Emotional Volatility and Cultural Success

Jonah Berger

Yoon Duk Kim

Robert Meyer

University of Pennsylvania

Jonah Berger (jberger@wharton.upenn.edu) is an associate professor of marketing at the

Wharton School at the University of Pennsylvania, 3730 Walnut St. Philadelphia PA 19104. The

authors thank Bowen Liu and Russell Richie for assistance with various aspects of the project

and participants at the Ohio State University Psychology seminar, Behavioral Insights from Text

Conference, and Marketing Science conference for helpful feedback.

1

ABSTRACT

Some cultural products (e.g., movies and books) catch on and become popular, but less is known

about why certain succeed and others fail. While some have argued that success is unpredictable,

we suggest that period-to-period shifts in emotional tone—what we term emotional volatility—

plays an important role. Automated sentiment analysis of thousands of movies demonstrates that

more emotionally volatile movies are evaluated more positively. This relationship holds

controlling for a range of other factors, and, consistent with the notion that emotional volatility

makes experiences more stimulating, is stronger in genres where evaluations are more likely to

be driven stimulation (i.e., thrillers rather than romance). By manipulating emotional volatility in

a follow up experiment, we underscore its causal impact on evaluations, and provide preliminary

evidence for the role of stimulation and engagement in driving these effects. Taken together,

these results shed light on why things become popular, the time dynamics of emotion, and the

psychological foundations of culture more broadly.

Cultural success | natural language processing | psychological foundations of culture

2

Why do some cultural products succeed while others fail? Cultural propagation, artistic

change, and the diffusion of innovations have been examined across disciplines with the goal of

understanding why things catch on (Bass 1969, Boyd and Richerson 1985, Kashima 2014,

Cavalli-Sforza and Feldman 1981, Rogers 1995, Salganik et al. 2006, Simonton 1980). Some

movies become blockbusters, while others languish. Some stories are adored while others flop.

What leads certain cultural products to win out in the marketplace of ideas?

One possibility is that success is random. Even domain experts are notoriously bad at

recognizing hits and failures in advance (Bielby and Bielby 1994, Hirsh 1972) and Hollywood

often spends millions of dollars promoting movies that end up being duds. This has led some to

argue that success is driven more by patterns of social influence than anything about the cultural

items themselves (Adler, 1985; Salganik et al. 2006).

In contrast, we suggest that individual-level psychological processes play an important

role. Research on cross-cultural psychology demonstrates how culture shapes psychological

processes (Markus and Kitayama 1991), but the reverse is also true: psychological processes

influence what people remember, like, and share, which in turn shapes collective culture

(Kashima 2008, Schaller and Crandall 2004). Arousal shapes social transmission (Berger 2011)

for example, leading stories that evoke more high arousal emotions to be more likely to go viral

and make the most emailed list (Berger and Milkman 2012). Thus, psychological processes may

act as a selection mechanism, determining which things prosper and which fall flat (Norenzayan

et al. 2006).

Along these lines, we suggest that period-to-period shifts in emotional tone—what we

term emotional volatility—helps shape success. We test this possibility in both the lab and the

field. First, we perform automated sentiment analysis on thousands of movies, testing whether

3

more emotionally volatile movies are received more positively. Second, by directly

manipulating emotional volatility in an experiment, we can rule out alternative explanations and

underscore its causal impact.

This work makes three main contributions. First, while a great deal of research has

examined how emotions influence evaluations, emotional dynamics (i.e., period to period

changes) have received less attention (Kuppens and Verduyn 2017). We demonstrate how one

dynamic feature, emotional volatility, shapes responses, and show that this effect holds even

controlling for a range of other features. Further, in the general discussion, we outline other

potential dynamic features that might deserve further attention.

Second, while researchers have long speculated about narrative structure, there have been

few empirical tests. In his unpublished master’s thesis, Kurt Vonnegut argued that “stories have

shapes which can be drawn on graph paper,” and suggested that that there were eight or so main

trajectories. Others have argued that there are seven (Booker, 2004), twenty (Tobias, 1993) or

thirty-six (Polti 1921) basic plots. But while these suggestions are intriguing, little work has

actually empirically tested them (cf. Reagan et al 2016). Further, while identifying and

classifying key plot types is certainly worthwhile, it leaves open the question of whether and why

certain narrative structures might be more engaging. We address this question, quantifying one

aspect of narrative, suggesting why it might be valuable, and demonstrating its impact

Third, we illustrate how natural language processing can be used to study cultural

success. While there has been great interest in why things catch on across disciplines,

measurement has proved difficult. Recently, however, researchers have highlighted the value of

using automated textual analysis for consumer research (Humphreys and Wang 2017; Netzer,

Feldman, Goldenberg, and Fresko 2012; Netzer, Lemaire, and Herzenstein 2018; Moore and

4

McFerran 2017; Packard, Moore, and McFerran 2018; Rocklage and Fazio 2015; Rocklage,

Rucker, and Nordgren 2018) and begun to apply these insights to narratives (Eliashberg, Hui,

and Zhang 2014). We demonstrate how natural language processing can help address these

challenges, measuring key features at scale, and opening up a range of interesting questions for

future research.

EMOTIONS AND EVALUATIONS

Decades of research have examined how aspects of experiences shape evaluations. At

the most basic level, experiences can be described by valence. Some experiences, like eating

tasty food, relaxing, or winning an award, are generally positive. Other experiences, like getting

fired, missing a flight, or waiting on hold, are generally negative. Not surprisingly, people

generally prefer positive experiences to negative ones, and a great deal of evidence supporting

the hedonistic principle (i.e., approach pleasure and avoid pain) is consistent with this

perspective. [That said, as the consumption of horror movies or roller coaster rides

demonstrates, in some situations people do value consuming negative feelings (Andrade and

Cohen 2007).]

Peaks and ends of experiences also play a role. Research on duration neglect

(Fredrickson and Kahneman 1993; Kahneman, Fredrickson, Schreiber, and Redelmeier 1993)

suggests that rather than adding up every moment of an experience and summing them together,

people often judge an experience based on two snapshots: how they felt at its peak, or most

intense point, and how they felt at the end. One might imagine, for example, that making a

painful colonoscopy longer should make it more unpleasant. But adding an additional less

5

negative period to the end of the experience actually made it less unpleasant (Redelmeier, Katz,

and Kahneman 2003), consistent with the notion that experiences are evaluated by their

endpoints.

But while a good deal of work has examined how the valence, or specific moments of an

experience shape evaluations, less attention has been given to emotion dynamics (Kuppens and

Verduyn 2017). While watching a movie, for example, emotions may fluctuate over time.

Might those fluctuations shape evaluations, and if so, how?

EMOTIONAL VOLATILITY

We suggest that emotional volatility plays an important role in shaping evaluations and

cultural success. The term volatility is often used to describe variation or dispersion. In the case

of the stock market, for example, volatility refers to the standard deviation or variance of a given

security or market index. Volatile stocks are those that frequently fluctuate up and down.

Volatility can also be used to describe the emotional nature of an experience. Consider,

for example, two different four-part sequences, such as chunks of a movie or book. The first

sequence has two positive periods followed by two negative ones (i.e., P1P2N1N2), while the

second sequence has the same four periods but alternates positive and negative, positive and

negative (i.e., P1N1P2N2). Both experiences have the same average valence, the same peak and

end, and even the same average distance from the mean, but the second is more emotionally

volatile. It is characterized by greater period-to-period shifts in valence.

We suggest such volatility may increase evaluations. While we are unaware of research

that has directly examined emotional volatility, work in other domains indicates that variation

6

can be beneficial because it increases stimulation. When doing the same thing over and over,

people tend to become bored. They tire of eating the same foods (Rolls et al. 1981; Rolls, van

Duijvenvoorde, and Rolls 1984) or listening to the same music (Ratner et al. 1999).

Consequently, adding variety can be beneficial. Variation can be stimulating (McAliser and

Pessemier 1982; Pessemier and Handlesman 1984), and this stimulation, in turn, can have

beneficial effects. Etkin and Mogilner (2016), for example, found that doing more varied

activities over the course of an hour could make that time feel more stimulating, exciting, and

engaging, which increased how happy or satisfied people were with the time.

Do the benefits of variation extend to emotional experiences? Variety research usually

compared whether doing different positive things is better than doing the same thing over and

over. In Rolls et al (1984), for example, participants found the meal more pleasant if they eat

sausages for one course and bread and butter for the next, rather than easting sausage for all the

courses.

But while it makes sense that switching between liked foods makes a meal more

enjoyable, it’s less obvious that volatility in emotional experiences would be similarly beneficial

because it adds negative (or at least less positive) experiences. Rather than variety coming from

different but liked items or experiences (e.g., chocolate and vanilla ice cream), variation among

emotional experiences necessarily involves adding more and less positive things (e.g., chocolate

ice cream and SPAM). Adding positive things to negative ones seems good, but adding negative

things to positive ones seems less ideal. Compare two situations: a week in which you get one

acceptance letter and then another acceptance letter versus a week in which you get one

acceptance letter and then a rejection letter. The second situation is certainly more emotionally

variable, but most people would likely prefer the first and evaluate it more positively. Similarly,

7

research on child development shows that parental conflict induces stress, anxiety, and

depression and leads to difficultly focusing and worse performance at school (Reynolds, et al.

2014). Consequently, while emotional volatility may increase stimulation, no one has looked at

whether the overall effect will be positive or negative.

In the case of cultural products, we suggest that emotional volatility may be beneficial

because of the distance between the negative things and the self. Work on humor, for example,

suggest that humor occurs when a violation (i.e., a threatening stimulus) is made benign (i.e.,

okay, McGraw and Warren 2010). Related work (McGraw, et al 2012) highlights the important

role of psychological distance in this process. A bad thing happening to you would be terrible,

but that same thing happening to an acquaintance can be funny because it is distant enough to

feel benign. Along these lines, Lazarus (1993) finds that psychologically distancing oneself

from a bloody accident reduces stress reactions such as heart rate and skin conductance.

Similarly, while emotional volatility may be stimulating whether it is happening to you or a main

character, distance should reduce the negative moment’s negative impact.

Consequently, though most people would prefer to have positive things happen to

themselves, emotional volatility may lead cultural products to be evaluated more favorably

because it makes them more stimulating and engaging.

THE CURRENT RESEARCH

Unfortunately, empirically testing the link between emotional volatility and cultural

success has been constrained by the ability to easily quantify volatility at scale. Measuring

emotional responses would usually require having people turn dials to rate their on-line

8

experiences (e.g., Ruef and Levenson 2007). But while possible for a small number of items,

applying such methods on the scale needed to truly study culture would be challenging.

To address this issue, we use natural language processing. We collect data on thousands

of movies, and use sentiment analysis to quantify the positivity or negativity of different parts.

Then, we calculate emotional volatility and examine whether more volatile movies are rated

more highly. While the results are supportive, one could wonder whether they are truly causal,

so we address this in several ways. First, we control for a range of other factors (e.g., overall

valence, peak, end, and emotional extremity) to test whether the relationship between volatility

and success persists. Second, we use moderation. If emotional volatility makes experiences

more stimulating and engaging, as we suggest, then volatility should be more beneficial in genres

where evaluations are more likely to be driven by how stimulating a movie seems (i.e., mysteries

and thrillers). We test this possibility. Third, to best identify emotional volatility’s effect, one

would ideally keep all other movie aspects the same, vary emotional volatility, and measure its

influence on success. To approximate this in the field data, we examine movies with sequels

(e.g., Harry Potter). While many focal actors and production team members remain the same,

emotional volatility varies across different movies in the series, providing a stricter test of

volatility’s impact.

Finally, to provide an even stronger test, we also conduct an experiment. We take the

same four chunks of a movie, manipulate their order, and measure evaluations. This allows us to

directly manipulate volatility and measure its impact. We also test and provide evidence for at

least one potential mechanism underlying volatility’s effects.

9

STUDY 1: EMPIRICAL ANALYSIS OF THOUSANDS OF MOVIES

To start, we conduct an empirical analysis of thousands of movies. We use natural

language processing to measure emotional volatility and predict that more volatile movies will

achieve higher ratings.

Method

First, we collected data on movies. We analyzed English subtitles in the

OpenSubtitles2013 corpus, which is a collection of movie subtitles gathered from

http://www.opensubtitles.org/ (Tiedemann 2012). Most movies were released between 1981 and

2013, and include everything from small indie films (e.g., The Marsh, Juno, and An Invisible

Sign) to big blockbusters (e.g., Star Wars). They span all genres (see Table S1 for distribution of

genre information), but the most frequent are dramas, comedies, romances, and thrillers. To

ensure similar movies are being compared, we ignored shorts (e.g., movies less than 30 minutes)

or those with very few words (i.e., less than 2000), leaving 4118 movies. To focus on text that

appeared as spoken dialogue, parenthetical indicators (e.g. [music], (laughter), and [gunshot])

were filtered out.

Second, we measure the sentiment of each word in the script. We build on work (Dodds

et al. 2011) which scored over 10,000 words based on how positive or negative they made people

feel on a nine-point scale. Words like laughter, happiness, and love were rated as highly positive

while words like terrorist, suicide, and murder were rated as highly negative. Following prior

work (Reagan et al. 2016), we focus on words with clear emotional content (i.e., ≥ 6 or ≤ 4).

10

Third, we calculate emotional volatility. We focus on volatility between sizable chunks

of a movie, like scenes, or portions of them. While volatility may also occur at a more granular

level (e.g., second-to-second), this is less likely to leave an enduring impression and more likely

to be measured with error. Further, if a movie repeatedly oscillated back and forth between

highly negative and positive in a matter of seconds, it might exhaust the viewer. Consequently,

we examine volatility on a larger scale, looking at how emotional variations across chunks of

dialogue (e.g. 1% of the movie) relates to success.

One challenge in constructing emotional trajectories is determining chunk length (i.e.,

how to break up the text). Unfortunately, there is no clear answer. While one could imagine

chunking movies by scene, scenes length varies greatly. Some movies have shorter scenes and

others have longer ones. Even within movies, some scenes are longer than others. Thus, using

scene boundaries would involve comparing apples and oranges as scenes would be of different

lengths, which could influence the degree of emotional volatility displayed. Further, in many

movies, emotional variation occurs within scenes, suggesting that scenes may not be the ideal

unit of analysis.

Consequently, our main analysis relies on the chunking strategy used in prior work

(Reagan et al. 2016), applying the same overlapping window to all parts of all movies. We

divide each movie into the same number of segments (i.e., 100) of consistent lengths (i.e., 500

words each) and average sentiment across the words within that segment. As shown below,

results are robust to a variety of number, size, and alternative approaches to segmentation.

Emotional volatility is measured by the standard deviation of differences in sentiment

between adjoining chunks of a movie. While the standard deviation of sentiment would be a

simpler measure, this ignores where in narratives different sentiment occurs, and thus does not

11

capture volatility. Regardless of whether two mildly positive periods are followed by two mildly

negative ones, or positive and negative periods alternate, the standard deviation would be the

same. But alternating positive and negative periods is more volatile. The standard deviation in

differences thus more accurately captures this up and down, and consequently, emotional

volatility.

Fourth, we measure cultural success. We recorded user ratings of each movie from

IMDB.com. Each movie is rated by many different people (Mean number of ratings = 55,709)

and ratings are on a 1-10 scale. We focused on this measure of cultural success, rather than say,

critics’ reviews, because it is more likely to be driven by individual preferences rather than a

small number of institutionalized actors.

Finally, an OLS regression examines the relationship between volatility and movie

success.

Results

Results indicate that more emotionally volatile movies receive higher ratings (b = 9.31,

s.e. = 0.78, p < 0.001, Table 1, model 1 and Figure 1). A 10.74% increase in volatility, for

example, is linked to a 1% increase in ratings.

Robustness Checks. We included numerous covariates in the model to assess the stability

of the main result and test alternative explanations. First, one might wonder whether longer

movies or movies from certain genres are both more emotionally volatile and positively

evaluated, and thus one of these factors, rather than volatility itself, is driving the observed

relationship. We test this possibility by controlling for movie length (using either number of

word or running time) and genre fixed effects.

12

Second, rather than volatility, one might wonder whether the mere presence of emotion

itself (i.e., some movies are more emotional) or valence (i.e., some movies are more positive or

negative) is driving the effect. To test this possibility, following prior work (Berger and

Milkman 2012) we control for emotionality by the proportion of affect-laden words in the script

using Linguistic Inquiry and Word Count (Pennebaker et al. 2015) and control for valence using

average sentiment across all segments.1

Third, given peaks and ends can shape evaluations of experiences (Kahneman et al.

1993), one could wonder whether emotionally variable movies have higher peaks or ends, and

those factors, rather than volatility itself, is driving the effect. To address this possibility, we

control for each movie’s peak emotion (using the difference between the global maximum and

mean of the overall emotional trajectory) and end (using the film’s last segment minus the

mean). Results are robust to different approaches to measuring peak (e.g. maximum minus

minimum) and end (e.g., the last segment, last few segments, or last minus minimum).

Fourth, maybe emotionally volatile movies are also more complex, and complexity is

driving success. To address this possibility, we control for complexity using Flesch-Kincaid level

(Kincaid et al. 1975) which measures the number of years of education required to understand a

text. We find the same results using an alternate measure of complexity (i.e., number of named

entities, or noun phrases that refer to specific individuals, places or organizations). Compared to

movies that have fewer characters, places, etc., movies that refer to more of those entities should

be perceived as more complex.

Fifth, maybe rather than short-term emotional volatility, our measure is picking up

extremeness, or how much a movie’s emotional trajectory diverges from the mean, and that is 1 While one might suggest using hedonometer (Dodds et al.2011) for this analysis, it is less ideal because it would require picking cut points. Hedonometer provides a more continuous measure of sentiment, but to determine whether certain movies are more or less emotional, we need a way to determine whether words are emotional or not rather than how positive or negative they are. Thus LIWC seemed more appropriate.

13

driving the effect. To address this possibility, we control for how much a movie’s emotional

trajectory diverges from the mean

Sixth, one might wonder whether recent movies are both more volatile and rated more

positively, and thus time explains the observed relationship. To address this possibility, we

control for time using year dummies. Results are the same using a continuous measure of time.

Even controlling for all these factors, however, the link between emotional volatility and

success persists (Table 1, model 2).

Table 1: Relationship between Emotional Volatility and Cultural Success

  (1) (2)

Emotional Volatility 9.31 *** 3.15 ***

(0.78) (0.90)Movie Length 0.01 ***

(0.00)Emotionality -0.01

(0.02)Avg. Valence -0.46 ***

(0.13)Peak 0.27

(0.23)End 0.03

(0.07)Complexity 0.01

(0.01)Extremeness -0.60

(0.55)Genre Dummies No YesTime Dummies No YesIntercept 5.58 *** 8.09 ***

(0.06) (0.96)Adjusted R2 0.0334 0.2429Note: *** p < .001, ** p < .01, * p < .05

14

Chunking Strategy. Results are also robust to different ways of dividing movies into

chunks. Whether using the same segment size but different number of segments (e.g., 500

words, 50 segments: b = 5.54, s.e. = 0.48, p < 0.001) or different segment sizes (e.g., 1000 words

and 50 segments: b = 10.79, s.e. = 0.92, p < 0.001 or 100 segments: b = 17.42, s.e. = 1.51, p <

0.001), results remain the same.

We also examined alternate ways of chunking, such as fixing the segment size as well as

the overlap between segments (e.g., 500 word segments with 100-word overlap), or fixing the

segment size but having no overlap between segments. The latter allows us to avoid any

concerns about forcing movies of different lengths into the same number of segments. We

control for movie length in these approaches, as some movies have more segments than others.

In all these approaches, however, results remain the same: More emotionally variable movies

receive higher ratings (500 word, 100 overlap: b = 1.64, s.e. = 0.32, p < 0.001; 500 word, 250

overlap: b = 2.48, s.e. = 0.47, p < 0.001; 1000 word, 900 overlap: b = 11.24, s.e. = 2.12, p <

0.001; 250 words, no overlap: b = 0.96, s.e. = 0.25, p < 0.001; 500 words, no overlap: b = 0.65,

s.e. = 0.22, p = 0.003; 750 words, no overlap: b = 0.73, s.e. = 0.22, p = 0.001).

Identifying the Effect. While these relationships are suggestive, to best identify emotional

volatility’s effect, one would ideally keep all other movie aspects the same, vary emotional

volatility, and measure its influence on success. To approximate this, we examine movies with

sequels (e.g., Harry Potter). While many focal actors and production team members remain the

same, emotional volatility varies across different movies in the series, providing a stricter test of

volatility’s impact. If among movies in a series, the ones that are more successful are the ones

that are more emotionally volatile, this would support the notion that volatility, rather than some

other factor, is driving success.

15

To test this possibility, we analyze the 175 movies that are part of a series (either the

original or a sequel) using an analysis approach similar to difference-in-differences. We

calculate the emotional volatility of each movie in a given series and compare its success relative

to others within that same series.

Underscoring the prior findings, even within a series, more emotionally volatile films are

evaluated more favorably (b = 8.86, s.e. = 4.32, p = 0.04). While one could argue that this is

driven by the original films being evaluated more positively and having greater emotional

volatility, this is not the case. Ignoring originals and just examining later films in a series (i.e.,

the 2nd vs. 3rd) shows the same relationship; even looking among sequels in the same series, those

sequels that are more emotionally volatile are more successful (b = 19.37, s.e. = 6.19, p = 0.04).

Variation by Genre. Further evidence for the notion that emotional volatility increases

ratings comes from examining how the relationship varies across genres (Figure 1). What make

a thriller emotionally engaging, for example, is different than what makes for a successful

romance. While thrillers should be more appealing the more stimulation they elicit, these aspects

should have less of an impact in genres like romance. Consequently, if emotional volatility is

truly shaping evaluations, as we suggest, it should have a stronger effect in genres where

stimulation matters more (e.g. thrillers rather than romances).

Consistent with this notion, the relationship was strongest in genres like thrillers (b =

12.50, s.e. = 1.52, p < 0.001) and mysteries (b = 12.93, s.e. = 2.41, p < 0.001) and weakest in

genres like music, documentaries, and romance (b = 2.55, s.e. = 3.48, p = 0.46, b = 3.40, s.e. =

2.23, p = 0.13, and b = 2.94, s.e. = 1.51, p = 0.05 respectively). Note, the effect still holds in

almost all genres (except those in which the standard errors are larger because there are fewer

16

films) but the fact that it is weaker in genres where stimulation is less likely to drive evaluations

provides further support for our hypotheses.

Figure 1: Impact of Emotional Volatility across Genres

Note: Coefficient estimates and 95% confidence intervals for all genres (first row) and for each individual genre (subsequent rows). The effect is significant for a genre if the confidence interval does not intersect with zero. Numbers in parentheses are the number of films in each genre. Most films are tagged with multiple genres.

Discussion

Natural language processing of thousands of movies provides preliminary evidence of the

value of emotional volatility. More emotionally volatile movies were evaluated more positively.

Ancillary analyses cast doubt on a number of alternative explanations. First, the

relationship between volatility and success relationship holds controlling for other features

shown to impact evaluations (i.e., average valence, peaks, and ends) and a variety of other

17

factors (i.e., genre, movie length, emotionality, complexity, extremity, and release year).

Second, the results are also robust to a range of other chunking approaches.

While it is often challenging to identify causal effects in field data, two additional

approaches are at least consistent with the notion that volatility boosts evaluations. First, even

within movies that are part of the same series, those that are more emotionally volatile are liked

more. Second, rather than being equally beneficial across genres, volatility has a larger impact

on evaluations in genres where stimulation is more desirable (e.g., thrillers rather than romance).

A few additional points are worth note. First, while one might suggest chunking movies

by scene breaks, as noted above this is less than ideal for several theoretical reasons. Further, at

a practical level, the subtitles data we used does not demarcate where scenes end and begin,

making it impossible to analyze at that level. We explored acquiring data on movie scripts,

while would include scene breaks, but such data is only available for a much smaller number of

movies.

Second, one could suggest analyzing the data using Fourier transform and power spectra.

While this approach is nice in some ways in that it breaks down waves into component

frequencies, there’s no real way to link those frequencies to underlying behavioral theory.

Third, one might why we examined on consumer rating rather than ticket sales. We

focused on this measure because it is more likely to be driven by individual preferences rather

than a smaller number of institutionalized actors (i.e., studio and theater executives). Ticket

sales depend on how many theaters show the movie, the size of the advertising budget, and

several other factors. Not only do we not observe all those factors, but they could potentially be

endogenous. If volatile movies tend to be shown in more theaters, do they sell more tickets

18

because they are more volatile or because they were more widely available? Consequently,

ratings are cleaner to examine in some ways.

Overall, the results of Study 1 are consistent with the notion that emotional volatility

shapes evaluations, but for a more direct test, in Study 2, we manipulate it directly.

STUDY 2: EXPERIMENTALLY MANIPUATING VOLATILITY

To more directly test the impact of emotional volatility on evaluations, we conducted an

experiment. We take the same four chunks of a movie, two positive and two negative, and by

manipulating their order, manipulate emotional volatility. Then, we measure the resulting

influence on evaluations. The scenes are completely independent (they don’t build on one

another), and the action is them is taking place simultaneously, so they could occur in any order.

By exposing participants to the exact same content, and just manipulating volatility, we can

examine its causal impact on evaluations. We predict that alternating positive and negative

scenes will make people like the movie more, and that this will be driven by emotional volatility.

While the main focus of Study 2 is to provide a causal test of emotional volatility, we

also measure at least one of the potential mechanisms that could contribute to the effect. As

noted in the introduction, one reason emotional volatility might increase evaluations is by

providing stimulation and engagement. To test this possibility, we measure how stimulating and

engaging the movies seemed and test whether it mediated the effect.

19

Method

Participants (N = 142) completed a film study as part of a larger group of studies for

compensation. They were told the experimenters were “interested in how people react to

different film scripts,” and that they would “read some scenes from a movies script that was

currently being written and respond to some questions based on them.” They were randomly

assigned to either a high or low volatility condition.

All participants read an overview of the movie (see Appendix for more detail). They were

told about some college students who loved a singer that was coming to perform at their school,

and volunteered to help the company putting on the show in exchange for tickets. At the last

minute, though, the company goes back on their word and won’t give the students the tickets, so

the students hatch a four-part plan to steal tickets from the box office. Each part of the plan is

independent, occurs simultaneously, and involves different pairs of the students, so they could

occur in any order.

Participants were then exposed to each of the four different scenes (see Appendix). Two

of the scenes (P1 and P2) were pretested to be positive (e.g., they successfully sneak into the box

office) and two (N1 and N2) were pretested to be negative (e.g., try to hack into the surveillance

system but leave a trail that makes them easy to find). 2

The only difference between conditions was the order in which the scenes occurred.

Participants were told that each scene was happening simultaneously, and the scenes were

explicitly designed so that they could be considered in any order and still make sense. In the low

volatility condition, same valence scenes were grouped together. Participants were either

exposed to two positive scenes followed by two negative ones, or two negative scenes followed 2 Participants (N = 156) were exposed to one of the scenes as part of a larger packet of studies, and rated how they felt afterwards (-3 = very negative, 3 = very positive). As expected, the two positive scenes evoked a positive mood (M = 0.54 and 1.00, compared to the scale midpoint t’s > 49, p’s < .001) and the two negative scenes evoked a negative mood (M = -0.78 and -1.01, , compared to the scale midpoint t’s > 29, p’s < .001)

20

by two positive. In the high volatility condition, however, participants were exposed to the same

four scenes, but positive and negative scenes were interspersed, to generate a more emotionally

volatile experience (i.e., negative, positive, negative, positive, or positive, negative, positive,

negative). The exact appearance of difference positive or negative scenes was fully randomized

across participants, so among participants who saw a positive scene first, for example, some saw

P1 first while others saw P2. Supporting the effectiveness of the manipulation, pretest data

confirmed that the movie was more emotionally volatile when it alternated between positive and

negative scenes.3

Next, we measured the hypothesized underlying process. Participants were asked how

engaging and how stimulating the plot was (1 = not at all, 7 = extremely, r = .88 averaged to

form a measure of stimulation and engagement).

Then, after reading a brief summary of the end of the movie (instead of stealing the

tickets they write the singer a letter, she gets them tickets, and all ends well), participants

completed the main dependent variable. They reported their attitude towards the movie on two

separate scales, how much they liked the movie (1 = didn’t like at all, 7 = liked a great deal) and

how interested they would be in watching the rest of it (1 = not at all interested and 7 =

extremely interested, averaged to form an index r = .81).

Finally, we tested some alternative explanations. First, to examine whether mood could

explain the results, participants rated the mood of the movie overall (1 = very positive, 7 = very

negative). Second, to test whether the emotionally volatile script was simply more natural or the

plot made more sense, participants rated “how natural was the order of the four scenes you read”

3 As part of a larger group of studies, participants (N = 258 from the same population as the main study) we randomly assigned to condition. After reading the script, instead of rating the main dependent variables, they were asked “Thinking back to the different scenes, how emotionally volatile was the movie? That is, how much did the emotion valence change between scenes?” As expected, alternating between positive and negative scenes made the movie seem more emotionally volatile (M = 4.73 vs. 4.32; F(1, 257) = 11.26, p = .01), confirming the effectiveness of the manipulation.

21

(1 = not at all, 7 = extremely) and “how much sense did the plot make” (1 = very little, 7 = a

great deal).

Results

First, as predicted, a one-way ANOVA found that scene order influenced movie

evaluations. Compared to the low volatility condition (M = 3.41), participants liked the movie

more in the high volatility condition where it alternated between positive and negative scenes (M

= 4.01; F(1, 141) = 4.39, p = .038).

Second, as expected, the manipulation also influenced engagement and stimulation.

Participants thought the movie was more stimulating and engaging when it was more

emotionally volatile (M = 4.87 vs. 4.23; F(1, 141) = 7.34, p = .008)

Third, as predicted, stimulation and engagement drove the impact of scene order on

evaluations [ab = .54, 95% CI .15 to .96]. Emotional volatility made the movie seem more

stimulating and engaging, which made people like it more.

Discussion

Building on Study 1, Study 2 provides experimental support for the results observed in

the field data. Reorganizing a movie to intersperse positive and negative chunks made people

like it more. Further, this was mediated by stimulation and engagement.

Alternative explanations. Ancillary analyses rule out a number of alternative

explanations. First, one could wonder whether the effects were simply driven by mood, but this

was not the case. There was no difference between conditions on mood (M = 5.04 vs. 4.95; F(1,

141) = 0.28, p > .6). Second, rather than making the movie more stimulating and engaging, one

22

could argue that emotional volatility increases liking because alternating positive and negative

scenes just seemed more natural, or made the plot make more sense. But ancillary data casts

doubt on these possibilities. There was no difference between conditions on either of these two

measures (natural: M = 4.85 vs. 4.66; F(1, 141) = .75, p = .387; plot: M = 5.22 vs. 5.20; F(1,

141) = 0.01, p = .94).

Alternatively, one might wonder whether emodiversity could explain the results. Some

work suggests that experiencing a diverse set of emotions is associated with more positive

mental and physical health (Quoidback et al 2014). Correlational evidence finds that people who

report experiencing a range of emotions also report being less depressed. One could wonder

whether something similar could be going on here. While possible, though, this seems unlikely,

as both experimental conditions are equivalent on emodiversity. Emodiversity looks at the

number of emotions people experience in a period of time, but is agnostic about when those

emotional experiences occur. So PNPN would have the same emodiversity as PPNN. This casts

doubt on the notion that emodiversity could explain the effects.

GENERAL DISCUSSION

Why are some cultural items more successful than others? This question has long vexed

academics and practitioners alike, with some concluding that success is random or driven by

patterns of social influence. While others have suggested that all stories can be clustered into a

small number of key narrative structures, empirical work in this area has been stymied by

measurement challenges. Further, while classifying narratives is interesting, it leaves open the

question of whether certain narrative structures actually lead to success.

23

We use natural language processing to begin to answer some of these questions.

Sentiment analysis of thousands of movies demonstrates that more emotionally volatile moves

are evaluated more positively (Study 1). This result is robust to including a variety of controls,

and holds even focusing on just movies in a series. Experimental evidence underscores

emotional volatility’s causal impact (Study 2). By directly manipulating volatility, and

measuring its impact on evaluations, we demonstrate it increases evaluations.

While our focus is on the effect of emotional volatility, additional results provide some

insight into the underlying process. Consistent with the notion that these effects may be driven

by stimulation, the relationship is stronger in genres where evaluations are more likely to be

driven by how stimulating a movie seems (e.g., thrillers rather than romances, Study 1). Further,

mediational evidence (Study 2) demonstrates that stimulation and engagement mediate the effect

of emotional volatility on evaluations. While this is not to say that other process don’t also play

a role, it provides at least some evidence for one potential mechanism.

Finally, by examining combining controlled experiments with field data, we can

rigorously test causality while also demonstrating external validity.

Contributions

This work makes a number of contributions. First, these findings contribute to the

burgeoning stream on literature on the psychological foundations of culture (Kashima 2014,

Kashima 2008, Schaller and Crandall 2004, Norenzayan et al. 2006). When shared across

individuals, psychological processes can shape the processing, sharing, and evaluation of cultural

items, which in turn, shapes their success. In this case, more emotionally volatile movies receive

higher ratings.

24

Second, the results shed further light on emotional dynamics. While average valence and

specific points (e.g., the end) are clearly important, this work suggests that how sentiment

evolves over the course of an experience is also important. As noted below, future work might

examine a number of other aspects of emotional trajectories and how they shape evaluations.

Third, the findings contribute to research on variety and hedonic adaptation. While

decades of variety research have examined the impact of varied consumption experiences (e.g.,

eating different foods or listening to different songs), there has been less attention to variation in

emotional experiences. Hopefully our research will open new avenues to study an old and

important topic. Similarly, while research demonstrates that inserting breaks (Nelson and Meyvis

2008) or other experiences (Nelson, Meyvis, and Galak 2009) between chunks of an experience

can stem hedonic adaptation, this work suggests that variation within an experience itself may

also provide benefits. Ads, for example, may make an enjoyable television show more

enjoyable, but even within the show itself, emotional volatility should shape evaluations.

Fourth, the research demonstrates the value of natural language processing to extract

behavioral insight from text. Researchers have long been interested in why some things succeed

and fail, but measurement has been a key challenge. Natural language processing, however,

provides a reliable method of extracting features, and doing so at scale (Humphreys and Wang

2017; Netzer, et al. 2012; Netzer, et al. 2018; Moore and McFerran 2017; Packard, et al. 2018;

Rocklage and Fazio, 2015; Rockladge et al., 2018). This emerging toolkit can hopefully shed

light not only on cultural success, but a range of interesting questions more generally.

Fifth, the findings have obvious implications for cultural producers. Boosting emotional

volatility (at least within a reasonable range) should increase evaluations. While we focused on

25

movies, similar effects may extend to other narrative domains (e.g., books and maybe even

songs).

Limitations

This investigation is not without limitations. First, we do not mean to suggest that the

words used by actors are the only features of movies that influence viewer experience. Auditory

(e.g., how words are said or music) and visual information (e.g., what is happening on screen)

certainly play a role. Automatically measuring these additional features, however, is not trivial.

Subtitle data is available for thousands of movies, but less data is available for auditory and

visual information. Further, extracting behavioral insight from this data is challenging. Software

packages like Praat (Boersma 2001) can be used to extract paralinguistic features (e.g., pitch and

tone) from social interactions, but there is less existing literature linking these features to

psychological processes. Advances in computer vision and image recognition have made it

possible to begin to automatically extract features from images, but doing this for video, at scale,

and linking it to psychological processes is still in its infancy.

That said, the fact that a measure constructed from words alone helps explain variation

suggests that the effect of emotional volatility could be even larger once other features are taken

into account. A more multifaceted measure should more accurately capture the full scope of

viewer experience, and better capture the underlying psychological construct of interest.

Second, as with many effects, the effect of emotional volatility is likely multiply

determined. While stimulation and engagement help explain emotional volatility’s impact on

evaluations, hedonic contrasts also likely plays a role. We like cultural products, in part, for their

ability to arouse emotion, but emotions become muted if they are experienced repeatedly (e.g.,

26

Carver and Scheier 1990; Frederick and Loewenstein 1999). While the first bite of a tasty

sandwich is delicious, we soon adapt to the pleasure, and the tenth bite is not as hedonically

positive. Volatility, however, may stem this reduction in subjective intensity. While five positive

scenes in a row may make the last scene feel less positive, alternating between more and less

positive periods, should intensify the experience. Future work might examine this additional

process more directly.

Future Research

Future work might examine other narrative features. We examined volatility, but there

are a number of other aspects of emotional trajectories that could be interesting to study.

Volatility focuses on a micro level, or period-to-period change, but more macro shifts, or

changes across a number of periods in a row, might also play a role. A number of famous

movies, for example, involve a seemingly similar structure, where before they reach a happy

ending, the characters have to overcome various trials and tribulations, or what can be thought of

as barriers to success. In both Star Wars and Harry Potter, the protagonist must overcome death

of their parents, they meet friends along the way and things look good, but then something bad

happens to them, and so on. In these, and other similar examples, the more macro, or broader,

emotional trajectory seems to follow a long wave-like pattern. Starting low and then slowly

building up to a high point over a before going down low again and building up again. One

could imagine that overcoming these lows make the highs more impactful. Said another way,

reaching the top of a mountain is more emotionally impactful if you hiked up from a valley

rather than got dropped off by a helicopter. Consequently, the most enjoyable narratives may

27

blend volatile moment to moment changes combined with larger aggregate waves of ups and

downs.

The distance between such peaks might also influence evaluations. Vertical change, from

negative to positive, may have an emotional impact, but the time period over which that change

occurs is likely also important. Put differently, part of the reason hiking up the mountain all the

way from the valley is so impactful is that it took a while to get there. Wins are savored more if

they took a while to develop. Consequently, beyond period to period emotional volatility, too

much aggregate change, too quickly, is likely not as positive.

The way peaks develop might also play a role. Levels of a video game often build on one

another. In the first level, the character has to overcome a small challenge. The next level, a

slightly larger challenge, and so on. Part of the reason is to engage a range of players early, and

let them practice and build their skills, but there may also be a narrative benefit. It makes sense

that the biggest challenge would be saved for last.

Applied to emotional trajectories, this suggests that slowly increasing extremity may

increase evaluations. At the beginning, both the lows and highs are small. But as the narrative

develops, they get larger and larger. That said, one could also make an argument for an alternate

structure. The largest valley near the beginning to set the stage, and smaller ones thereafter. It

would be interesting to empirically examine this more deeply.

Beyond emotional trajectories, other narrative dynamics also deserve further attention.

One such example is the speed of plot development. Some stories move quickly from plot point

to plot point while others take longer to develop. Screenplay writer Aaron Sorkin, for example,

is known for creating rapid fire narratives where something important happens almost every

other moment. Rather than one extreme being better than the other, the ideal point is likely in

28

between. Moving slowly enough to care about the characters and understand the flow, but

quickly enough to keep the viewer or reader engaged.

The value of these different trajectories, however, may vary across genres. People may

like thrillers more when the plot moves faster, but romantic comedies more when the plot moves

slower.

Conclusion

In conclusion, academics and practitioners alike have long been interested in why some

cultural products succeed while others fail. While there is certainly an art to writing an engaging

narrative, this work suggests that there may be some underlying science as well. Natural

language processing provides an exciting method for extracting features of narratives and doing

so at scale. Hopefully this method will help begin to unlock the mysteries of why some stories

are so powerful.

29

REFERENCES

Adler, Moshe (1985), “Stardom and Talent,” The American Economic Review, 75(1), 208-212.

Andrade, Eduardo B. and Joel B. Cohen (2007), “On the Consumption of Negative Feelings,”

Journal of Consumer Research, 34(3), 283-300.

Bass, Frank (1969), “A New Product Growth for Model Consumer Durables,” Management

Science, 15(5), 215–227.

Berger, Jonah (2011), “Arousal Increases Social Transmission of Information,” Psychological

Science, 22(7), 891-893.

Berger, Jonah and Katherine L. Milkman (2012), “What Makes Online Content Viral?” Journal

of Marketing Research, 49(2), 192–205.

Berscheid, Ellen and Hilary Ammazzalorso (2001), “Emotional Experience in Close

Relationships,” in Blackwell Handbook of Social Psychology: Interpersonal Processes,

Fletcher Garth J. O. and Clark Margaret S., eds, 308–330. NJ: John Wiley & Sons.

Bielby, William T. and Denise D. Bielby (1994), “All Hits are Flukes: Institutionalized Decision

Making and the Rhetoric of Network Prime-time Program Development,” American Journal

of Sociology, 99(5), 1287–1313.

Booker, Christopher (2004), The Seven Basic Plots: Why We Tell Stories. London: A&C Black.

Boyd, Robert and Peter J. Richerson (1985), Culture and the Evolutionary Process. IL:

University of Chicago Press.

Boersma, Paul (2002), “Praat, a System for Doing Phonetics by Computer,” Glot International,

5.

Carver, C. S., and Scheier , M. F. (1990). Origins and functions of positive and negative affect:

A control-process view. Psychological Review, 97, 19 – 35.

30

Cavalli-Sforza, Luigi Luca, and Marcus W. Feldman (1981), Cultural Transmission and

Evolution, Vol. 16. NJ: Princeton University Press.

Dodds, Peter Sheridan, Harris Kameron Decker, Kloumann Isabel M., Bliss Catherine A., and

Danforth Christopher M. (2011), “Temporal Patterns of Happiness and Information in a

Global Social Network: Hedonometrics and Twitter,” PLoS ONE, 6(12), e26752.

Eliashberg, Joshua, Sam K. Hui, and John J Zhang (2014). Assessing Box Office Performance

Using Movie Scripts: A Kernel-Based Approach. IEEE Transactions on Knowledge and Data

Engineering, 26, 2639-2648.

Etkin, Jordan and Cassie Mogilner (2016), “Does Variety Among Activities Increase

Happiness?” Journal of Consumer Research, 43(2), 210-229.

Fredrickson, Barbara L. and Daniel Kahneman (1993), “Duration Neglect in Retrospective

Evaluations of Affective Episodes,” Journal of Personality and Social psychology, 65(1), 45.

Frederick, Shane and George Loewenstein (1999), “Hedonic Adaptation,” in Well-Being: The

foundations of Hedonic Psychology. Daniel Kahneman, Edward Diener, and Norbert Schwarz,

eds. NY: Russell Sage, 302-329.

Golder, Scott A. and Michael W. Macy (2011), “Diurnal and Seasonal Mood Vary with Work,

Sleep, and Daylength Across Diverse Cultures,” Science, 333(6051), 1878–1881.

Hirsch, Paul M. (1972), “Processing Fads and Fashions: An Organization-Set Analysis of

Cultural Industry Systems,” American Journal of Sociology, 77(4), 639–659.

Humphreys, Ashlee and Rebecca Jen-Hui Wang (2018), Automated Text Analysis for Consumer

Research," Journal of Consumer Research, 44(6), 1274–1306.

31

Kahneman, Daniel, Fredrickson Barbara L., Schreiber Charles A., and Redelmeier Donald A.

(1993), “When More Pain Is Preferred to Less: Adding a Better End,” Psychological Science,

4(6), 401–405.

Kashima, Yoshihisa (2008), “A Social Psychology of Cultural Dynamics: Examining How

Cultures are Formed, Maintained, and Transformed,” Social and Personality Psychology

Compass, 2(1), 107–120.

Kashima, Yoshihisa (2014), “How Can You Capture Cultural Dynamics?” Frontiers in

Psychology, 5, 995.

Kincaid, J. Peter, Fishburne Jr. Robert P., Rogers Richard L., and Chissom Brad S. (1975),

“Derivation of New Readability Formulas (Automated Readability Index, Fog Count and

Flesch Reading Ease Formula) for Navy Enlisted Personnel,” Institute for Simulation and

Training.

Kuppens, Peter and Philippe Verduyn (2017), “Emotion Dynamics,” Current Opinion in

Psychology, 17, 22–26.

Lazarus, RS (1993), From Psychological Stress to the Emotions: A History of Changing

Outlooks, Annual Review of Psychology, 44, 1-21.

Vonnegut, Kurt (1999), Palm Sunday: An Autobiographical Collage. New York: Dial Press.

Markus, Hazel R. and Shinobu Kitayama (1991), “Culture and the Self: Implications for

Cognition, Emotion, and Motivation,” Psychological Review, 98(2), 224–253.

McAlister, Leigh and Edgar Pessemier (1982), “Variety Seeking Behavior: An Interdisciplinary

Review,” Journal of Consumer research, 9(3), 311-322.

McGraw, Peter A and Caleb Warren (2010), “Benign violations: Making immoral behavior

funny,” Psychological Science, 21, 1141-1149.

32

McGraw, Peter A, Caleb Warren, Lawrence Williams, and Bridget Leonard (2012), “Too close

for comfort, or too far to care? Finding humor in distant tragedies and close mishaps,”

Psychological Science, 25, 1215 - 1223.

Mehl, Matthias R., Vazire Simine, Ramírez-Esparza Nairán, Slatcher Richard B., and

Pennebaker James W. (2007), “Are Women Really More Talkative Than Men?” Science,

317(5834), 82.

Mellers, Barbara A., Schwartz Alan, Ho Katty, and Ritov Ilana (1997), “Decision Affect Theory:

Emotional Reactions to the Outcomes of Risky Options,” Psychological Science, 8(6), 423–

429.

Moore, Sarah G. and Brent McFerran (2017), “She Said, She Said: Differential Interpersonal

Similarities Predict Unique Linguistic Mimicry in Online Word of Mouth,” Journal of the

Association for Consumer Research: The Connected Consumer, 2 (2).

Norenzayan, Ara, Atran Scott, Faulkner Jason, and Schaller Mark (2006), “Memory and

Mystery: The Cultural Selection of Minimally Counterintuitive Narratives,” Cognitive

Science, 30(3), 531–553.

Nelson, Leif D. and Tom Meyvis (2008), “Interrupted Consumption: Disrupting Adaptation to

Hedonic Experiences,” Journal of Marketing Research, 45(6), 654-664.

Nelson, Leif D., Tom Meyvis, and Jeff Galak (2009), “Enhancing the Television-viewing

Experience Through Commercial Interruptions,” Journal of Consumer Research 36(2), 160-

172.

Netzer, Oded, Ronen Feldman, Jacob Goldenberg and Moshe Fresko (2012), "Mine Your Own

Business: Market Structure Surveillance through Text Mining." Marketing Science, 31 (3),

521-543

33

Netzer, Oded, Alain Lemaire, and Michal Herzenstein (2018), “When Words Sweat: Identifying

Signals for Loan Default in the Text of Loan Applications,” Working Paper.

Packard, Grant, Sarah Moore, and Brent McFerran (2018), “(I’m) Happy to Help (You): The

Impact of Personal Pronoun Use in Customer-Firm Interactions,” Journal of Marketing

Research, 55(4), 541-555.

Pessemier, Edgar and Moshe Handelsman (1984), “Temporal Variety in Consumer Behavior,”

Journal of Marketing Research, 435-444.

Pennebaker, James W., Boyd Ryan L., Jordan Kayla, and Blackburn Kate (2015), “The

Development and Psychometric Properties of LIWC2015,” TX: University of Texas at Austin.

Pennebaker, James W., Mehl Matthias R., and Niederhoffer Kate G. (2003), “Psychological

Aspects of Natural Language Use: Our Words, Our Selves,” Annual Review of Psychology

54(1), 547–577.

Polti, Georges (1921), The Thirty-Six Dramatic Situations. OH: J.K. Reeve.

Quoidbach, Jordi, Gruber June, Mikolajczak Moïra, Kogan Alexsandr, Kotsou Ilios, and Norton

Michael I. (2014), “Emodiversity and the Emotional Ecosystem,” Journal of Experimental

Psychology: General, 143(6), 2057.

Ratner, Rebecca K., Barbara E. Kahn, and Daniel Kahneman (1999), “Choosing Less-Preferred

Experiences for the Sake of Variety,” Journal of Consumer Research, 26 (June), 1–15.

Reagan, Andrew J., Mitchell Lewis, Kiley Dilan, Danforth Christopher M., and Dodds Peter

Sheridan (2016), “The Emotional Arcs of Stories are Dominated by Six Basic Shapes,” EPJ

Data Science, 5(1), 31.

Redelmeier, Donald A., Joel Katz, and Daniel Kahneman (2003), “Memories of Colonoscopy: A

Randomized Trial,” Pain, 104(1-2), 187-194.

34

Reynolds, Jenny, Catherine Houlston, Lester Coleman, Gordon Harold (2014), Parental

Conflict: Outcomes and Interventions for Children and Families. Bristol University Press.

Rocklage, Matthew D. and Russell H. Fazio (2015), “The Evaluative Lexicon: Adjective Use as

a Means of Assessing and Distinguishing Attitude Valence, Extremity, and Emotionality,”

Journal of Experimental Social Psychology, 56, 214–27.

Rocklage, Matthew D., Derek D. Rucker, and Loran F. Nordgren (2018), “Persuasion, Emotion,

and Language: The Intent to Persuade Transforms Language via Emotionality,” Psychological

Science, 29, 749–60.

Rogers, Everett M. (1995), Diffusion of Innovations. New York: The Free Press.

Rolls, Barbara J., Edmund T. Rolls, Edward A. Rowe, and Kevin Sweeney (1981), “Sensory

Specific Satiety in Man,” Physiology & Behavior, 27 (July), 137–42.

Rolls, Barbara J., P. M. van Duijvenvoorde and Edmund T. Rolls (1984), “Pleasantness Changes

and Food Intake in a Varied Four-Course Meal,” Appetite, 5 (4), 337–48.

Ruef, Anna Marie and Levenson Robert W. (2007), “Continuous Measurement of Emotion: The

Affect Rating Dial,” in Handbook of Emotion Elicitation and Assessment, James A. Coan and

John J. B. Allen, eds. New York: Oxford University Press, 286–297.

Rule, Alix, Jean-Philippe Cointet, and Peter S. Bearman (2015), “Lexical Shifts, Substantive

Changes, and Continuity in State of the Union Discourse, 1790-2014,” Proceedings of the

National Academy of Sciences, 112(35), 10837–10844.

Salganik, Matthew J., Peter Sheridan Dodds, and Duncan J. Watts (2006), “Experimental Study

of Inequality and Unpredictability in an Artificial Cultural Market,” Science 311(5762), 854–

856.

35

Schaller, Mark and Crandall Christian S. (2004), The Psychological Foundations of Culture. NJ:

Psychology Press.

Scherer, Klaus R. and James S. Oshinsky (1977), “Cue Utilization in Emotion Attribution from

Auditory Stimuli,” Motivation and Emotion, 1(4), 331–346.

Simonton, Dean K. (1980), “Thematic Fame, Melodic Originality, and Musical Zeitgeist: A

Biographical and Transhistorical Content Analysis,” Journal of Personality and Social

Psychology, 38(6), 972–983.

Tiedemann, Jörg (2012), “Parallel Data, Tools and Interfaces in OPUS,” Proceedings of the 8th

International Conference on Language Resources and Evaluation, Turkey, (May 21-27).

Tobias, Ronald (1993), 20 Master Plots and How to Build Them. OH: Writers Digest Books.

36

Appendix

BackgroundCollege students Elly, Darell, Ron, Andy, Nick, Sarah and Clarissa are close friends, who are all huge fans of the singer Luna. This year, Acrala entertainment (to which Luna belongs) announced Luna would perform for their college music festival. Excited, Elly and her friends decided to help the entertainment company set up everything - they were promised front row tickets, so they agreed to work for no pay. At the very last moment, however, the company went back on their promise, saying that the initial agreement was void since the one who hired them, Lizzie, no longer worked for them. Elly and her friends protested, but the company fired all of them. Furious, they decided to steal the tickets they deserve. They hatched up a four-part plan to steal tickets from the box office, located in the art center where the concert is to be held.

Elly and Darell will sneak into the college administration building (separate from the art center) and steal the keys to the box office.

Nick will persuade the security guard at the art center to turn a blind eye when they sneak into the art center.

Andy and Ron will hack into the surveillance system and disable the cameras in the art center. Clarissa and Sarah will call the new project manager one last time to give her a last chance to give them

the tickets.They decided to steal the tickets on Tuesday. It is Monday night, and the friends are carrying out each of their part of the plan simultaneously.

Positive Scene #1ADMINISTRATION BUILDING - HALLWAY - NIGHTFlashlight dashes across the walls, followed by QUIET APPROACHING FOOTSTEPS from around the corner. ELLY AVERILL and DARELL HOWLAND, dressed in dark clothing, ENTER FRAME, walking slowly. DARELL: Are you sure the keys are still here? ELLY: For the hundredth time, yes, I'm sure. Unless the new kid is crazy paranoid and is expecting ex-employees to break in in the middle of the night, I'm sure they keep the keys where I used to… ELLY flashes the light to a sign that says "ROOM 456". ELLY: Here. My old office. ELLY pulls the doorknob, but the door is locked. ELLY then shakes the doorknob until something clicks, and pulls it towards herself. The door pops open. DARELL looks at ELLY, surprised. ELLY: I mean, if there's one door that should actually lock on campus, it should be the one leading to the room that has all the keys, right? DARELL laughs. BOTH walk into room 456.

ADMINISTRATION BUILDING - ROOM 456ELLY flashes the light to a wall where dozens of keys are hanging. ELLY: We're looking for a yellow key with a small "T" inscribed on it. The handle part looks kinda like a flower… but you know what? Just show me all the yellow –

37

DARELL: This one? ELLY points the flashlight to DARELL, who is holding a key with a giant card attached to it that says "BOX OFFICE". ELLY: Uh, yeah, that one. DARELL: Huh. I guess the new kid got tired of explaining the shape every time. ELLY smiles, and nods her head towards the door. ELLY: Well that was easy, now let's get out of here! BOTH hurriedly EXIT the room.

38

Positive Scene #2ART CENTER – SECURITY STATION - NIGHTNICK HOPKINS gets out of his car, walks up to the security booth, where CHARLIE BOWMAN, a friendly security guard in his mid 50s, is working his night shift. NICK: Hey Charlie, how’s it going? CHARLIE (adjusting his glasses): Oh hey there, Nick! I’m doing well, thanks. What brings you here at night? NICK: Oh, I was just around the area, thought I’d drop by and say hi to a friend. CHARLIE (smiling): Well that’s sweet of you. What were you doing around here? NICK: Uh, you know, things. Art things. CHARLIE (chuckles): Nick, why are you here? NICK looks left and right, walks a few steps towards the booth, and leans towards Charlie. NICK (whispers): I need your help. CHARLIE: Well, let’s hear it out. From a FARTHER ANGLE, we see Nick explain to Charlie him and his friend’s situation. We then come back to a CLOSE SHOT of the two. NICK (CONT’D): So yeah, do you think you could, I don’t know, “doze off” tomorrow around this time for us? CHARLIE taps his fingers on the table for a while. He looks at the ceiling for a moment, then looks at NICK. CHARLIE: You know what? You and your friends are good kids. And those people can’t do what they did to you guys. That’s just plain wrong. CHARLIE leans back on his chair. CHARLIE (CONT’D): So maybe I’ll be real mad about them tonight. So mad I’ll get very little sleep. So little sleep I might end up dozing off for a minute or two tomorrow. CHARLIE looks at NICK and smiles. NICK (smiling): Thanks so much, Charlie! You’re literally the best. CHARLIE: No worries, kid. NICK runs to his car, excited. He stops, turns around, sticks his thumb up to CHARLIE, and continues to run.

39

Negative Scene #1RON AND ANDY’S DORM ROOM - NIGHTRON GARNER and ANDY CLARKSON are sitting in front of a laptop, lights off, the only light coming from the laptop screen. ANDY types on the laptop furiously. RON: Dude, I’m still not sure why I have to be here. Like, you can do whatever you want with my laptop, and I just don't know anything about hacking, so - ANDY: Ron, you’re my rubber duck. I just need somebody to talk to while I think. RON: I mean, okay, I guess. RON opens a bag of potato chips and starts eating. ANDY keeps typing on the laptop. ANDY: I’ve been telling the team to set up their firewall for forever, but no, nobody has time for that. So yeah, let’s shut down these cameras - ANDY suddenly freezes. His eyes widen, fixated on the screen. RON leans in towards the screen, and looks at ANDY. RON (with chips in his mouth): Andy? ANDY doesn’t respond. RON (tapping Andy’s shoulder): You okay, buddy? No response. RON (looking at screen): What’s wrong? ANDY points to the VPN icon on the browser. It’s gray, indicating that it’s off. RON: Oh, it’s the thing you downloaded for me. Is that bad? ANDY: It’s off. RON: So, bad. ANDY: It means that we were using our own IP address this whole time. RON: Oh. ANDY: It means that everything we were doing can be tracked back to this laptop. RON: Oh no. RON looks at ANDY’s face, and looks back at the screen. RON (CONT’D): Oh no.

40

Negative Scene #2CLARISSA’S ROOM – NIGHT CLARISSA PEYTON is sitting on her bed, staring at her phone. SARAH LINTON is sitting on CLARISSA’s desk chair. SHOT of CLARISSA’s phone screen: she has already put in the number for RACHEL PARKS, the project manager of Acrala entertainment. CLARISSA takes a deep breath, presses the call button, and rests the phone on her ear. A few seconds later, RACHEL picks up. RACHEL: My god Clarissa, what is it now? Is this about the tickets again? CLARISSA: Well, first of all, hi. And actually, yeah - RACHEL: Oh god. See, I never promised you those tickets. That was Lizzie. She doesn’t work here anymore. What part of this do you not understand? CLARISSA clenches her jaws. SARAH mouths the words “what’s wrong”, concerned. CLARISSA: Okay, yeah, we know that she doesn’t work there anymore. But we made a deal with the company, which - RACHEL (condescendingly): Oh honey, that’s why you need to get everything down in writing next time, okay? Otherwise I just can’t care, even if I wanted to. And frankly, I don’t want to either. CLARISSA closes her eyes, brows lowered, visibly angry. CLARISSA (angrily): You know what? My friends and I have been civil with you this whole time, hoping you’d have the common decency to reciprocate. But no, we overestimated you. You couldn’t be a decent human being if you tried, could you? SARAH’s eyes widen, and she waves her hands to tell CLARISSA to stop. RACHEL: What did you just say? You listen – CLARISSA (voice growing louder): No, you listen for once. You think you can do anything you want. You think we can’t fight back. You know what? Forget it! We don't need you to give us the tickets, we’ll take them from you! SARAH snatches the phone from CLARISSA. RACHEL (from the phone, muffled): What? Oh no, if you kids do anything – SARAH hangs up. SARAH looks at CLARISSA, who’s breathing heavily, still visibly angry. CLARISSA drops her head and starts sobbing. SARAH sits next to her and pats her

41