Qual Quant (2012) 46:10131024

DOI 10.1007/s11135-011-9445-x
Online image content analysis of political gures:
an exploratory study
Bulent Ozel Han Woo Park
Received: 31 July 2010 / Accepted: 17 February 2011 / Published online: 24 April 2011
Springer Science+Business Media B.V. 2011
Abstract This study investigates the emotional content of facial images of South Korean
politicians. The data were drawn fromofcial websites of 18th National Assembly members.
We categorized facial expressions (smiling, frowning, and no expression) by using photos
on the websites. The smiling image was the most prevalent facial expression regardless of
the socio-political-demographic attributes of the politicians. The existence and strength of
the smiling image had significant positive correlations with the politicians web visibility.
Opposition parties were significantly more likely to show frowning or expressionless images
than the ruling party. The more experienced politicians were more likely to continue their
use smiling images than similarly aged politicians with less experience.
Keywords Visual content analysis Emotional content analysis Online visibility
Web-based campaigns Facial expressions Candidate websites
1 Introduction
In new attention-grabbing cyberspace (e.g., McAllister and Turow 2009), one aim of indi-
vidual and organizational websites is to induce active responses (e.g., click, post, trackback,
etc.) from visitors. Because the publics interest in politics has been decreasing, political
An earlier version of this paper was presented at the ICCC2009 (International Conference on Convergence
Content), December 1719, 2009, Hanoi University of Culture, Vietnam.
B. Ozel
Computer Science Department, Istanbul Bilgi University, Istanbul, Turkey
e-mail: bulento@bilgi.edu.tr
H. W. Park (B)
World Class University Webometrics Institute, Gyeongsan-Si, South Korea
e-mail: hanpark@ynu.ac.kr
H. W. Park
Department of Media and Communication, YeungNam University, Gyeongsan-Si, South Korea
1 3
1014 B. Ozel, H. W. Park
strategies have been increasingly focused on gaining peoples attention and turning it into
their participation and, ultimately, votes. One objective of homepage maintenance, from the
politician perspective, is to present people with current news, policy reports, and photos so
that they could verify and appreciate politicians performance (Foot and Schneider 2006).
Accordingly, politicians have been increasingly uploading visual content (e.g., snapshots,
video clips, etc.) in either the static or dynamic format on their websites to interact more
effectively with their electorate.
The use of images can be a powerful strategy for a politician because people can copy
and modify the images freely and spread the politicians message on cyberspace (Druck-
man et al. 2010; Heaney et al. 2010; Druckman et al. 2007). Online images empower young
born-digitals (Palfrey and Gasser 2008) by allowing them to share political information
more efciently. Online visuals are critical when website visitors are relatively unfamiliar
with places, objects, and activities displayed on the site (Hellemans and Govers 2005).
Therefore, in this digital era, politicians can benet from online photo images in terms
of civic engagement and mobilization. A website can enable politicians by effectively dis-
seminating political messages. Previous studies have focused mostly on the content of online
communication and political services of politicians.
1
However, fewhave examined the effects
of photographic images displayed on politicians websites, which is the topic of this research.
There are many different types of visual content on political websites. This paper examines
the emotional content embodied in facial pictures of Korean politicians. Facial expressions
are important in web-based campaigns because a smiling photo can be immensely useful for
promoting a positive image to voters. On the other hand, negative online facial images can be
easily copied, manipulated, and distributed. Negative facial images can drawunwanted atten-
tion to politicians. However, previous studies have not systematically examined the effects
of smiling and/or frowning faces on voters political interest and voting intentions. In this
regard, this research does not report the linear causality between the number of certain facial
expressions and political performance in the Korean context. Rather, it makes a preliminary
attempt to apply an image-based content analysis technique to the political process to better
understand the topology of online campaigns. More specifically, this study addresses the
following interrelated set of research questions, which are exploratory in nature and require
future empirical analysis:
What types of facial expressions are displayed on ofcial homepages of politicians? More
specifically, how do facial images differ among politicians in terms of their socio-politi-
cal-demographic attributes and web visibility index?
2 Related literature
Previous image research has focused mostly on the content analysis of photographic images
in media and communication elds (for a complete review, see Bauer 2000); studies of online
images represent a relatively recent phenomenon. In such research on online image content,
social scientists have generally taken two approaches, whereas those in the computer and
information science community have taken one. The two approaches of social scientists are
as follows: (i) descriptive studies investigating the usage practice of digital photos (e.g., to
improve the presentation quality of messages and provide rich and contextual information)
and (ii) a theoretical perspective based on the (statistically tested) relationship between the
1
For a comparison of political website features across countries, see Kluver et al. (2007).
1 3
Analysis of political gures 1015
intended strategy of website producers and the specic use of online images for a particular
purpose.
A third stream of studies, which take a technical approach to the study of online images,
is usually methodological in nature, including digitizing images and organizing databases
(e.g., parsing based on the use of new algorithms and digital tools available online). Angus
et al. (2011) is a representative example of the third approach. However, we do not deal with
technical literature because such discussions are beyond the scope of our study.
More specifically, the rst social science approach addresses mainly the existence and/or
prevalence of online images on websites. In a pioneering study of web campaigns, Foot and
Schneider (2006) examined the frequency of political event photos on US campaign web-
sites from1998 to 2004. They found that both political parties and individual politicians have
gradually embraced digital images on their websites. Further, they argued that the practice
of displaying many campaign activity photos facilitates cognitive afliation between the site
producer and site visitors. Similarly, Park and Bae (2007) performed a content analysis of 277
Korean National Assembly members in 2004. They classied images into four categories and
found that portraits of individual politicians were the most prevalent (87.4%). The next image
category was political activities (53.4%), followed by cartoons/comics (23.1%). Noteworthy
is that the progressive party (Democratic Labor Party; DLP) members (66.7%) included more
political activities photos than other party members did. Kim et al. (2006) study is similar
to our research. They compared website images of major Korean political gures and found
that the number of smiling images of current President Myung-Bak Lee was slightly lower
than that of his female competitor, Geun-Hye Park (29.3 and 32.8%, respectively), although
the difference was not statistically significant.
Whereas the rst social science approach focuses on online images to discern the adoption
of digital photos by websites, the second applies the existing social science theories to exam-
ine the ways in which website authors use images to induce the involvement of site visitors
in the activities promoted by the sites.Verser and Wicks (2006) examined the role of visual
images during the 2000 electoral campaigns in the US by taking the impression management
theory perspective. They found that Gore and Bush used their websites to improve negative
media portrayals. Whereas Gore presented campaign photographs that might have looked
less serious, Bush focused on developing a respectable gure. In a study of the use of web
technologies among US congressmen, Druckman et al. (2007) found that politicians chose
visual technologies (e.g., static photos, dynamic images, etc.) by considering practical and
strategic political implications, reecting the technology acceptance model (Davis 1989).
Electoral candidates in a tight contest tended to present more visual images on their web-
sites to better covey their central campaign messages. In other words, online images served
as an important means to present political activities. More recently, Page and Duffy (2008)
evaluated the visual strategies the candidates employed in the 2008 US presidential cam-
paign. They selected photographic images found on biographical sections of the sites and
analyzed the overall themes of the images by considering symbolic convergence theory. Their
research showed that politicians used online images to build intimate relationships with var-
ious electorates. More specifically, candidates, regardless of their political afliation, tried
to boost positive image attributes such as patriotism, family, heritage, multi-culturalism, and
populism.
Content analysis has been widely used in a number of elds (particularly in tourism) for
examining website images (Choi et al. 2007; Govers et al. 2004). For instance, although
Govers et al. (2004) did not examine political gures in cyberspace, the content analysis
of digital images extracted from 20 Dubai-based tourism-related websites enabled them to
examine the digitalized characteristics of its tourism industry.
1 3
1016 B. Ozel, H. W. Park
Table 1 Classication of facial expressions
Types Content
Smiling face Turning up the corners of the mouth, usually showing their teeth; an upward curving of
the corners of the mouth, revealing pleasure, happiness, or amusement; a downward
curving of the corners of the eyes, expressing moderate joy.
Frowning face Wrinkling of the brow, showing displeasure, anger, unrest, disapproval, and tiredness; a
downward curving of the corners of the mouth; staring at something with anger,
discontent, or unkindness.
No expression No movement around the mouth, eyes, or eyebrows, revealing no emotional information.
Some other recent studies have started to recognize the potential of online image content
analysis. However, efforts have been largely limited to the English-speaking webosphere.
Further, previous research has suggested that internet researchers seek detailed evidence to
gain a deeper and richer understanding of visual strategies of political sites within/across party
lines. However, fewefforts have been made to examine the relationship between socio-demo-
graphic attributes of political gures and the presentation style of visual images on websites.
It is in this regard that the current study focuses on Korean congressional members facial
images on their ofcial campaign websites.
3 Method
The images for this research were drawn from ofcial homepages of members of South
Koreas 18th National Assembly, who were elected in April 2008.
Of the 292 members, we analyzed 277; eight did not have a homepage at the time of data
collection (August 13, 2009), and the websites of seven were not accessible as a result of
technical maintenance. We obtained all the photos from these homepages by using a freely
available software package (Teleport Pro) and then manually collected a list of images posted
only on the front page of the politicians homepages. During the combined process of com-
puter-assisted collection and manual verication, we minimized the possible exclusion of
some photographic images using single software package. All forms of images, including
static photos and dynamic Flash animations on the front page, were collected. Later, images
in video les were excluded from further analysis.
We then classied facial expressions into the following categories: smiling, frowning, and
no expression. Coding schemes were drawn from previous studies (Coleman and Wasike
2004;Verser and Wicks 2006). Given that this is an exploratory study, one coder was used.
During this process, we used the criterion in Table 1.
The R statistical tool and programming environment was used for all of the analysis and
graphical visualization.
2
4 Results
The number and ratio of facial expressions are summarized in Table 2. A total of 1,879
facial images were collected; 66.74% (1,254) were smiling, whereas more than 25% were
expressionless.
2
R is a free software environment for statistical computing and graphics. See http://www.r-project.org/.
1 3
Analysis of political gures 1017
Table 2 Number and percentage
of facial images by type
Types
Frowning No expression Smiling Sum
Frequency 154 471 1254 1879
% 8.2 25.07 66.74 100
Table 3 Coding labels for facial
expressions
Expression Label
Frowning 1
No-expression 2
Smiling 3
Other 4
It is widely acknowledged that inter-coder reliability is a critical component of content
analysis: Without the establishment of reliability, content analysis measures are useless
(Neuendorf 2002, p. 141). Thus, we performed an experiment to assess the reliability of the
coding system. Ten randomly selected facial images for the 292 politicians (as of Febru-
ary 16, 2010) were used. The images were downloaded randomly from Naver
3
and coded
according to the three types of facial expressions, resulting in a set of 2,920 images. This set,
together with the classication scheme (Table 1), was given to two coders, who then coded
each image independently. Four numeric labels were used to classify each image (Table 3).
Note that a facial expression other than smiling or frowning was classied as other.
The labeling results for the two coders were then compared. We employed two inter-
rater reliability methods: percent agreement (Riffe et al. 1998) and Krippendorffs Alpha
(Krippendorff 2004). We chose the percent agreement method because it is one of the most
widely used methods. However, previous research has suggested that although the method
is intuitively appealing and simple to calculate, it should be accompanied by a more robust
method (White and Marsh 2006). Krippendorffs Alpha is known to be the most conservative
method (i.e., it produces lower values than other methods); however, its validity depends on
the use of multiple coders, different sample sizes, and missing data (Hayes et al. 2007).
In terms of the three facial expressions, the percent agreement measure showed a 0.71
level of reliability for the coding scheme, which was greater than the accepted 0.70 threshold
(Riffe et al. 1998). As shown in Table 2, whereas smiling images were common, frowning and
no-expression images were relatively rare. Based on this observation, we dichotomized the
classications as smiling and non-smiling to further examine the reliability of our ndings.
Both of the methods suggested supportive indices: Krippendorffs Alpha (=0.81) and percent
agreement (=0.92).
5 Analysis and discussion
5.1 Age distribution
As of time of data collection, the ages of assembly members exhibited a normal-like distri-
bution with a mean/median around 56. Figure 1 displays the normal QQ plot and box plot
3
Naver is the most popular portal/search engine in South Korea. See http://www.naver.com.
1 3
1018 B. Ozel, H. W. Park
Fig. 1 Age distribution of 18th National Assembly members in South Korea (as of September 2009)
of the age distribution. The distribution plots conrmed a normal-like age distribution for the
politicians. The youngest member was 38years old, whereas the oldest was 79. The gender
distribution showed that males dominated the National Assembly; of the 292 members, there
were only 41 female legislators.
5.2 Distribution of facial expressions
The smiling image was the most prevalent facial expression regardless of the politicians
socio-political-demographic attributes (Table 2). The box whisker plots shown in Fig. 2
(which displays the distribution of values attributed to each category of facial expressions)
support this nding. Thus, it further examines the frequency of expressions on a comparative
basis. In other words, it checks the prevalence of each expression for each politician. The
distribution of numbers of smiling images on websites was bell-shaped, with a tail for large
values. Each politician had approximately ve smiling images on average. Expressionless or
frowning images, however, were very rare.
5.3 Experience and facial expressions
We examined the relationship between the age of politicians and their facial expressions. We
took the frequency of each facial expression as its intensity. We rst examined their exact age.
There was no correlation between their age and the existence or intensity of any particular
facial expression. We then grouped the politicians by age (in decades) to determine whether
there were any significant differences in facial expressions between age groups. Although
there were no significant correlations, there was some relationship between age (in decades)
and facial expressions. The box plots shown in Fig. 3 indicates that politicians are opting
to wear a facial expression on their public images on the Web. Younger age groups clearly
preferred smiling images; older groups increasingly preferred frowning images. Based on
these results, we tested whether age (in decades) was related to experience (measured by
the number of terms served). As shown in Table 5, there was a strong relationship between
1 3
Analysis of political gures 1019
Fig. 2 Distribution of values for each facial expression category
Fig. 3 Facial expressions by age (coded in decades, for example, 40s, 50s, 60s, etc). The test result is given
in Table 4, second column
the two. We then examined the relationship between the politicians experience and their
preference for facial expressions. Only smiling faces were correlated with experience. The
more experienced a politician, the more smiling he/she was. Although smiling expressions
generally decreased with age, those politicians with more experience were more likely to use
smiling expressions. This suggests that the more experienced politicians were more likely to
continue their use of smiling images than similarly aged politicians with less experience.
1 3
1020 B. Ozel, H. W. Park
Table 4 Correlation between
facial expressions and Naver
visibility and experience
Naver (Log) Experience
Smiling df 275 275
t 3.2484 2.0934
p-value 0.0013 0.0372
cor. 0.1922 0.1253
Fig. 4 Smiling and frowning pictures according to party afliation
5.4 Gender, hometown, and constituency
We explored whether there were any gender differences in facial expressions but found no
significant difference. Only the hometown of politicians had a significant relationship with
smiling pictures (p-value=0.012). There were no differences in facial expressions in terms of
the politicians original election constituencies. However, there were significant differences
in smiling expressions between Seoul and other regions; the politicians from Seoul were
more likely to use smiling images than those from other regions.
5.5 Party afliation
Figure 4 shows the distribution of smiling and frowning pictures by political party. Although
the distribution seems to suggest differences between political parties, the Kruskal-Wallis
rank-sum test indicates a significant difference (p-value=0.034) only in the frequency of
frowning expressions.
However, when we compared the ruling party to the opposition (all other parties), we
found significant differences. Opposition groups were more likely to use frowning
(p-value=0.0014) or expressionless images (p-value=0.0415) than the ruling party. On the
other hand, the ruling party was more likely to use smiling images (p-value=0.1159) than
opposition groups, although the result was not very significant. Figure 5 shows the results.
1 3
Analysis of political gures 1021
Fig. 5 Ruling party vs. opposition parties
5.6 Visibility and facial expressions
We examined the relationship between the politicians web visibility and their facial expres-
sions. Web visibility was estimated by the number of the politicians in-links and Naver
visibility. The number of links that a politician received (from others) on his or her personal
homepage was used as the number of in-links; the number of links through Yahoo was used.
The number of mentions a politician received on Naver was used for Naver visibility. In
addition, we estimated the size of a politicians website by using the number of the websites
webpages. Again, we used Yahoo to estimate website size.
The rows in Fig. 6 show the distributions of the politicians web presence and visibility
(webpages, mentions on Naver, and in-links). All of the distributions were heavily skewed.
There were a few politicians with very large numbers of in-links, very high Naver visibility,
and very large websites.
We determined whether there were any significant positive relationships between the
parameters. As shown in Table 6, there were positive pairwise correlations. Naver visibility
was significantly correlated with the number of in-links, but it had not that signicant cor-
relation with website size. Note that numbers in the table denote estimated correlations and
stars denote the signicance levels.
Finally, we tested pairwise correlations between facial expressions and web visibility. As
shown in Table 4, rst column there was a positive correlation between smiling images and
Naver visibility. However, there was no significant relationship between website size and
facial expressions or between website size and facial expressions.
6 Conclusions
6.1 Summary
The results indicate that the smiling image was the most prevalent facial expression
on the websites of the politicians regardless of their socio-political-demographic attributes.
1 3
1022 B. Ozel, H. W. Park
Fig. 6 Web visibility
Table 5 Correlation between
experience and age, site size, and
Naver visibility
Age Site Size Naver
Experience df 290 278 290
t 6.84 3.05 3.9
p-value 4.76E-011 0.0025 0.0001
cor. 0.3726 0.1798 0.2234
Furthermore, the existence and strength of the smiling image had a significant positive corre-
lationwiththe politicians webvisibility, but not withtheir website size or in-linknumbers. No
significant relationships were found for the other facial expressions. Opposition parties were
significantly more likely to show frowning or expressionless images than the ruling party,
whereas the ruling party was more likely to be smiling than opposition parties. Although
there were significant differences in facial expressions between the politicians hometowns
and between their election constituencies, there were no gender differences. The level of
political experience was positively correlated with smiling images. This suggests that the
more experienced politicians were more likely to continue their use smiling images than
similarly aged politicians with less experience.
6.2 Limitations and future research
Although this study of facial expressions on politicians websites provided a number of note-
worthy results, it cannot be considered as complete and consistent in its current form. To
1 3
Analysis of political gures 1023
Table 6 Pairwise correlations
between visibility parameters
* p = 0.002, **p << 0.001
Site Size In-link
Naver 0.183* 0.536**
Site size 0.251**
improve the validity of the ndings, three coders can be used for the classication. Further,
to better understand the relationship between online images and socio-political-demographic
attributes of politicians, facial expression categories can be extended.
Lastly, the following research questions are proposed for future research:
Longitudinal analysis: What facial expressions did politicians from opposition parties
make when they were in power?
International comparison: What are the effects of socio-political-demographic attributes
of politicians from different countries on the regionally primed facial expressions con-
veyed online?
Image versus text: How do the expressive content of prole images and the content of
textual narratives found on politicians websites couple or decouple?
Acknowledgements This research was partly supported by the World Class University (WCU) program
through the National Research Foundation of Korea, funded by the Ministry of Education, Science and
Technology (No. 515-82-06574, http://english-webometrics.yu.ac.kr). This paper is part of the Investigat-
ing Internet-based Politics using e-Research Tools Project. The authors are grateful to Byung-Gyu Park,
Chien-Leng Hsu, and Yon-Soo Lim for their assistance and comments in early stages of data collection and
analysis.
References
Angus, E., Stuart, D., Thelwall M.: Flickrs potential as an academic image resource: an exploratory study.
J Librarian Inf Sci (2011, in press)
Bauer, M.: Classical content analysis: a review. In: Bauer, M.W. and Gaskell, G. (eds.) Qualitative researching
with text, image and sound, pp. 131151. Sage, Thousand Oaks, CA (2000)
Choi, S., Lehto, X.Y., Morrison, M.A.: Destination image representation on the web: content analysis of macau
travel related websites. Tourism Manag. 28, 118129 (2007)
Coleman, R., Wasike, B.: Visual elements in public journalism news-papers in an election: a content analysis
of the photographs and graphics in campaign 2000. J. Commun. 54(3), 456473 (2004)
Davis, F.: Perceived usefulness perceived ease of use and user acceptance of information technology. MIS
Quarterly 13(3), 319340 (1989)
Druckman, J., Kifer, M., Parkin, M.: The technological development of candidate websites: How and why
candidates use web innovations. Social Science Computer Review 25, 425442 (2007)
Druckman, J., Hennessy, C.L., Kifer, M.J., Parkin, M.: Issue engagement on congressional candidate websites
(20022006). Social Science Computer Review. 28(1), 323 (2010)
Foot, K., Schneider, S.M.: Web campaigning. MIT Press, Cambridge, MA (2006)
Govers, R., Go, M.F.: Projected destination image online: website content analysis of pictures and text. Inf.
Tech. Tour. 7(2), 7389 (2004)
Hayes, A.F., Krippendorff, K.: Answering the call for a standard reliability measure for coding data. Commun.
Method Measure 1(1), 7789 (2007)
Heaney, M.T., Newman, M.E., Sylvester, D.E.: Campaigning in the Internet Age In: Craig, S.C. (ed.) The
electoral challenge: theory meets practice. CQ Press, Washington (2010)
Hellemans, K., Govers, R.: European tourism online: comparative content analysis of the ETC website
and corresponding national NTO websites. Information and communication technologies in tourism
2005, Proceedings. International Conference. Innsbruck, Austria, pp. 205214. Springer-Verlag, Wien,
New York (2005)
Kim, W.-Y., Hyun, K.-M., Chung, H.-K.: Analyzing visual images of Korean politicians pictures in web sites
(in Korean). Korean J. Cybercommun. 20(4), 4584 (2006)
1 3
1024 B. Ozel, H. W. Park
Kluver, R., Jankowski, N., Foot, K.A., Schneider, S.M.: The internet and national elections: a comparative
study of web campaigning. Routledge, London (2007)
Krippendorff, K.: Reliability in content analysis: some common misconceptions and recommendations. Hum.
Commun. Res. 30, 411433 (2004)
McAllister, M., Turow, J.: Thinking critically about advertising and consumer culture. Routledge, New York
and London (2009)
Neuendorf, K.A.: The content analysis guidebook. Sage, Thousand Oaks, CA (2002)
Page, J.T., Duffy, M.E.: US Presidential Campaign 2008: a fantasy theme analysis of visual stories spun on the
web. Paper presented at the annual meeting of the International Communication Association, Montreal,
Quebec (2008).
Palfrey, J., Gasser, Urs.: Born digital: understanding the rst generation of digital natives. Basic Books, New
York (2008)
Park, H.W., Bae, A.J.: Political use of the internet among the 17th national assembly members in South Korea:
focusing on the characteristics of ofcial website. Digi. Commun. Rev. (in Korean) 3(1), 5793 (2007)
Riffe, D., Lacy, S., Fico, G.F.: Analyzing media messages: using quantitative content analysis in research.
Lawrence Erlbaum, Mahwah (1998)
Verser, R., Wicks, R.H.: Managing voter impressions: the use of images on presidential candidate websites
during the 2000 campaign. J. Commun. 56, 178197 (2006)
White, M.D., Marsh, E.E.: Content analysis: a exible methodology. Library Trends 55(1), 2245 (2006)
1 3