DHQ: Digital Humanities Quarterly
2020
Volume 14 Number 2
Volume 14 Number 2
Tracking the Consumption Junction: Temporal Dependencies between Articles and Advertisements in Dutch Newspapers
Abstract
Historians have regularly debated whether advertisements can be used as a viable source to study the past. One of their main concerns centered on the question of agency. Were advertisements a reflection of historical events and societal debates, or were ad makers instrumental in shaping society and the ways people interacted with consumer goods? Using techniques from econometrics (Granger causality test) and complexity science (Adaptive Fractal Analysis), this paper analyzes to what extent advertisements shaped or reflected society. We found evidence that indicates a fundamental difference between the dynamic behavior of word use in articles and advertisements published in a century of Dutch newspapers. Articles exhibit persistent trends. Contrary to this, advertisements have a more irregular behavior characterized by short bursts and fast decay, which, in part, mirrors the dynamic through which advertisers introduced terms into public discourse. On the issue of whether advertisements shaped or reflected society, we found particular product types that seemed to be collectively driven by a Granger causality going from advertisements to articles. Generally, we found support for a complex interaction pattern, analogous to Cowan’s concept of the consumption junction. Finally, we discovered noteworthy patterns in terms of Granger causality and long-range dependencies for specific product groups. All in, this study shows how methods from econometrics and complexity science can be applied to humanities data to improve our understanding of complex cultural-historical phenomena such as the role of advertising in society.
Introduction
Advertisements as a Lens on the Past
Over the course of the twentieth century, branded consumer goods turned into an
integral part of society [Cross 2000]
[de Grazia 2005]. Consequently, researchers turned to consumer
goods as an object of research. In addition to writing histories of particular
consumer goods, researchers have also conceptualized consumer goods as entry
points into broader cultural phenomena. Scholars have, for instance, studied
how
advertisements represented consumerism, gender identities, and the globalization
of food cultures (see Lears, 1994; Parkin, 2007; Sivulka, 2012). In these studies, advertisements functioned as a
lens on the past. Roland Marchand argues that advertisements provide an insight
into the ideals and aspirations of past realities. He argues that they show
the
state of technology, the social functions of products, and that they provide
information on the society in which a product was sold. Furthermore, Marchand
poses that advertisements contributed to the shaping of a “community of discourse”. He claims that advertisements infused
public discourse with a particular type of language [Marchand 1985].
At the same time, scholars have debated to what extent adverts actually offer a
meaningful depiction of the past. In their effort to sell more products,
producers, and ad makers amplified or distorted certain social and cultural
aspects to make products more appealing to consumers [Brandt 2004]. Erving Goffman points out that ads shaped consumer’s lived experience by
prescribed certain conceptions of identity [Goffman 1985]. Put
differently, ad makers used adverts to shape society by tweaking
the desires of consumers [Fox 1997]. This form of social
engineering was rooted in psychological and sociological theories of the day.
Especially the tobacco industry was actively involved in developing new ways
of
advertising that could entice people to turn to smoking cigarettes [Brandt 2004].
These theoretical debates raise the question whether we can actually use
advertisements to study the past, or are we merely studying a version of the
past constructed by ad makers? Theories on the relationship between
advertisements and society can be summarized by three positions. The first
position contends that advertisements reflected the desires and aspirations
of
consumers. The second argues that advertisements merely represented the interest
of advertisers and the companies that produced the commodities. This approach
attributes more agency to the advertisers. The third approach proposes the
existence of a more complex consumption junction, in which
producers, distributors, consumers, and advertisers collectively negotiated
the
meaning and success of a consumer product [Cowan 1987].
This article examines the validity of these three positions in a specific
historical context (i.e. twentieth-century Netherlands) and sets out to answer
to what extent did advertisements reflect or shape society. We studied this
interplay between advertisers and society by analyzing advertisements and
articles in newspapers. Newspapers are a well-read rich historical source that
contains “conscious representations of
conditions and events” as well as “unconscious reflection[s] of the tastes, the interests,
the desires, and the spirit of its day”
[Smail 2008, 48]. As such, newspapers function as a proxy for
public discourse [Marshall 1995]
[Schudson 1982]. Newspapers contained a large number of
advertisements. Even though advertisements were a major source of revenue for
newspaper publishers, the editors writing the news were generally separated
from
the people selecting the advertisements [Rooij 1974]
[Schreurs 2001]. For this reason, we argue that we can view the
articles and advertisements in the same source as separate sources of
information. The possible relationships between the two are, therefore, more
likely to be reflective of relations existing in the real world
rather than reflective of choices made by editors. The availability of digitized
newspapers enables researchers to use computation to explore the archive, locate
particular instances of language use, or extract specific linguistic patterns
[Wevers and Verhoef 2017].
This study combines techniques from econometrics and complexity science to examine
the dynamics of word use in articles and advertisements. Complexity science
has
been used successfully in the past to model large-scale dynamics of human
language, increasing our understanding of social systems and collective dynamics
on the macro-scale [Helbing 2015]
[Petersen et al. 2012]. We set out to answer the following three
questions. First, did advertisements shape or reflect newspaper discourse?
Second, did word use in ads differ in dynamics from articles? And finally, were
these characteristics more pronounced for particular product groups? Answering
these questions will further our understanding of the dynamics underpinning
a
complex cultural-historical phenomenon, such as advertising.
Beyond Counting Word Frequencies
With the proliferation of large databases that hold temporally-dispersed text
content, time series plots of word frequencies have become a valuable source
of
exploration and validation of cultural trends [Gao 2012]
[Michel et al. 2011]. Google has popularized this approach through their
web-based n-gram viewer based on its digitized book collection.[1] The popularity of the
Google Ngram viewer has sparked other digital archives to also develop n-gram
viewers based on their collection.[2]
N-gram viewers can help researchers to determine when words appeared and how they
evolved. Plots of word frequencies, however, only offer an overview of
particular trends in discourse. To gain more insight into the trends in and
between advertisements and articles, we applied two techniques from econometrics
and complexity science to the (relative) frequencies of words. First, we applied
the Granger causality test to determine whether trends in word use in ads
followed trends in articles or vice versa. Second, we analyzed the persistence
of words using fractal analysis to identify whether ads differed from articles
in terms of dynamics related to word use. Put differently, we determined whether
advertising discourse was distinct in its behavior from discourse in articles.
Also, we set out to identify specific words that stuck with people?
This stickiness can be indicative of social dynamics on a macro-scale. Can we
detect behavior of words that transcends individual uses and is more reflective
of the existence of “memory” in communication?
Memory in a time series is modeled as the presence of self-similarity, or more
precisely persistent correlation, between the values of these features at
various time steps. Merely glancing at visualizations produced by n-gram viewers
might show sudden peaks or slow decays in word use, which might suggest dynamic
processes reflective of memory. In this paper, however, we quantify such
behavior and show the extent to which word use exhibits particular memory
functions exhibited in cultural expressions. Jan Assman has described cultural
memory as “a collective concept for all knowledge that
directs behavior and experience in the interactive framework of a society
and one that obtains through generations in repeated societal practice and
initiation”
[Assman 2011]
[Assman and Czaplicka 1995]
[Donald 2002]. One of these repeated societal practices is
language use, which also shapes our collective understanding of a shared
culture. According to Assman, cultural memory is formed over large periods of
time whereas communicative memory represents memories shaped over shorter time
spans (80-100 years). In a way, communicative memory can be viewed as the
short-term memory of a society [Assman 2008]
[Assman and Czaplicka 1995]. Can we use time series of word frequencies in
newspapers to detect communicative memory?
Methods
Data
The National Library of the Netherlands (KB) has digitized thousands of Dutch
historical newspapers using optical character recognition (OCR)
software.[3]
This software turns scans of physical pages into machine-readable data.
Unfortunately, the text extracted from the digital scans is often flawed due
to
imperfections in the original material or limitations of the recognition
software. These material blemishes cause the software to not recognize and
transcribe every word correctly, which has resulted in conjoined words, complete
gibberish, or words in which certain characters were replaced. The age and
quality of the original material are important determinants of the software’s
ability to correctly recognize the text; hence, older newspapers contain many
more errors than more recent papers. For this reason, we focused on
twentieth-century newspapers. Also, the KB does not provide suitable metrics
on
the quality of the OCR’ed text [Traub et al. 2015]. The study, therefore,
assumes that OCR errors are uniformly distributed over the period.
For analysis, we relied on two subsets of the digitized newspaper data. The first
subset consisted of the entire set of advertisements (
n
1
=
18,564,411
) in the KB’s digitized newspaper archive. The second
set held newspaper articles (
n
2
=
11,465,220
) from two national newspapers: De
Tijd (1890-1974) and De Telegraaf 1893-1989.
During digitization, the OCR software separated articles from advertisements
and
stored the document type in the metadata, allowing us to select these two types
of documents. Because advertisements made up a smaller portion of the
newspapers, we selected the entire set of advertisements to make it more
comparable to the corpus of articles in terms of size. Also, we narrowed our
focus to two national newspapers because these are more likely to represent
wider public discourse than regional newspapers.
We calculated keyword frequencies, more specifically, normalized relative daily
term frequency per document for these two subsets. We explicitly looked at 265
words (singular and plural forms collapsed) that denoted consumer products.
Based on exploratory data analysis using an n-gram viewer, we selected words
that appeared in both advertisements and articles throughout greater periods
of
time with considerable frequency. Brand names were excluded for two reasons.
First, brand names often appeared as part of logos in advertisements, making
it
more difficult to convert these images to machine-readable text. Second, the
techniques used in this paper necessitate the existence of the same words over
longer periods of time. More often than not, there existed multiple brand names
for the same products, which were also not used over longer periods of time.
Causal Dependencies
Several techniques can be used to compare lagged values of time series
X
with values of a second time series
Y
to model variation in their correlation coefficient
as a function of temporal displacement. The most widely used technique is
cross-correlation, which is simply used to detect the variation in the
correlation between two time series as a function of lag. The Granger causality
test goes beyond mere correlation and tests for the existence of causal-like
dependencies between temporally disjunctive time series of, for instance, words
from two sources [Granger 1969]. The test, which originated in
econometrics, is based on the assumption that causality is more than temporal
disjunction, it involves directionality between time series. The relation tested
by the Granger causality test is often characterized as predictive causality
and
represented as
X
G
r
a
n
g
e
r
c
a
u
s
e
Y
to distinguish it from more direct causality [Sugihara et al. 2012]. At its core, Granger causality, which is related
to correlation, expresses if values of time series
X
contain information that is uniquely predictive of
subsequent values in time series
Y
.
For our study, we used the Granger causality test as follows. To identify a shaping relation, we test if variation in a specific
word frequency for newspaper discourse (
Y
) at time
t
is predicted by variation in the frequency for the
same word in advertisement discourse (
X
) at earlier time steps
t
-
1
…
t
-
k
. We test for
X
G
r
a
n
g
e
r
c
a
u
s
e
Y
, by comparing the performance of the ‘newspaper
discourse only’ model:
with the full newspaper and advertisement discourses
model:
to identify which one does the better job at explaining word frequency (
y
t
) based on the residuals. The zero-model for the
hypothesis is
H
0
:
α
i
=
0
for each
i
of the element
[
1
,
m
]
with the alternative hypothesis being
H
1
:
α
i
≠
0
for at least one
i
of the element
[
1
,
m
]
. We applied the test bi-directionally such that a
shaping relation finds support if we can confirm that ‘
X
G
r
a
n
g
e
r
c
a
u
s
e
Y
’, and in the case of a reflecting relationship we can
reject ‘
Y
G
r
a
n
g
e
r
c
a
u
s
e
X
’. Finally, if both ‘
X
G
r
a
n
g
e
r
c
a
u
s
e
Y
’ and ‘
Y
G
r
a
n
g
e
r
c
a
u
s
e
X
’ find support, we viewed this as indicative of a more
complex relationship between the two time series.
Long-range Dependencies
In addition to the Granger Causality test, we used fractal analysis to identify if
words exhibited long-range dependencies.[4] Fractal Analysis is a method to assess the
complexity of data. In our case, we look at long-range dependencies, which
indicate a rate of decay between two points with increasing time intervals that
is slower than exponential decay. Analysis of time-dependent change in complex
systems — systems composed of many interacting elements — is an important
application of Fractal Analysis.
Some random processes in complex systems are self-affine, that is, fluctuation
patterns at shorter time scales are (statistically) similar to fluctuations
at
longer time scales. In the case of reading, for instance, fluctuations in
reading speed are self-affine across multiple time scales, because both reading
fluency and comprehension are affected by elements at short time scales (e.g.,
words and sentences) and longer times scales (e.g., paragraphs and chapters)
[O'Brien et al. 2014]. Such fractal behavior is also found in a range
of processes related to psychology [Chater and Brown 1999], economy [Marchant 2008], sociology [Gao 2017], health [Eke et al. 2002], language [Gao 2012] and music [Voss and Clarke 1975]. We argue, therefore, that fractal analysis has great
potential for the study historical trends in cultural expressions [Nielbo et al. 2019]. This is particularly the case when we are dealing
with “big data”, consisting of large sets of mostly unknown parameters [Gao 2012].
We are interested in a particular kind of fractal processes called
1
/
f
2
H
+
1
processes, in which
H
refers to the Hurst exponent. The Hurst exponent
quantifies the degree of long-range dependencies in a time series (Figure 1), such that when
0
<
H
<
0.5
, the time series is an anti-persistent process (i.e.,
a jump up is followed by a jump down, or vice versa, in the incremental
process). When
H
=
0.5
, the time series only has short-range dependencies,
and when
0.5
<
H
<
1
, the time series is characterized by long-range
dependencies (i.e. a jump up is followed by another jump up, or vice versa,
in
the incremental process). It is possible the
H
>
1
indicates a non-stationary process. In this study,
persistence represents whether a word ‘stuck’ with people and it is in that
manner analogous to how scholars have viewed communicative memory.
Figure 1.
Figure 1: Left: Time series that exhibit anti-persistent (top),
short-range (middle), and long-range (bottom) dependencies. Anti-persistent
time series oscillate rapidly around its average, which is sometimes
referred to as mean-reverting or rigid behavior. Short-range dependencies
are indicated by the short cycles, while long-range dependencies show
repetitive cycles at multiple time scales. Right: Estimation of the Hurst
exponent as the slope of the residual fit F(w) on the time
window w for the matching time series in the left column.
Anti-persistent time series have a slope <0.5, the slope for short-range
dependencies is 0.5, and long-range dependencies >0.5.Adaptive Fractal Analysis
Adaptive Fractal Analysis (AFA) is a relatively new technique for determining
the Hurst exponent of a time series [Gao et al. 2011]
[Riley et al. 2012]. AFA improves the popular detrended fluctuation
analysis [Peng et al. 1994] by identifying a global smoothed trend
that can automatically deal with arbitrary, strong nonlinear trends (Gao et
al., 2011). The technique is based on a nonlinear adaptive multi-scale
decomposition algorithm [Gao et al. 2011].
After constructing a random walk process from the time series, the initial
step of AFA involves partitioning the time series into overlapping segments
of length
w
=
2
n
+
1
, in which neighboring segments overlap by
n
+
1
points. In each segment, the time series is
fitted with the best polynomial of order
M
, obtained using standard least-squares
regression. The fitted polynomials in overlapped regions are then combined
to yield a single global smoothed trend. Denoting the fitted polynomials
for the
i
-
t
h
and
(
i
+
1
)
-
t
h
segments by
y
i
(
l
1
)
and
y
(
i
+
1
)
(
l
2
)
, respectively, where
l
1
,
l
2
=
1
,
⋯
,
2
n
+
1
, we define the fitting for the overlapped region
as
The global trend can be used to maximally suppress the effect of complex
nonlinear trends on the scaling analysis. The parameters of each local fit
are determined by maximizing the goodness of fit in each segment. The
different polynomials in the overlapped part of each segment are combined
such that the global fit will be the smoothest fit of the overall time
series. Even if the local fits are linear,
M
=
1
, the global trend signal will still be
nonlinear. AFA then can be described accordingly: for an arbitrary window
size
w
, we determine, for the random walk process
u
(
i
)
, a global trend
v
(
i
)
,
i
=
1,2
,
⋯
,
N
, where
N
is the length of the walk. The residual of the
fit,
u
(
i
)
-
v
(
i
)
, characterizes fluctuations around the global
trend, and its variance yields the Hurst exponent
H
according to the following scaling equation:
By computing the global fits, the residual, and the variance between original
random walk process and the fitted trend for each window size
w
, we can plot
log
2
F
(
w
)
as a function of
log
2
w
. The presence of fractal scaling amounts to a
linear relation in the plot, with the slope of the relation providing an
estimate of
H
(see Figure 1).
Design
To determine whether advertisements reflected or shaped public discourse, we first
applied Granger causality tests to each of the 265 keywords, comparing time
series from newspaper and advertisement discourse. We hypothesize the existence
of the following three Granger causal-like patterns:
- Advertisements shaped newspaper articles as expressed by Granger causality directed exclusively from advertisements to articles;
- Advertisements reflected newspaper articles as expressed by Granger causality directed exclusively from articles to advertisements;
- A complex, possibly externally-driven, causal pattern as evidenced by cases where Granger causality goes from articles to advertisements and vice versa.
For the second step of the analysis, we used AFA to model the persistence for each
keyword in both types of discourse. This enabled us to identify possible dynamic
properties of either advertisements and articles as a whole, and possibly the
dynamic properties of particular words. Similar to Granger causal-like patterns,
the Hurst exponent has three possible patterns of persistence: anti-persistent
processes, short-term correlation processes, and persistent processes. Each
keyword’s behavior can thus be described by one of nine possible combinations
of
causality (Granger causality test) and persistence (AFA). Insights into these
dynamic properties alongside the causal patterns can help to increase our
understanding of the relationship between advertisements and articles, and by
extension, between advertisements and society.
Data Analysis
Statistical tests were conducted with an
α
level of .005 [Benjamin et al. 2017].
Pearson’s correlation coefficient
R
was used to measure the non-lagged association
strength between the time series. We converted Pearson’s
R
using Fisher’s Z-transformation to normally
distributed z-values to permit averaging. Before applying the Granger causality
test for comparison of discourses, lag-1 differencing was used to obtain a
stationary keyword time series.
For the analysis of the Hurst exponent for each keyword time series per discourse,
we used a simple linear regression and compared this with the constant model.
This allowed us to test differences in long-range dependencies between the two
different discourses. The Shapiro-Wilk
W
test confirmed that the distribution of the Hurst
exponent did not deviate significantly from normality [Shapiro and Wilk 1965].[5]
Results
Directionality
On average, the variance in correlation for each keyword in all the time series
was similar between advertisement and articles. The mean correlation
coefficient,
R
¯
, between advertisements and the articles in the
newspapers De Tijd and De
Telegraaf was
R
¯
=
.
25
and
R
¯
=
.
27
respectively. Sixty-two percent (62%) of these
correlations were statistically reliable. The within-newspaper correlation,
that
is, the correlation between De Tijd and De Telegraaf, was considerably stronger,
R
¯
=
.
42
. Seventy-three percent (73%) of these correlations
were significant, suggesting that word use over time in articles between these
two newspapers was more similar than between the articles and
advertisements.
Figure 2.
Figure 2: Six keyword frequencies plotted at bi-annual intervals for the
two newspapers (News 1: De Tijd; News 2:
De Telegraaf) and the advertisements
(Ads). The line is smoothed using a simple moving average filter (window
size of five years) and gray bands represent confidence intervals at 95%.
Notice how ‘Cinema’ shows a distinct shaping Granger causal pattern, where
fluctuations in Ads antecede fluctuations in News. In contrast ‘Potato’
displays a complex Granger causal pattern, where fluctuations in Ads both
seem to antecede and succeed News fluctuations. The ‘Potato’ system’s spiky
behavior during the Dutch famine (1944-45) indicates that war is part of
the external cause.Analysis showed that there was no overarching causal pattern, but rather multiple
Granger causal patterns that were keyword-dependent. Twenty percent (20%) of
the
product terms show evidence of a shaping causality, in which
discursive trends in advertising discourse uniquely predict those in articles.
For 17% of the product terms, we found the causal pattern in which
advertisements reflect articles. Almost half of the product terms
(49%) belong to the complex externally-driven category, that is, fluctuations
in
both advertisements and articles seem to be predicted by extraneous events that
perturb the reflecting-shaping dynamic between the time series. Finally, 14%
of
the terms show no indication of predictive causality.
Almost half of the product terms exhibit complex behavior, pointing towards an
external source driving changes in discourse in advertisements and newspapers.
These could include economic developments or possibly the invention of new
products or technologies. In a way, this underpins that the relationship between
advertisements and articles was one of negotiation with external developments.
Noteworthy keywords in this category were related to produce (“apple”,
“cauliflower,”
“lettuce”), energy (“stove”, “cokes”, “furnace”,
“gasoline”) and audiovisual technology (“tape recorder”,
“gramophone”, “radio”, “television”).
There are slightly more keywords for which advertisements were shaping articles
than ads reflecting articles. In case of shaping, we detected that behavior
in
word use in advertisements was related to behavior in articles. Words that
exhibited this behavior referred quite generally to fashion and clothing
(“men’s clothing”, “sweater”, “fur”, “wool”,
“flannel”, “jeans”, “heels”), interior design (“living
room”, “dining room”, “bedroom”) and movies (“cinema”,
“film”). The keywords that exhibited reflecting behavior were more
difficult to categorize into particular categories. They ranged from words such
as “teapot”, “dictionary”, to “cheese”. Generally, the words with
reflecting behavior seemed to be more specific than the words in the shaping
category. For instance, the keywords “chair cushions” and “winter
coat” are specific types of cushions and coats.
Detection of Persistence in Word Use
The Hurst exponent
H
for both articles (
t
1058
=
32.8
,
p
<
.
00001
) and advertisements (
t
538
=
38.5
,
p
<
.
00001
) was significantly higher when compared to a
no-memory baseline (
H
:
M
=
0.5
,
S
D
=
0.18
). In terms of persistence, discourse in articles and
advertisements was different from processes that only showed short-range
correlations. AFA found an average difference (
Δ
H
) between articles of 0.21. The dynamics related to
word use in articles thus differed from advertisements. Word use in articles
was
more likely to be a persistent process (
H
:
M
=
0.89
,
S
D
=
0.19
) than it was in advertisements (
H
:
M
=
1.1
,
S
D
=
0.17
). Notice that advertisements display non-stationary
dynamics with
H
>
1
. To test whether the difference between ads and
articles was significant, we ran linear regression to predict H as a function of advertisement and articles (advertisement as
baseline). Compared to the constant model, a statistically significant
regression model was found (
χ
1
2
=
149.1
,
p
<
.
00001
) showing that H was, indeed,
reliably lower for articles than for advertisements (articles:
β
= -0.18 ,
S
E
±
0.01,
F
1,793
=
163.9
p
<
.
00001
).
On the whole, word use in articles more clearly expressed persistent trends, while
word use in advertisements tended to be more irregular, displaying bursts of
high activity followed by little or no activity. This indicates that articles
more closely express behavior that could be interpreted as communicative memory,
while ads seem more haphazard and perhaps catalytic to the existence of memory
in wider public discourse, as represented through articles.
Our findings indicate three distinct types of persistent trends (see Table 1). Persistence in only articles, persistence
in advertisements and articles, and a lack of persistence in either source.
Words that only exhibited persistence in articles included products related
to
interior design (“living room”, “couch”, “lamp”, “bedroom”).
This suggests that discourse about these products was part of a shared language
but that it was not clearly part of an advertising discourse. Conversely, words
that showed persistence in both discourses included those related to cigarettes
but not to cigars and tobacco (“cigarettes”), fashion (“fur”,
“jeans”), energy (“cokes”, “furnace”, “gasoline”) and
produce (“apple”, “cauliflower”, “lettuce”). This suggests that
advertising discourse for these products was much more persistent and relied
on
an established frame of reference. Interestingly, cigarettes and fashion are
often presented as typical examples of strongly branded products [Blaszczyk 2008]
[Brandt 2004]
[Hill 2002]
[White et al. 2012]. Advertisements for produce, on the other hand, might
exhibit persistent processes due to its highly seasonal and reoccurring nature
(cf. Gao et al., 2012). Keywords that showed
no persistence were related to technology (“cinema”, “tape recorder”,
“gramophone”, “radio”).
| Persistence in articles | Persistence in ads & articles | No persistence |
| Living room | Cigarettes | Cinema |
| Dining room | Fur | Film |
| Bedroom | Wool | Tape recorder |
| Chair | Flannel | Gramophone |
| Couch | Jeans | Radio |
| Cupboard | Heels | Television |
| Seat | Apple | |
| Lamp | Cauliflower | |
| Lettuce | ||
| Cokes | ||
| Furnace | ||
| gasoline |
Table 1.
Table 1: Example of keywords grouped on type of persistent trend
(persistence in articles only, persis- tence in articles and
advertisements, and no persistenceDiscussion
Using AFA, we found a significant difference in persistent behavior between word use
in
advertisements and articles. The latter exhibited long-term dependencies whereas
advertisements displayed more non-stationary and irregular behavior. In general,
advertisements introduced terms, but many of these terms did not persist and
their
decay was rapid. For articles, on the other hand, keywords that denoted products
showed more persistent behavior and were either mentioned recurrently in a
self-reinforcing manner or decayed much slower than advertisements. We speculate
that
this reflects an overarching media dynamic in which ads introduced keywords and
articles represented how these products became part of public discourse. However,
this dynamic does not hold for all products as evidenced by Table 1, and, at
least
partially, by keywords that exhibit a reflecting causal pattern.
The time series of keywords between the two newspapers were more clearly correlated
than the time series between the newspapers and advertisements. This shows that
word
use in newspapers more closely followed each other than word use between
advertisements and newspapers. Along with the found difference in
H
between articles and advertisements, the dissimilarity in
correlation adds evidence to the hypothesis that the dynamics of discourse in
ads are
different from articles. This demonstrates that advertisements are not merely
a lens
on the past, but more clearly a “distorted mirror” that is shaped to a certain
degree by advertisers and its own dynamics.
In terms of directionality, we did not find one dominant pattern. For 20% percent
of
the keywords, advertisements reflected articles, and for 17% of the keywords,
advertisements shaped articles. But for almost half of the keywords, there was
a more
complex causal relationship, indicative of external forces. This lends support
to
Cowan’s argument for a complex interaction pattern in which advertisers,
distributors, producers, and consumers negotiated the meaning of a product [Cowan 1997].
Product Groups
The causal direction and type of persistence seems to be, to some extent, related
to product type. We were not able to identify specific categories of keywords
in
the reflecting causal category. However, the complex relationship
and shaping category offered interesting groupings of words. The
groupings made on the basis of the existence of memory and causal directionality
leads to the following four points of discussion.
First, products with a shaping dynamic and long-term dependencies in
articles might point towards products that are not constantly
advertised — expressed by the lack of persistence in ads — but that nevertheless
are
part of the cultural life of Dutch consumers throughout the twentieth century,
such as bikes, pets, interiors, and clothes. The shaping dynamic reveals that
ad
makers might have pushed the popularity of these products, which can be
described as lifestyle products. One could argue that advertisers might have
been able to affect the longevity of these products, effectively installing
them
within a shared “community of discourse”.
Second, one of the most noteworthy behaviors is associated with the cigarette.
This product exhibits persistence in advertisements and in
articles, and it shows a shaping causal behavior. This suggests that in
advertising discourse and in newspaper discourse, cigarettes were a recurring
topic that built upon earlier discourse. Moreover, advertisers seemed to be
able
to shape newspaper discourse on cigarettes. This finding is in line with
scholars that view advertisements for cigarettes as the prime example of social
engineering [Brandt 2004]. Our study finds that, at least for the
Dutch context, cigarette advertisements were a steadily successful form of
advertising. The unique behavior of cigarettes was underlined by the fact that
related products such as tobacco and cigars behaved dissimilarly. Tobacco and
cigars exhibit no persistence and are driven by a complex causal relationship,
underscoring different advertising dynamic than found for cigarettes.
Third, some products revealed persistence in both advertisements and articles
without displaying a uniform causal relationship. These products include
produce, energy sources, and computer systems. One interpretation might be that
produce was of prolonged importance, indicated by the existence of long-range
dependencies, yet its importance was not driven by advertisers but by an
external source. In the case of produce, this external source could be seasonal
or economic shifts. The other two product groups (energy sources and computer
systems) were also quite instrumental in society, albeit not for prolonged
periods during the twentieth century. The keywords associated with energy were
most dominant in the first half of the century, whereas, the words associated
with computers only appeared in the latter quarter of the century. Nonetheless,
they both still exhibited persistence in newspaper articles.
Finally, keywords associated with technological innovations showed two distinct
types of behavior. First, keywords such as “cinema”, “tape recorder”
and “television” did not exhibit any persistence, which could have resulted
from the constant innovation and disruptions in the field of audiovisual
technology. Another explanation could be the use of different keywords to refer
to similar technologies. Further research is needed to explore this behavior
related to technology. Second, we found a distinction in the causal relationship
between types of technology. “Cinema” and “film” showed a clear causal
relationship between ads and articles. The causal relationship might have
resulted from the fact that advertisements played an important role in pushing
these innovations to a wider audience. Keywords associated with household
technology (“radio” and “television”), on the other hand, displayed
the complex type of causality. These technological products might be more
closely related to particular economic, seasonal, or innovative cycles. Again,
further research is needed to untangle these dynamics.
Conclusion
Through two data experiments, we have found evidence of a fundamental difference
between the dynamic behavior of word use related to consumer products in articles
and
advertisements published in newspapers. Articles — taken as a proxy for public
discourse — exhibit persistent trends that are likely to be reflective of communicative
memory. Contrary to this, advertisements have a more irregular behavior characterized
by short bursts and fast decay, which, in part, mirrors the dynamic through which
advertisers introduced terms into public discourse. On the issue of whether
advertisements shaped or reflected society, we found particular product types
that
seemed to be collectively driven by a causality going from advertisements to
articles. Generally, we found support for a complex interaction pattern that
Cowan
dubbed the “consumption junction”.
Finally, we discovered noteworthy patterns in terms of causality and long-range
dependencies for specific product groups.
These findings resonate with Marchand’s claim that advertisements contributed to the
“shaping of a ‘community
of discourse,’ an integrative common language shared by an
otherwise diverse audience” [Marchand 1985]. In
other words, ads seem to behave as a driver of processes in newspaper articles.
Their
behavior clearly differs from general discourse, which might stem from the fact
that
ads are to a large extent driven by ad makers and particular technological
innovations.
This study shows how methods from fields of econometrics and complexity science can
be
applied to improve our understanding of complex cultural-historical phenomena.
Further research that employs more extensive keyword lists that also includes
brand
names and cross-cultural comparisons will make it possible to propose a more
detailed
and general account of the mechanics that underpin the aforementioned consumption
junction.
Acknowledgements
Part of this research was performed while the authors were visiting the Institute
for Pure and Applied Mathematics (IPAM), which is supported by the National
Science Foundation. The newspaper data was provided by the National Library
of
the Netherlands (KB).
Notes
[1] Google
Books Ngram Viewer: https://books.google.com/ngrams
[2] Examples of ngram viewers: The British
Library: https://www.webarchive.org.uk/ukwa/ngram/, Danish Royal
Library: http://labs.statsbiblioteket.dk/smurf/, and National Library of
the Netherlands http://kbkranten.politicalmashup.nl
[3] These newspapers can be accessed through Delpher: https://www.delpher.nl
[4] Long-range dependency is also
called persistent behavior or long-memory in time series. The terms will be
used interchangeably.
[5] While some keyword time series did show indications of
multifractal structure (i.e. local fluctuations with either small or large
variation), this information was discarded from the final analysis for the
purpose of simplification.
Works Cited
Assman 2008 Assmann, J. (2008), “Communicative and Cultural Memory”, in Erll, A., Nünning, A. and Young,
S.B. (Eds.), Cultural Memory Studies: An International and
Interdisciplinary Handbook, Walter de Gruyter, New York, pp.
109–119.
Assman 2011 Assmann, J. (2011), Cultural Memory and Early Civilization: Writing, Remembrance, and Political
Imagination, Cambridge University Press, New York.
Assman and Czaplicka 1995 Assmann, J. and Czaplicka, J.
(1995), “Collective Memory and Cultural Identity”, New German Critique, No. 65, p. 125.
Benjamin et al. 2017 Benjamin, D.J., Berger, J.O.,
Berk, R., Bollen, K.A., Brembs, B., Brown, L., Camerer, C., et al. (2017), “Redefine statistical significance”, Nature Human Behaviour, available at: https://doi.org/10.1038/s41562-017-0189-z.
Blaszczyk 2008 Blaszczyk, R.L. (Ed.). (2008), Producing Fashion: Commerce, Culture, and Consumers,
University of Pennsylvania Press, Philadelphia.
Brandt 2004 Brandt, A. (2004), “Engineering Consumer Confidence in the Twentieth Century”, in Zhou, X.
and Gilman, S. (Eds.), Smoke: A Global History of
Smoking, Reaktion Books, London.
Chater and Brown 1999 Chater, N. and Brown, G.D. (1999),
“Scale-invariance as a unifying psychological
principle”, Cognition, Vol. 69 No. 3, pp.
B17–B24.
Cowan 1987 Cowan, R.S. (1987), “The
consumption junction: A proposal for research strategies in the sociology of
technology”, in Bijker, W., Hughes, T. and Pinch, T. (Eds.), The Social Construction of Technological Systems: New Directions
in the Sociology and History of Technology, MIT Press, Cambridge, pp.
261–80.
Cowan 1997 Cowan, R.S. (1997), “A
Proposal for Research Strategies in the Sociology of Technology”, in
Bijker, W.E., Hughes, T. and Pinch, T. (Eds.), The Social
Construction of Technological Systems: Papers of a Workshop Held at the
University of Twente, The Netherlands, in July 1984, MIT Press,
Cambridge, pp. 261–280.
Cross 2000 Cross, G. (2000), An
All-Consuming Century: Why Commercialism Won in Modern America, Columbia
University Press, New York.
Donald 2002 Donald, M. (2002), A
Mind So Rare: The Evolution of Human Consciousness, Norton, New
York.
Eke et al. 2002 Eke, A., Herman, P., Kocsis, L. and Kozak,
L.R. (2002), “Fractal characterization of complexity in temporal
physiological signals”, Physiological
Measurement, Vol. 23 No. 1, pp. R1–R38.
Fox 1997 Fox, S.R. (1997), The Mirror
Makers: A History of American Advertising and Its Creators, University
of Illinois Press, Urbana.
Gao 2012 Gao, J., Hu, J., Mao, X. and Perc, M. (2012),
“Culturomics meets random fractal theory: insights into
long-range correlations of social and natural phenomena over the past two
centuries”, Journal of The Royal Society
Interface, Vol. 9 No. 73, pp. 1956–1964.
Gao 2017 Gao, J., Fang, P. and Liu, F. (2017), “Empirical scaling law connecting persistence and severity of
global terrorism”, Physica A: Statistical Mechanics
and Its Applications, Vol. 482, pp. 74–86.
Gao et al. 2011 Gao, J., Hu, J. and Tung, W. (2011), “Facilitating Joint Chaos and Fractal Analysis of Biosignals
through Nonlinear Adaptive Filtering”, PLOS
ONE, Vol. 6 No. 9, p. e24331.
Goffman 1985 Goffman, E. (1985), Gender Advertisements, Macmillan, London.
Granger 1969 Granger, C.W.J. (1969), “Investigating Causal Relations by Econometric Models and
Cross-spectral Methods”, Econometrica, Vol.
37 No. 3, pp. 424–438.
Helbing 2015 Helbing, D., Brockmann, D., Chadefaux, T.,
Donnay, K., Blanke, U., Woolley-Meza, O., Moussaid, M., et al. (2015), “Saving human lives: What complexity science and information
systems can contribute”, Journal of Statistical
Physics, Vol. 158 No. 3, pp. 735–781.
Hill 2002 Hill, D.D. (2002), Advertising to the American Woman, 1900-1999, Ohio State University
Press, Columbus.
Lears 1994 Lears, T.J. (1994), Fables of Abundance: A Cultural History of Advertising in America,
Basic Books, New York.
Marchand 1985 Marchand, R. (1985), Advertising the American Dream: Making Way for Modernity,
1920-1940, University of California Press, Berkeley.
Marchant 2008 Marchant, T. (2008), “Scale invariance and similar invariance conditions for bankruptcy
problems”, Social Choice and Welfare, Vol.
31 No. 4, pp. 693–707.
Marshall 1995 Marshall, M. (1995), Contesting Cultural Rhetorics: Public Discourse and Education,
1890-1900, University of Michigan Press, Ann Arbor.
Michel et al. 2011 Michel, J.-B., Shen, Y.K., Aiden,
A.P., Veres, A., Gray, M.K., The Google Books Team, Pickett, J.P., et al. (2011),
“Quantitative Analysis of Culture Using Millions of
Digitized Books”, Science, Vol. 331 No.
6014, pp. 176–182.
Nielbo et al. 2019 Nielbo, K.L., Baunvig, K.F., Liu, B.
and Gao, J. (2019), “A curious case of entropic decay: Persistent
complexity in textual cultural heritage”, Digital
Scholarship in the Humanities, Vol. 34 No. 3, pp. 542–557.
O'Brien et al. 2014 O’Brien, B.A., Wallot, S.,
Haussmann, A. and Kloos, H. (2014), “Using Complexity Metrics to
Assess Silent Reading Fluency: A Cross-Sectional Study Comparing Oral and Silent
Reading”, Scientific Studies of Reading,
Vol. 18 No. 4, pp. 235–254.
Parkin 2007 Parkin, K. (2007), Food
Is Love: Advertising and Gender Roles in Modern America, University of
Pennsylvania Press, Philadelphia.
Peng et al. 1994 Peng, C.-K., Buldyrev, S.V., Havlin, S.,
Simons, M., Stanley, H.E. and Goldberger, A.L. (1994), “Mosaic
organization of DNA nucleotides”, Physical
Review, Vol. 49 No. 2, p. 1685.
Petersen et al. 2012 Petersen, A.M., Tenenbaum, J.N.,
Havlin, S., Stanley, H.E. and Perc, M. (2012), “Languages cool as
they expand: Allometric scaling and the decreasing need for new words”,
Scientific Reports, Vol. 2, p. 943.
Riley et al. 2012 Riley, M.A., Bonnette, S., Kuznetsov,
N., Wallot, S. and Gao, J. (2012), “A tutorial introduction to
adaptive fractal analysis”, Frontiers in
Physiology, Vol. 3, available at: https://doi.org/10.3389/fphys.2012.00371.
Rooij 1974 Rooij, M. (1974), Kranten: dagbladpers en maatschappij, Wetenschappelijke Uitgeverij,
Amsterdam.
Schreurs 2001 Schreurs, W. (2001), Geschiedenis van de Reclame in Nederland, Het Spectrum,
Utrecht.
Schudson 1982 Schudson, M. (1982), The Power of News, Harvard University Press,
Cambridge.
Shapiro and Wilk 1965 Shapiro, S.S. and Wilk, M.B.
(1965), “An Analysis of Variance Test for Normality (Complete
Samples)”, Biometrika, Vol. 52 No. 3/4, pp.
591–611.
Sivulka 2012 Sivulka, J. (2012), Soap, Sex, and Cigarettes: A Cultural History of American Advertising,
2nd ed., Wadsworth, Boston.
Smail 2008 Smail, D.L. (2008), On
Deep History and the Brain, University of California Press,
Berkeley.
Sugihara et al. 2012 Sugihara, G., May, R., Ye, H.,
Hsieh, C., Deyle, E., Fogarty, M. and Munch, S. (2012), “Detecting causality in complex ecosystems”, Science, Vol. 338 No. 6106, pp. 496–500.
Traub et al. 2015 Traub, M.C., van Ossenbruggen, J. and
Hardman, L. (2015), “Impact analysis of OCR quality on research
tasks in digital archives”, International Conference
on Theory and Practice of Digital Libraries, Springer, pp.
252–263.
Voss and Clarke 1975 Voss, R.F. and Clarke, J. (1975),
“‘1/ f noise’ in music and speech”, Nature, Vol. 258 No. 5533, pp. 317–318.
Wevers and Verhoef 2017 Wevers, M. and Verhoef, J.
(2017), “Coca-Cola: An Icon of the American Way of Life. An
Iterative Text Mining Workflow for Analyzing Advertisements in Dutch
Twentieth-Century Newspapers”, Digital Humanities
Quarterly, Vol. 11 No. 4.
White et al. 2012 White, C., Oliffe, J.L. and Bottorff,
J.L. (2012), “From the Physician to the Marlboro Man:
Masculinity, Health, and Cigarette Advertising in America, 1946–1964”,
Men and Masculinities, Vol. 15 No. 5, pp.
526–547.
de Grazia 2005 de Grazia, V. (2005), Irresistible Empire: America’s Advance Through Twentieth-Century
Europe, Kindle Edition., Belknap Press of Harvard University Press,
Cambridge.
Recommendations
DHQ is testing out three new article recommendation methods! Please explore the links below to find articles that are related in different ways to the one you just read. We are interested in how these methods work for readers—if you would like to share feedback with us, please complete our short evaluation survey. You can also visit our documentation for these recommendation methods to learn more.
SPECTER Recommendations
Below are article recommendations generated by the SPECTER model:
- Coca-Cola: An Icon of the American Way of Life. An Iterative Text Mining Workflow for Analyzing Advertisements in Dutch Twentieth-Century Newspapers, 2018, Melvin Wevers, DH Group, KNAW HUC, Amsterdam, The Netherlands; Jesper Verhoef, TU Delft, Delft, The Netherlands
- The Page Is an Image Again: Bleedmapping as an Analysis Technique for Historical Newspapers, 2023, Quintus van Galen, University of Amsterdam
- Attention Ecology: Trend Circulation and the Virality Threshold, 2016, Nicholas M Van Horn, Capital University; Aaron Beveridge, University of Florida; Sean Morey, University of Tennessee, Knoxville
- Using word vector models to trace conceptual change over time and space in historical newspapers, 1840–1914, 2022, Jaap Verheul, Utrecht University; Hannu Salmi, University of Turku; Martin Riedl, University of Stuttgart; Asko Nivala, University of Turku; Lorella Viola, University of Luxembourg; Jana Keck, German Historical Institute Washington; Emily Bell, University of Leeds
- Finding Narratives in News Flows: The Temporal Dimension of News Stories, 2022, Blanca Calvo Figueras, University of Gronigen; Tommaso Caselli, University of Groningen; Marcel Broersma, Centre for Media and Journalism Studies, University of Groningen
DHQ Keyword Recommendations
Below are article recommendations generated by DHQ Keywords:
- Mining Public Discourse for Emerging Dutch Nationalism, 2016, Maarten van den Bos, Utrecht University; Hermione Giffard, Utrecht University
- Avanços no estudo das redes de itinerários da Península Ibérica no século XVI. Aplicando os SIGH para estudar a história da arquitetura, 2020, Patricia Ferreira-Lopes, Escuela Técnica Superior de Arquitectura de la Universidad de Sevilla
- Mining Embodied Emotions: A Comparative Analysis of Sentiment and Emotion in Dutch Texts, 1600-1800., 2018, Inger Leemans, Faculty of Humanities, Vrije Universiteit Amsterdam, The Netherlands; Janneke M. van der Zwaan, Netherlands eScience Center, Amsterdam, The Netherlands; Isa Maks, Faculty of Humanities, Vrije Universiteit Amsterdam, The Netherlands; Erika Kuijpers, Faculty of Humanities, Vrije Universiteit, Amsterdam, The Netherlands; Kristine Steenbergh, Faculty of Humanities, Vrije Universiteit, Amsterdam, The Netherlands
- Studying Large-Scale Behavioral Differences in Auschwitz-Birkenau with Simulation of Gendered Narratives, 2022, Gábor Mihály Tóth, USC Shoah Foundation / USC Viterbi School of Engineering; Tim Hempel, Freie Universität Berlin, Department of Mathematics and Computer Science / Department of Physics; Krishna Somandepalli, Google Research; Shri Narayanan, USC Viterbi School of Engineering
- Unraveling reported dreams with text analytics, 2018, Iris Hendrickx, Centre for Language Studies, Radboud University, Nijmegen, The Netherlands; Centre for Language and Speech Technology, Radboud University, Nijmegen, The Netherlands; Louis Onrust, Centre for Language Studies, Radboud University, Nijmegen, The Netherlands; Florian Kunneman, Centre for Language Studies, Radboud University, Nijmegen, The Netherlands; Ali Hürriyetoğlu, Centre for Language Studies, Radboud University, Nijmegen, The Netherlands; Wessel Stoop, Centre for Language and Speech Technology, Radboud University, Nijmegen, The Netherlands; Antal van den Bosch, Meertens Institute, Amsterdam, The Netherlands
TF-IDF Recommendations
Below are article recommendations generated by the TF-IDF Model:
- Coca-Cola: An Icon of the American Way of Life. An Iterative Text Mining Workflow for Analyzing Advertisements in Dutch Twentieth-Century Newspapers, 2018, Melvin Wevers, DH Group, KNAW HUC, Amsterdam, The Netherlands; Jesper Verhoef, TU Delft, Delft, The Netherlands
- Word Processor Art: How "User-friendly" Inhibits Creativity, 2016, Renée Farrar, United States Military Academy at West Point
- Writing to be Found and Writing Readers, 2011, John Cayley, Brown University
- Comic Book Markup Language: An Introduction and Rationale, 2012, John A. Walsh, Indiana University
- Crowdsourcing Image Extraction and Annotation: Software Development and Case Study, 2020, Ana Jofre, SUNY Polytechnic; Vincent Berardi, Chapman University; Kathleen P.J. Brennan, University of Queensland; Aisha Cornejo, Chapman University; Carl Bennett, SUNY Polytechnic; John Harlan, SUNY Polytechnic
URL: https://dhq.digitalhumanities.org/vol/14/2/000445.html
Comments: [email protected]
Published by: The Alliance of Digital Humanities Organizations and The Association for Computers and the Humanities
Affiliated with: Digital Scholarship in the Humanities
DHQ has been made possible in part by the National Endowment for the Humanities.
Copyright © 2005 -
Unless otherwise noted, the DHQ web site and all DHQ published content are published under a Creative Commons Attribution-NoDerivatives 4.0 International License. Individual articles may carry a more permissive license, as described in the footer for the individual article, and in the article’s metadata.
Comments: [email protected]
Published by: The Alliance of Digital Humanities Organizations and The Association for Computers and the Humanities
Affiliated with: Digital Scholarship in the Humanities
DHQ has been made possible in part by the National Endowment for the Humanities.
Copyright © 2005 -
Unless otherwise noted, the DHQ web site and all DHQ published content are published under a Creative Commons Attribution-NoDerivatives 4.0 International License. Individual articles may carry a more permissive license, as described in the footer for the individual article, and in the article’s metadata.
