI [0.094,
0.005], 98.8% <0<1.2%) and age (M¼0.002, 95% CI
[0.003, 0.000], 98.0% <0<2.0%) have a credible nega-
tive main effect on non-decision time. The main effect for age
indicates that older participants take a shorter time to process
and execute a media selection task compared to younger par-
ticipants. The negative main effect for movie arousal means
that participants take a shorter time to process and execute a
media selection task when there is an arousal difference be-
tween the choices.
Finally, there is a credible positive main effect of both
mood valence (M¼0.015, 95% CI [0.000, 0.029], 1.8% <
0<98.2%) and mood arousal (M¼0.016, 95% CI [0.001,
0.031], 1.6% <0<98.4%) on decision bias (Figure 4B). This
means that the experimental manipulation changes a partici-
pant’s starting point to be biased toward positively valenced
or high-arousal choices. There is also a credible negative inter-
action effect between mood arousal and movie arousal on de-
cision bias (M¼0.025, 95% CI [0.043, 0.006], 99.7%
<0<0.3%), which means that a high-arousal mood biases
people’s decision starting point toward low-arousal movie
choices.
Discussion
Our confirmatory model in study three failed to replicate the
drift rate results observed in study two. However, the confir-
matory model did replicate the decision boundary result ob-
served in study two, namely that a valence difference between
media choice options makes people less cautious in their deci-
sion making.
However, prior research has shown that people’s use of me-
dia for mood management is moderated by age (Mares et al.,
2008;Shifriss et al., 2015). Therefore, we constructed an ex-
ploratory model that included participant age. When doing
so, we were able to reproduce the drift rate and decision
boundary results observed in study two, this time in a more
representative and heterogeneous sample of U.S. adults.
People have a preference for negatively valenced and high-
arousal media. When there is a valence difference between
media choices, people’s decision making becomes less
cautious.
We also found that age has a critical moderating effect on
media decision-making. Younger adults showed a strong pref-
erence for negatively valenced and high-arousal movies, while
by comparison, older adults showed a preference for posi-
tively valenced and low-arousal movies. This distinct media
preference among people of different ages is not a unique
finding. Mares et al. (2008) found that younger adults were
more inclined toward negative and arousing content, while
older adults preferred emotional stability and positively
valenced films. In news selection contexts, Bachleda et al.
(2020) and Soroka et al. (2021) found that age is significantly
associated with people’s negativity bias such that younger
adults have a stronger preference toward negative valenced
news compared to older adults. Ossenfort & Isaacowitz
(2018) found that older adults selected more positive-
valenced games compared to young adults, and Livingstone
and Isaacowitz (2015) found that older adults select positive
media as a mood management strategy.
In line with this previous work, our results provide more
evidence that people’s media preferences change as they grow
older. Our study is not well suited to explain why, as we did
not examine change longitudinally. But it does suggest that
age, which has been under-theorized in the literature, plays an
important role in media selection, particularly when selection
is framed as a decision-making process.
General discussion
In three studies, we applied a computational decision-making
model (HDDM) to MMT, a central media selection theory.
The results of all three experiments lead to a robust conclu-
sion that people prefer negatively valenced over positively
valenced movies, and high arousal over low-arousal movies,
in decision-making tasks between two options. This prefer-
ence holds, even after inducing different mood states (studies
two and three). We also show that age moderates media con-
tent preferences (study three). Younger adults prefer nega-
tively valenced and high-arousal media content, whereas these
preferences are reversed among older adults. Finally, our
results show that arousal and valence shape the subjective
value of a media choice option, and that people’s selection of
media content is determined by the comparison of the subjec-
tive value differences between two media choice options, con-
sistent with value-based decision theory.
Our project illustrates the utility of formal modeling for
testing communication theory by applying a computational
model of decision making to a well-known and established
theory of media selection. Most communication theories and
models are instantiated in what is known as a “verbal model”
in which constructs central to a theory, and their relationships
to each other, are verbally described, defined, and articulated
(Shoemaker et al., 2004). Testing of these models is often con-
ducted by collecting data and subjecting specific relationships
between variables to NHST. The results of these tests are used
Journal of Communication (2023) 13
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
to support, or fail to support, the verbally articulated relation-
ships. MMT is a verbal theory. It has primarily been tested us-
ing experimental methods and NHST, receiving mixed
support (Reinecke, 2016).
As noted by Fisher and Hamilton (2021), one limitation of
testing verbal theories is that there is room for individual
experimenters to argue, for example, what exactly is specified
by a theory, and how support for that hypothesis will be dem-
onstrated or falsified. In this case, conceptual ambiguity on
the part of scholars examining MMT hypotheses has led to
mixed support for the theory (Knobloch-Westerwick, 2014).
Computational or formal modeling, in contrast, translates
verbal descriptions of variables into mathematical relation-
ships between variables, which can be supported via mathe-
matical proofs (in mathematical models), by using simulated
data, or as in our study, experimental data to test the extent
to which models hold under specific parameters (Fisher &
Hamilton, 2021;Smaldino, 2020;van Rooij & Baggio,
2021). Thus, the risk of misinterpretation or inconsistent ap-
plication of relationships is reduced, as are potential argu-
ments about what constitutes disconfirming evidence.
An additional benefit of formal models is that when they
fail, they fail “forwards,” or productively. Even when formal
models are tested and demonstrate evidence unsupportive of
verbal theory, they do so in a way that allow for generative
theory building. In the following section, we examine our
results from the current studies in light of formal models. In
doing so, we provide a generative path forward for media
scholars interested in media selection.
Implications for MMT
Perhaps the most striking finding is the lack of confirmation
for MMT’s valence hypothesis. In three studies, we showed
that people have a preference for negatively valenced content.
In study one, we showed that people’s prevailing mood state
was unrelated to their media preferences. In study two we
showed that a preference for negatively valenced media is am-
plified when people are experimentally induced into a nega-
tively valenced mood. And in study three, we showed that age
moderates media preference with younger adults preferring
negatively valenced media whereas older adults prefer posi-
tively valenced media.
This preference for hedonically unpleasant media is not
unique to our study. Indeed, research on why viewers would
select tragic or sad films for entertainment (i.e., “the paradox
of tragedy”) has received substantial theoretical and research
attention for three decades (Oliver, 1993). Explanations put
forth for rationalizing why audiences select tragic films has in-
cluded the implicit anticipation of meta-emotions form tragic
film (Bartsch et al., 2008), self-determination theory-based
psychological need satisfaction (Tamborini et al., 2010), the
evolutionary benefits of fictional play (Steen & Owens,
2001), or the role of sad stories in evoking appreciation, nos-
talgia, or meaning (Oliver & Raney, 2011). We also note that
this finding was moderated by age, in line with past research
showing that younger viewers, in comparison to older, prefer
negatively valenced media (Mares et al, 2008). Therefore, and
in line with these studies, and from our own work here, we
can confidently state that affect is not the sole or primary
driver of media selection, as posited in the valence hypothesis
of MMT.
We did find that arousal partially operates in line with
MMT predictions. Individuals in a state of low arousal do se-
lect high-arousal films. However, we also saw a preference for
high-arousal films among participants in a high-arousal state.
This is similar to findings from Bryant and Zillmann (1984)
who found no difference between stressed and relaxed partici-
pants in terms of the excitatory potential of selected television
shows, as well as Strizhakova and Krcmar (2007), who found
that nervous subjects preferred horror movies and sad sub-
jects preferred crime dramas; both violations of MMT’s excit-
atory homeostasis hypothesis.
One could argue that the preference for high-arousal films
could be due to specific characteristics of our selection task or
the sample we selected. Yet, additional analyses demonstrate
that participants responded to each type of decision task in
systematically similar ways (Supplementary Section 15). This
means that our results were not driven by a small number of
idiosyncratic choices in our study. It could be that the pan-
demic played a moderating role in media selection (Eden et
al., 2020) that we did not account for (Holbert et al., 2022).
However, it is worth noting that our mood induction in stud-
ies two and three were successful. Therefore, such a critique
likely only applies to study one; although our successful
mood induction nullifies this critique for studies two and
three.
What does this mean for MMT and related theories of
affect-based media choice? We would note that some core
predictions from MMT were supported by our studies. Put
differently, valence does influence decision-making, but does
so by influencing the cautiousness of decisions. Studies two
and three show that people make less cautious decisions when
the subjective value difference between two choices is high.
Therefore, to the scholar interested in studying media choice
from a mood optimization perspective, we would suggest the
following amendments to MMT and related theories.
First, media choice may be an insufficiently specified depen-
dent variable for media selection studies. Decision cautious-
ness, in conjunction with selection, may more completely
reflect the cognitive processes occurring during the selection
period. Second, presentation order and comparative proper-
ties of the stimuli must be taken into account in future studies.
Extensive theoretical accounts of the role of stimulus presen-
tation or comparative effects are largely absent in media psy-
chology studies, yet our finding suggests that presentation
effects may be important to understand selective exposure
broadly. When considering the practical implications, we may
look towards Netflix or Hulu queues, which present several
similarly-valanced films in a row. Based on our findings, this
type of stimulus arrangement may push more cautious deci-
sion making, rather than if films with different attributes were
presented together. However, the role of stimulus presenta-
tion is largely absent from selection theories. Third, media se-
lection requires more extensive understanding and modeling
of demographically-based moderators of choice such as age
and context, as well as selected predictors of selection.
Clearly, people make preferential decisions on entertainment
movies not only based on the affective features, but also other
features like empathy, psychological needs, genre, and emo-
tion (Oliver, 1993;Tamborini et al., 2010;Oliver & Raney,
2011), as well as intrapersonal variables such as, individual
goals and characteristics (Knobloch-Westerwick, 2014). To
the extent to which these factors can be formally modeled, the
potential for falsifying specific hypotheses increases. Fourth,
14 A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
we would suggest that media choice research move beyond
the mood optimization hypothesis to examine other mecha-
nisms driving media selection and more faithfully modeling
the process. We turn to some of those models now.
Strengths and weakness of two-choice
decision-making models for media research
As Smaldino (2017) notes, “Models are stupid, and we need
more of them” (p. 311). When we use simple algorithmic rep-
resentations of complex and dynamic relationships, there is
room for disagreement about the manner in which variables
have been selected and represented. Clearly, DDM comes
with several practical constraints that potentially limit its ap-
plicability to test verbal theories. First, the parameterization
procedure for transitioning theoretical constructs from verbal
theories into latent variables requires accurate and solid mea-
surement and manipulation of constructs. Second, DDM
model fitting needs a relatively large sample of ubiquitous and
homogeneous observations within experimental conditions
for each participant with high consistency in experimental
manipulation across trials. MMT itself suggests two addi-
tional predictions—semantic affinity and intervention poten-
tial—which we did not formalize in our work for the above
two reasons; even if it may ultimately be possible to do so, in
principle.
However, despite the narrow focus of our work, our study
demonstrates how formalized “stupid” models can still elicit
valuable theoretical insights. Our project represents the first
ever formalization of MMT. To achieve this, we returned to
first-principles by formalizing two core MMT predictions (the
valence and excitatory homeostasis hypotheses). Future
researchers might investigate alternative model parameteriza-
tions in an attempt to better formalize MMT, or investigate
how previously undiscovered mechanisms (e.g., response cau-
tion) shape media selection. MMT assumes that affect-related
media selection functions as an automatic and uncontrolled
process (Zillmann, 1988). Given that this is a core assumption
of MMT, it has not been rigorously tested. But decades of re-
search show that automatic and uncontrolled processes are
just one element that guides decision making. Decision mak-
ing, including media selection, can also be governed by habit
and intentional or goal-oriented drives, which can also be in-
tegrated into the parameterization of DDM. This is because
DDM merges different preference-influencing factors (pavlov-
ian, habit, and goal-oriented, to be exact) together into a sin-
gular subjective value (Rangel et al., 2008), which is later
processed as evidence accumulation to reach the decision
boundary. Therefore, the value-based decision-making frame-
work functions as a generic theoretical framework for media
selection studies by integrating, comparing and testing differ-
ent verbal theories concurrently. With that said, it is possible
to separate these three preference-influencing factors. For in-
stance, future studies could fit a DDM with time-varying evi-
dence to investigate the extent to which automatic
and uncontrolled, or highly controlled processes guide media
selection (Diederich & Trueblood, 2018;Roberts &
Hutcherson, 2019). This is what we mean when we say that
computational models are generative. It was previously im-
possible to test this MMT assumption. But now, this assump-
tion can be tested, and media selection theory can be refined.
Alternatively, the DDM may be questioned in terms of its
utility as an ideal decision-making model under which to
examine media selection. Media selection is, after all, rarely a
binary choice between repeated alternatives. Research using
binary choice decision tasks has also shown a dependency be-
tween chosen and unchosen options. Over time, the value of
unchosen options is down weighted relative to the value of
chosen options such that repeatedly choosing one type of op-
tion negatively reinforces selection of a different type of op-
tion (Biderman & Shohamy, 2021). Admittedly, DDM and
the two-choice decision task lack a certain level of generaliz-
ability and external validity because they inadequately simu-
late and approximate real-world media selection scenarios.
Real-world media selection behavioral data requires tailored
complex behavioral modeling such as a multivariate version
of DDM for multi-alternative choice tasks (Krajbich &
Rangel, 2011), reinforcement learning models combined with
a probabilistic function for media engagement (Fisher &
Hamilton, 2021), or random walk models for sequential me-
dia selection (Lydon-Staley et al., 2020). However, despite its
low generalizability, DDM is still a valid and sufficient meth-
odology to reveal people’s preference for hypothetical media
content, and can be used to explain (why) media selection
process under an algorithmic (what) and implemental (how
media selection phenomenon relies on biological process)
level (Huskey et al., 2020).
Our study shows just how much can be learned by simplify-
ing the decision landscape and that the mathematical simplic-
ity necessary to formalize a verbal model is heuristically
provocative, a theoretical contribution in its own right
(DeAndrea & Holbert, 2017). Additionally, formalizing dy-
namic, complex, and multichoice decision making is notori-
ously non-trivial, although state-of-the-art approaches are
beginning to emerge (Yoo et al., 2021). We offer a starting
point from which others may design and test their own media
selection models (for an extended theoretical treatment, see
Gong & Huskey, in press).
Conclusion
In conclusion, we argue that inconsistent conceptualization
via verbal models of media choice may have insufficiently
specified the role of valence and arousal on media selection
processes. If we evaluate verbal models of media choice, such
as MMT, based on established criteria for evaluating theory
in communication science (DeAndrea & Holbert, 2017), it is
clear from our own data that past verbal models may lack ex-
planatory power in terms of accurately depicting media choice
behavior. We argue that this lack of explanatory power likely
stems from the multiple definitions used to describe affect,
arousal, and selection in past work, which makes MMT’s hy-
potheses nearly impossible to falsify. In contrast to past work,
by fitting a computational model (HDDM), we were able to
demonstrate that MMT’s valence hypotheses are fully falsi-
fied, and its arousal hypothesis is partially falsified.
Therefore, our work helps to improve verbal theories of me-
dia choice by increasing falsifiability. In addition, we hope to
increase the accuracy and explanatory power of these models
by including previously unaccounted for boundary
conditions.
How can we be confident our findings may argue against a
35-year-old theory? Using a method that depends on compu-
tational power and cognitive understanding of decision-
making that were not present 35 years ago when MMT was
formulated, we were able to more accurately specify under
Journal of Communication (2023) 15
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
which conditions specific propositions received support, as
well as identify a previously undiscovered mechanism, re-
sponse caution, as central to affect-driven media selection
processes. Thus, our contribution is a timely update to the
theory, based on understanding media selection as a value-
based decision-making task under specific conditions.
However, we would also note that our study does not reduce
the undeniable heuristic provocativeness or organizing power
of MMT. Indeed, we identify that affect and arousal are pow-
erful determinants of media choice under specific decision-
making conditions. We hope that our work drives further in-
terest in value-based decisions making tasks across communi-
cation subfields beyond media choice.
Finally, our work is in line with calls for communication
scientists to create theory which can be specified at computa-
tional (why does a behavior exist), algorithmic (what mathe-
matical rules govern the behavior), and implementation (how
is that behavior biologically implemented) levels of explana-
tion in order for the field to progress as a science (Huskey
et al., 2020). Formal models, which exist at the algorithmic
level, connect the computational and implementation levels.
Communication research has made great strides explaining
and describing behavior at computational and implementa-
tion levels separately. Formal models have the capacity to
clearly and unambiguously link these levels (for an example,
see Wang et al., 2015). Using formal models allows communi-
cation scientists to bridge individual and group level processes
studied at the different levels of explanation, that is, they are
scalable. The individual-level phenomena of interest to com-
munication researchers may be directly connected via formal
modeling to group-level phenomena of interest (Fisher &
Hamilton, 2021;Wiradhany et al., 2021). We join the effort
(e.g., Chung et al., 2012;Chung & Fink, 2022;Fink, 1993;
Huskey et al., 2020;Wang et al., 2006,2011,2015) to move
communication science in this direction.
Citation diversity statement
Citation disparities exist in communication research
(Chakravartty et al., 2018;Trepte & Loths, 2020;Wang
et al., 2021). We quantify our citation practices by including
a citation diversity statement (Supplementary Section 16;
Zurn et al., 2020).
Supplementary material
Supplementary material is available at Journal of
Communication online.
Notes
1. MMT was originally called the “theory of affect-dependent stimulus
arrangement” and is part of a broader body of work which suggests
that media selection is a function of the affective state of media users
and that selection will follow the principle of mood optimization
(Reinecke, 2016).
2. It should be possible, in principle, to test the intervention potential
and semantic affinity hypotheses using DDM. Doing so requires de-
veloping reliable and valid manipulations for variables related to
these hypotheses, and verifying that these manipulations can be
done in a way that is invariant across a large number of trials.
3. Unlike frequentist approaches to null-hypothesis significance testing
(NHST), Bayesian inference eschews “significant/not-significant”
language in favor of “credible/not-credible” language. Whereas
NHST conventionally specifies a¼.05 as the threshold for a
“significant” result, there is not a conventional cutoff that demar-
cates a “credible” result from a “not-credible” result, at least when
conducting inference on the posterior probability distribution as we
do in this manuscript. Some HDDM studies set relatively liberal
thresholds for a “credible” result (e.g., 90%, see Eikemo et al.,
2017). In our case, and given the confirmatory nature of our project,
we have adopted a stricter threshold of 97.5%, which is roughly
equivalent to a two-tailed NHST at a¼.05. A complete treatment
of Bayesian inference on a posterior probability distribution is be-
yond the scope of our current paper, but we direct interested readers
to Kruschke (2013), which details the inferential approach used in
our project.
References
Bachleda, S., Neuner, F. G., Soroka, S., Guggenheim, L., Fournier, P., &
Naurin, E. (2020). Individual-level differences in negativity biases in
news selection. Personality and Individual Differences,155,
109675. https://doi.org/10.1016/j.paid.2019.109675
Bamman, D., O’Connor, B., & Smith, N. (August, 2013). Learning
Latent Personas of Film Characters. Proceedings of the Annual
Meeting of the Association for Computational Linguistics (ACL
2013), Sofia, Bulgaria.
Bartsch, A., Vorderer, P., Mangold, R., & Viehoff, R. (2008). Appraisal
of Emotions in Media Use: Toward a Process Model of
Meta-Emotion and Emotion Regulation. Media Psychology,11(1),
7–27. https://doi.org/10.1080/15213260701813447
Biderman, N., & Shohamy, D. (2021). Memory and decision making in-
teract to shape the value of unchosen options. Nature
Communications,12(1), 4648. https://doi.org/10.1038/s41467-021-
24907-x
Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The
self-assessment manikin and the semantic differential. Journal of
Behavior Therapy and Experimental Psychiatry,25(1), 49–59.
https://doi.org/10.1016/0005-7916(94)90063-9
Bradley, M. M., & Lang, P. J. (1999). Affective norms for English words
(ANEW): Instruction manual and affective ratings (Technical
Report No. C-1). University of Florida, NIMH Center for Research
in Psychophysiology.
Bryant, J., & Zillmann, D. (1984). Using television to alleviate boredom
and stress: Selective exposure as a function of induced excitational
states. Journal of Broadcasting,28(1), 1–20. https://doi.org/10.
1080/08838158409386511
Busemeyer, J. R., Gluth, S., Rieskamp, J., & Turner, B. M. (2019).
Cognitive and neural bases of multi-attribute, multi-alternative,
value-based decisions. Trends in Cognitive Sciences,23(3), 251–263.
https://doi.org/10.1016/j.tics.2018.12.003
Carpentier, F. R. (2020). Mood management. In J. Van den Bulck (Ed.),
The international encyclopedia of media psychology (pp. 1–8).
Wiley. https://doi.org/10.1002/9781119011071.iemp0255.
Chakravartty, P., Kuo, R., Grubbs, V., & McIlwain, C. (2018).
#CommunicationSoWhite. Journal of Communication,68(2),
254–266. https://doi.org/10.1093/joc/jqy003
Chung, S., & Fink, E. L. (2022). Mathematical models of message dis-
crepancy: Previous models and a modified psychological discounting
model. Communication Theory,32(4), 471–487. https://doi.org/10.
1093/ct/qtac010
Chung, S., Fink, E. L., Waks, L., Meffert, M. F., & Xie, X. (2012).
Sequential information integration and belief trajectories: An experi-
mental study using candidate evaluations. Communication
Monographs,79(2), 160–180. https://doi.org/10.1080/03637751.
2012.673001
16 A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
DeAndrea, D. C., & Holbert, R. L. (2017). Increasing clarity where it is
needed most: Articulating and evaluating theoretical contributions.
Annals of the International Communication Association,41(2),
168–180. https://doi.org/10.1080/23808985.2017.1304163
Diederich, A., & Trueblood, J. S. (2018). A dynamic dual process model
of risky decision making. Psychological Review,125, 270–292.
https://doi.org/10.1037/rev0000087
Dienlin, T., Johannes, N., Bowman, N. D., Masur, P. K., Engesser, S.,
Ku¨ mpel, A. S., Lukito, J., Bier, L. M., Zhang, R., Johnson, B. K.,
Huskey, R., Schneider, F. M., Breuer, J., Parry, D. A., Vermeulen, I.,
Fisher, J. T., Banks, J., Weber, R., Ellis, D. A., ... de Vreese, C.
(2021). An agenda for open science in communication. Journal of
Communication,71(1), 1–26. https://doi.org/10.1093/joc/jqz052
Eden, A. L., Johnson, B. K., Reinecke, L., & Grady, S. M. (2020). Media
for coping during COVID-19 social distancing: Stress, anxiety, and
psychological well-being. Frontiers in Psychology,11, 577639–21.
https://doi.org/10.3389/fpsyg.2020.577639
Eikemo, M., Biele, G., Willoch, F., Thomsen, L., & Leknes, S. (2017).
Opioid Modulation of Value-Based Decision-Making in Healthy
Humans. Neuropsychopharmacology: Official Publication of the
American College of Neuropsychopharmacology,42(9),
1833–1840. https://doi.org/10.1038/npp.2017.58
*PFaul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). GPower 3:
A flexible statistical power analysis program for the social, behav-
ioral, and biomedical sciences. Behavior Research Methods,39(2),
175–191. https://doi.org/10.3758/BF03193146
Fink, E. L. (1993). Mathematical models for communication: An intro-
duction. Journal of Communication,43(1), 4–7. https://doi.org/10.
1111/j.1460-2466.1993.tb01245.x
Fisher, J. T., & Hamilton, K. A. (2021). Integrating media selection and
media effects using decision theory. Journal of Media Psychology,
33(4), 215–225. https://doi.org/10.1027/1864-1105/a000315
Gong, X., & Huskey, R. (in press). Computational methods and formal
modeling in entertainment research. In N. D., Bowman (Ed.),
DeGruyter handbook of entertainment. (Volume 1.). DeGruyter.
Gong, X., & Huskey, R. (in press). Drift diffusion modeling and online
data collection: A tutorial and some recommendations. American
Behavioral Scientist.
Greifeneder, R., Bless, H., & Pham, M. T. (2011). When do people rely
on affective and cognitive feelings in judgment? A review.
Personality and Social Psychology Review: An Official Journal of
the Society for Personality and Social Psychology, Inc,15(2),
107–141. https://doi.org/10.1177/1088868310367640
Hartley, C., & Sokol-Hessner, P. (2018). Affect is the foundation of
value. In A. S. Fox, R. C. Lapate, A. J. Shackman, & R. J. Davidson
(Eds.), The nature of emotion (2nd ed., pp. 348–351). Oxford
University Press.
Holbert, R. L., Baik, E. S., Tallapragada, M., Hardy, B. W., Tolan, C.
M., & LaMarre, H. L. (2022). Pandemic as boundary condition in
service to communication theory building. Annals of the
International Communication Association,46(3), 231–246. https://
doi.org/10.1080/23808985.2022.2108878
Huskey, R., Bue, A. C., Eden, A., Grall, C., Meshi, D., Prena, K.,
Schma¨lzle, R., Scholz, C., Turner, B. O., & Wilcox, S. (2020).
Marr’s tri-level framework integrates biological explanation across
communication subfields. Journal of Communication,70(3),
356–378. https://doi.org/10.1093/joc/jqaa007
Knobloch-Westerwick, S. (2014). Choice and preference in media use:
Advances in selective exposure theory and research. Routledge.
Krajbich, I., & Rangel, A. (2011). Multialternative drift-diffusion model
predicts the relationship between visual fixations and choice in val-
ue-based decisions. Proceedings of the National Academy of
Sciences,108(33), 13852–13857. https://doi.org/10.1073/pnas.
1101328108
Krajbich, I., Armel, C., & Rangel, A. (2010). Visual fixations and the
computation and comparison of value in simple choice. Nature
Neuroscience,13(10), 1292–1298. https://doi.org/10.1038/nn.2635
Krajbich, I., Hare, T., Bartling, B., Morishima, Y., & Fehr, E. (2015). A
common mechanism underlying food choice and social decisions.
PLoS Computational Biology,11(10), e1004371. https://doi.org/10.
1371/journal.pcbi.1004371
Krajbich, I., Lu, D., Camerer, C., & Rangel, A. (2012). The attentional
drift-diffusion model extends to simple purchasing decisions.
Frontiers in Psychology,3, 193–18. https://doi.org/10.3389/fpsyg.
2012.00193
Kruschke, J. K. (2013). Bayesian estimation supersedes the t test.
Journal of Experimental Psychology. General,142(2), 573–603.
https://doi.org/10.1037/a0029146
Kuijsters, A., Redi, J., de Ruyter, B., & Heynderickx, I. (2016). Inducing
sadness and anxiousness through visual media: Measurement techni-
ques and persistence. Frontiers in Psychology,7, 1141, https://doi.
org/10.3389/fpsyg.2016.01141
Lewis, N. A. Jr, (2020). Open communication science: A primer on why
and some recommendations for How. Communication Methods and
Measures,14(2), 71–82.
Livingstone, K. M., & Isaacowitz, D. M. (2015). Situation selection and
modification for emotion regulation in younger and older adults.
Social Psychological and Personality science,6(8), 904–910. https://
doi.org/10.1177/1948550615593148
Lydon-Staley, D. M., Zhou, D., Blevins, A. S., Zurn, P., & Bassett, D. S.
(2020). Hunters, busybodies and the knowledge network building
associated with deprivation curiosity. Nature Human Behaviour,
1–10. https://doi.org/10.1038/s41562-020-00985-7
Mares, M. L., Oliver, M. B., & Cantor, J. (2008). Age differences in
adults’ emotional motivations for exposure to films. Media
Psychology,11(4), 488–511. https://doi.org/10.1080/15213260802
492026
Milosavljevic, M., Malmaud, J., Huth, A., Koch, C., & Rangel, A.
(2010). The drift diffusion model can account for value-based choice
response times under high and low time pressure. Judgment and
Decision Making,5(6), 437–449. https://resolver.caltech.edu/
CaltechAUTHORS:20130816-103405233
Oliver, M. B. (1993). Exploring the Paradox of the Enjoyment of Sad
Films. Human Communication Research,19(3), 315–342. https://
doi.org/10.1111/j.1468-2958.1993.tb00304.x
Oliver, M. B., & Raney, A. A. (2011). Entertainment as pleasurable and
meaningful: Identifying hedonic and eudaimonic motivations for en-
tertainment consumption. Journal of Communication,61(5),
984–1004. https://doi.org/10.1111/j.1460-2466.2011.01585.x
Ossenfort, K. L., & Isaacowitz, D. M. (2018). Video games and emotion
regulation: Aging and selection of interactive stimuli. GeroPsych,
31(4), 205–213. https://doi.org/10.1024/1662-9647/a000196
Rangel, A., Camerer, C., & Montague, P. R. (2008). A framework for
studying the neurobiology of value-based decision making. Nature
Reviews. Neuroscience,9(7), 545–556. https://doi.org/10.1038/
nrn2357
Ratcliff, R., & McKoon, G. (2008). The diffusion decision model:
Theory and data for two-choice decision tasks. Neural
Computation,20(4), 873–922. https://doi.org/10.1162/neco.2008.
12-06-420
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
Reinecke, L. (2016). Mood management theory. In P. Ro¨ ssler, C.
Hoffner, & L. Van Zoonen (Eds.), The international encyclopedia of
media effects (pp. 1–13). Wiley. https://doi.org/10.1002/9781118
783764.wbieme0085
Roberts, I. D., & Hutcherson, C. A. (2019). Affect and decision making:
Insights and predictions from computational models. Trends in
Cognitive Sciences,23(7), 602–614. https://doi.org/10.1016/j.tics.
2019.04.005
Schwarz, N. (2012). Feelings-as-information theory. In P. A. M. Van
Lange, A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theo-
ries of social psychology (pp. 289–308). Sage Publications Ltd.
https://doi.org/10.4135/9781446249215.n15.
Shifriss, R., Bodner, E., & Palgi, Y. (2015). When you’re down and trou-
bled: Views on the regulatory power of music. Psychology of Music,
43(6), 793–807. https://doi.org/10.1177/0305735614540360
Journal of Communication (2023) 17
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023
Shoemaker, P. J., Tankard, J. W., Jr., & Lasorsa, D. L. (2004). How to
build social science theories. Sage Publications.
Shulman, H. C., & Bullock, O. M. (2019). Using metacognitive cues to
amplify message content: A new direction in strategic communica-
tion. Annals of the International Communication Association,
43(1), 24–39. https://doi.org/10.1080/23808985.2019.1570472
Smaldino, P. E. (2017). Models are stupid, and we need more of them.
In R. Vallacher, S. Read, & A. Nowak (Eds.), Computational social
psychology (pp. 311–331). Routledge. https://doi.org/10.4324/
9781315173726-14
Smaldino, P. E. (2020). How to translate a verbal theory into a formal
model. Social Psychology,51(4), 207–218. https://doi.org/10.1027/
1864-9335/a000425
Song, H., Tolochko, P., Eberl, J.-M., Eisele, O., Greussing, E.,
Heidenreich, T., Lind, F., Galyga, S., & Boomgaarden, H. G. (2020).
In validations we trust? The impact of imperfect human annotations
as a gold standard on the quality of validation of automated content
analysis. Political Communication,37(4), 550–572. https://doi.org/
10.1080/10584609.2020.1723752
Soroka, S., Guggenheim, L., & Valentino, D. (2021). Valence-based
biases in news selection. Journal of Media Psychology,33(3),
145–154. https://doi.org/10.1027/1864-1105/a000292
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A.
(2002). Bayesian measures of model complexity and fit. Journal of
the Royal Statistical Society Series B: Statistical Methodology),
64(4), 583–639. https://doi.org/10.1111/1467-9868.00353
Steen, F. F., & Owens, S. A. (2001). Evolution’s pedagogy: An adapta-
tionist model of pretense and entertainment. Journal of Cognition
and Culture,1(4), 289–321.
Strizhakova, Y., & Krcmar, M. (2007). Mood management and video
rental choices. Media Psychology,10(1), 91–112. http://doi.org/10.
1080/15213260701301152
Tajima, S., Drugowitsch, J., & Pouget, A. (2016). Optimal policy for
value-based decision-making. Nature Communications,7(1),
12400. https://doi.org/10.1038/ncomms12400
Tamborini, R., Bowman, N. D., Eden, A., Grizzard, M., & Organ, A.
(2010). Defining media enjoyment as the satisfaction of intrinsic
needs. Journal of Communication,60(4), 758–777. https://doi.org/
10.1111/j.1460-2466.2010.01513.x
Trepte, S., & Loths, L. (2020). National and gender diversity in commu-
nication: A content analysis of six journals between 2006 and 2016.
Annals of the International Communication Association,44(4),
289–311.
van Rooij, I., & Baggio, G. (2021). Theory before the test: How to build
high-verisimilitude explanatory theories in psychological science.
Perspectives on Psychological Science: A Journal of the Association
for Psychological Science,16(4), 682–697. https://doi.org/10.1177/
1745691620970604
Wang, X., Dworkin, J. D., Zhou, D., Stiso, J., Falk, E. B., Bassett, D. S.,
Zurn, P., & Lydon-Staley, D. M. (2021). Gendered citation practices
in the field of communication. Annals of the International
Communication Association,45(2), 134–153. https://doi.org/10.
1080/23808985.2021.1960180
Wang, Z. (2014). Bridging media processing and selective exposure: A
dynamic motivational model of media choices and choice response
time. Communication Research,41(8), 1064–1087. https://doi.org/
10.1177/0093650214534963
Wang, Z, Busemeyer, J. R., & Lang, A. (2006). Grazing or staying
tuned: A stochastic dynamic model of channel changing behavior. In
Proceedings of the twenty-eighth annual conference of the Cognitive
Science Society (pp. 870–875). Erlbaum.
Wang, Z., Lang, A., & Busemeyer, J. R. (2011). Motivational processing
and choice behavior during television viewing: An integrative dy-
namic approach. Journal of Communication,61(1), 71–93. https://
doi.org/10.1111/j.1460-2466.2010.01527.x
Wang, Z., Vang, M., Lookadoo, K., Tchernev, J. M., & Cooper, C.
(2015). Engaging high-sensation seekers: The dynamic interplay of
sensation seeking, message visual-auditory complexity and arousing
content. Journal of Communication,65(1), 101–124. https://doi.org/
10.1111/jcom.12136
Warriner, A. B., Kuperman, V., & Brysbaert, M. (2013). Norms of va-
lence, arousal, and dominance for 13,915 English lemmas. Behavior
Research Methods,45(4), 1191–1207. https://doi.org/10.3758/
s13428-012-0314-x
Wiecki, T. V., Sofer, I., & Frank, M. J. (2013). HDDM: Hierarchical
Bayesian estimation of the drift-diffusion model in Python. Frontiers
in Neuroinformatics,7, 14, https://doi.org/10.3389/fninf.2013.
00014
Wilson, R. C., & Collins, A. G. (2019). Ten simple rules for the compu-
tational modeling of behavioral data. eLife,8, e49547. https://doi.
org/10.7554/eLife.49547
Wiradhany, W., Baumgartner, S., & de Bruin, A. (2021).
Exploitation–exploration model of media multitasking. Journal of
Media Psychology,33(4), 169–180. https://doi.org/10.1027/1864-
1105/a000303
Yoo, S. B. M., Hayden, B. Y., & Pearson, J. M. (2021). Continuous deci-
sions. Philosophical Transactions of the Royal Society of London.
Series B, Biological sciences,376(1819), 20190664. https://doi.org/
10.1098/rstb.2019.0664
Zillmann, D. (1988). Mood management through communication
choices. American Behavioral Scientist,31(3), 327–340. https://doi.
org/10.1177/000276488031003005
Zurn, P., Bassett, D. S., & Rust, N. C. (2020). The citation diversity
statement: A practice of transparency, a way of life. Trends in
Cognitive Sciences,24(9), 669–672. https://doi.org/10.1016/j.tics.
2020.06.009
18 A computational model of mood management theory
Downloaded from https://academic.oup.com/joc/advance-article/doi/10.1093/joc/jqad020/7194700 by Serials Records Section user on 13 June 2023