|
Representing frames of texts:
Computer-aided text analysis and
graphical representations of frame salience
____________________________
Submitted to the Division of
Communication Theory and Methodology
__________________
Joohoan Kim
Ph.D. Candidate
The Annenberg School for Communication
University of Pennsylvania
3620 Walnut Street
Philadelphia, PA 19104-6220
Phone: (215) 898-4775 (Office)
Fax/Tel: (610) 519-1506 (Home)
E-mail: [log in to unmask]
Representing frames of texts:
Computer-aided text analysis and
graphical representations of frame salience
(abstract)
This study tries to capture dynamic structural changes of three
different kinds of texts over a 52 week period. We suggested equations for
"degrees of prominence of a frame indicator," or "frame salience scores."
Combining with traditional methods for content analysis such as contingency
analysis and multi-dimensional scaling, framing salience scores can construct a
series of semantic networking maps to be presented with presentations tools such
as HyperCard or Powerpoint.
Representing frames of texts:
Computer-aided text analysis and
graphical representations of frame salience
The purpose of this paper is to explore some methods for graphical
representations of semantic structures of texts. The methods suggested here may,
hopefully, be used as basic algorithms for text analysis computer programs.
Neither do computers understand nor interpret texts in the same
ways that human coders would do (at least as of now). It is also true, however,
that we can still "compute" and draw useful information about semantic
structures of a text. As graphical representations (diagrams and graphs) with
computers will necessarily involve computing numbers, we need to determine,
first, what to compute with given texts, or how to convert word into numbers.
Our suggestion is that we should not concern too much about syntax structures
except using it as a text unit. Rather, by concentrating the structural
relations among key concepts, we believe, we may capture, at least, dynamic
changes of semantic structures over time as well as among different texts (e.g.,
among news stories, public discourse, congressional debates).
It is somewhat surprising how few efforts have been made by
communication scholars to develop theories and methods in computer-aided text
analysis. Since General Inquirer (Stone et al, 1962) was introduced, little, if
any, progresses have been made in text analysis tools, which cannot more than
calculating word frequencies in a given text unit. We cannot, however, ignore
the demands for better computer-aided text analysis any more.
First of all, more and more our research objects exit in the form
of digitized text: for example, most of news texts that we are studying now are
available in the form of digitized database such as Nexis/Lexis. Besides news
stories, all kinds of reports, hearings, debates, and public discourse now exist
in the form of large chunk of digitized texts. In qualitative studies, too, most
field notes and interview data are being stored in computers with pervasive uses
of word processors. Moreover, the Internet has recently begun to be recognized
as one of main research areas in communication studies.[1] And the Internet
itself is a bunch of digitized texts consisting of e-mails, messages on bulletin
boards, web pages, and so on.
In this paper, we compared semantic structures of three different
kinds of texts (news stories, congressional debates, and briefings by government
policymakers) about on a topic, the North Korean nuclear issue. Within each kind
of text, we followed the trends of "frame salience," rather than issue salience,
to compare dynamic changes in the semantic structures. By doing so, this study
would show (1) how to represent "frames" of texts with numbers and graphics and
(2) some empirical evidences for "framing effects."
Why frame analysis?
We believe that the reason why the number of communication studies
based on the method of content analysis had suddenly reduced since early
seventies: it's agenda-setting theory. Agenda-setting theory has argued that
news media "tell people what to think about, not what to think" since the
seminal work of McCombs and Shaw (1972). If we admit that media effects are not
more than of "setting an agenda," then all we need to do is to identify what are
the "salient" issues; we can find those "salient issues" simply by measuring the
length (for example, airing time, inches of newspaper stories, number of words
in news stories) of news stories about a certain issue. In other words,
agenda-setting studies do not require us to investigate meanings of news texts;
they are interested only in the physical characters of news stories such as
amount, location, or "coverage prominence".[2]
Interestingly enough, however, after reviewing the agenda-setting
studies of 25 years, MaCombs and Shaw, the forefathers of agenda-setting
studies, concluded that "media not only tell us what to think about, but also
how to think about it, and, consequently, what to think" (MaCombs and Shaw,
1993:65). This "new" understanding of media effects would provoke, we believe, a
revival of content analysis as a core method of communication studies.
Following Goffman (1974), we define a frame as the "schemata of
interpretation," which enables us to "locate, perceive, identify, and label"
incidents. In foreign policy issues, "frames" would have significant
implications since frames would tell us "how to think," and therefore, what
policy options we should take.
Although this study shares fundamental theoretical assumptions
with traditional agenda-setting studies, it has its own distinctive features:
First, this study traces trends of "frame salience" rather than issue salience.
Our hypothesis is that every issue can have only one "the most salient frame" at
a time among various possible frames. Herbert Simon argued that "[O]f all the
things we know, or can see or hear around us, only a tiny fraction influences
our behavior over any short interval of time." In short, people (journalists +
congressmen + policymakers) would consider only one aspect of an issue at a
given time, since "only one or a very few things can be attended to
simultaneously." This is so-called "bottleneck of attention"(Simon in
Baumgartner and Jones, 1993:104). Just as public issues are competing one
another due to "limited carrying capacity of public arena" (Hilgartner and
Bosk,1988), frames of an issue are also competing one another due to limited
considering capacity of human cognition. To frame-setting studies, therefore, "a
zero-sum theory" (Zhu,1992) and "an integrated model" (Zhu, Watt, Snyder, Yan, &
Jiang,1993 ) would be applicable as well.
Secondly, this study investigates media effects on policymakers'
discourse rather than on general public opinion. Our assumption is that news
media would more "directly" affect policymakers' discourse and opinion (and
presumably their decision making processes, too) than through public opinion,
especially in the case of foreign policymaking. Cook et. al. (1983) can be
considered as an example of empirical research on the "direct effects," which
found some "direct media influence" on (1) policymakers' perceptions of the
importance of the issue; (2) policymakers' perceptions of how the general public
views the importance of the issue; (3) policymakers' belief in the necessity of
policy action. For a theoretical argument based on "the third-person effects
theory," see Lasorsa (1992).
Lastly, this study employs computer text analysis programs (Mohler
and Zuell, 1990; Miller, 1993), and tries to clarify their limitations as well
as usefulness. Basically speaking, even the most sophisticated computer text
analysis programs today cannot do more than "counting" word frequencies. (Well,
sophisticated program can "count" words with more options). For example, see
Roberts and Popping (1993). Fan (1988) seems to ignore such limitations and had
his computer do the jobs beyond its capacity. we will use computers to identify
"frame indicators" and calculate their "prominence" scores.
Data and analysis
Through LEXIS/NEXIS database, we retrieved full texts and
transcriptions of news stories (from MAJPAP file), congressional records (from
RECORD file) and Capitol Hill hearings (from FEDNEW file), and department and
White House briefings (from FEDNEW file) that contain the two search terms
(North Korea and nuclear) within 20 words for 52 weeks period from 10/3/1993.[3]
Then, using TEXTPACK, we filtered out the paragraphs that did not
include the two search term within 20 words. Then we randomly selected 50
paragraphs per week from each of the three groups of the retrieved texts
("News", "Records", and "Briefings"). For Capitol Hill Hearings and Briefings,
which are full transcript texts, each "turn-taking" will be considered as a
paragraph unless it has more paragraphs in it.
Using VBPro, we identified the most frequently appearing
overall time periods (among the whole sample of 2600 paragraphs for
each
of three text sets). Irrelevant words like articles and pronouns
were
ignored. Then, a content analytic dictionary for categorization of
each
of 15 concepts was established.[4] Occurrences rates are presented
in the
Table 1 and 2.
To identify hidden structural relations among the concepts, factor
analysis was employed. The factor matrix for the newspaper concepts is shown in
the Table 3. From each factor, the top 4 concepts were selected and combined:
Factor 1 (NF1) combines IAEA, INTERNATIONAL, SANCTION, and INSPECTION. These
concepts pretty much represent "international" perspectives. We may call it
"international frame." Factor 2 (NF2) combines JAPAN, NORTH KOREA, US, and WAR.
We may call it "war frame," since the concept WAR ranked as the top in the
factor 2. Factor 3 (NF3) combines TALKS, NORTH KOREA, NUCLEAR, and CRISIS. For
the same rationale, we may regard it as "talk frame."
The factor matrix for the briefings and hearings appears on the
Table 4. The combinations of the factors here are a little bit confusing.
However, we may name the factors "war/weapon," (BF1) "China/US," (BF2) and
"Talks/South Korea" (BF3). From this data, we may argue that news stories and
policymakers' discourse have somewhat different semantic structure.
Reliability analysis showed that each four items of the 6 factors
can be combined into new variables: standardized item alpha was .90 (NF1), .77
(NF2), .67 (NF3), .83 (BF1), .77 (BF2), .and 64 (BF3). The correlation
coefficients among the 9 variables (the 6 newly constructed frames, 2 opinion
variables and the number of news stories) are presented on the Table 5. The war
frame of the news stories (NF2) shows strong correlation with "war/weapon" frame
of the briefings and hearings (r=.66, p<.05), and with "Talks/S. Korea" (r=.67,
p<.05). Interestingly enough, the "talk" frame (NF3) of the news stories has
positive correlation with the support for the military action (r=.73, p<.01),
and negative ones with the support of Clinton (r=-.76, p<.01). This implies that
the "talk" frame of the newspapers (NF3) is neither about supporting US's
policies to have talks with North Korea nor about peaceful diplomatic talks with
North Korea. Rather, we may suppose that even in reports of the talks,
newspapers are continuously stressing incredibility of North Korean government
and increasing perceived level of crisis.
Another possible interpretation from the Table 5 is that among
various frames of news stories, "talks" frame (NF3) is more effective on public
opinions, while "war" frame (NF2) is more influential to policymakers' frames of
both "talks" (BF3) and "war" (BF1), on the assumption that newspaper's frame has
some effects both on public opinions and policymakers' discourse; and this
assumption is supported by the previous studies that we have looked into above.
Frame salience
To capture framing effects along time periods, we should be able
to measure trends of frame salience. We would suggest a new conceptual
operationalization: frame indicator's degree of prominence. By "frame indicator"
we mean one or more concepts that may indicate a frame of a text set. To become
a frame indicator, a concept (a word or a cluster of synonyms) should have
higher degrees of prominence than others. we define "prominence" as a function
of "degrees of semantic connectivity" (Carley & Kaufer,1993; Kaufer and
Carley,1993) and rates of appearing frequencies. Every concept can have a degree
of connectivity and a rate of appearing frequencies in a given text set.
Here, we should distinguish a "text unit" from a "text set." Text
unit is a semantically meaningful unit, and it could be a word, a sentence, a
phrase, a paragraph, a chapter, a book, and so on. Text set is a set of text
units that are determined by analytical purpose. In this study, every paragraph
was regarded as a text unit, and randomly selected 50 paragraphs for each week
were dealt with as a text set. Thus, semantic connectivity here would mean,
simply speaking, how a concept is combined with other concepts in each of 50
paragraphs, and rates of appearing frequencies would mean how many paragraphs
contain the concept.
Semantic connectivity consists of the "density, conductivity and
consensus." According to Carley and Kaufer (1993), "density" of a concept means
"the number of links that connect it to other concepts"; "conductivity of a
concept is measured by multiplying the number of concepts linked into it by the
number of concepts linked out from it"; "consensus" could be determined by how
many "language users" agree on the "links." And intersection of these three
dimensions form "a typology of eight semantic categories": for example, when all
the three dimensions are low, the focal concept is "ordinary word," and all the
dimensions are high, it is "symbol," and so on.
We may measure three dimensions of connectivity as following: (1)
"Density" can be easily determined by the number of significant and positive
correlation coefficients with other concepts. (2) To measure "conductivity," we
first should know "directions of the links" ("in-link," or "out-link," or
"two-way link"), which can be determined by comparing rates of appearing
frequencies of each pair of concepts. When a "large concept" and a "small
concept" are connected, the direction can be counted as "in-link connection from
the small one to the large one. Otherwise (either two "small" concepts or two
"large" concepts), they should be regarded as "two-way links." When a concept's
rate of appearance is below the average, the concept can be regarded as "small"
one. Otherwise, it would be counted as a "large" concept. Conductivity score
will be computed by multiplying the number of in-links by that of out-links. (3)
Lastly, for "consensus" score, the mean of correlations coefficients with other
concepts can be used. Frame indicator's prominence scores will be computed for
each group of texts (news, congressional records, and briefings) and for each
time unit (a week). The formula for the prominence score of frame indicator i
in a given text set at a given time would be:
[--- Pict Graphic Goes Here ---]
,
where D is density, C is conductivity, S is Consensus and A is a
rate of appearing frequencies. Then,
[--- Pict Graphic Goes Here ---]
and
[--- Pict Graphic Goes Here ---]
, where
[--- Pict Graphic Goes Here ---]
,
[--- Pict Graphic Goes Here ---]
, Where
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
,
[--- Pict Graphic Goes Here ---]
where
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
, and
[--- Pict Graphic Goes Here ---]
Therefore,
[--- Pict Graphic Goes Here ---]
[5]
Our rationale for incorporating semantic connectivity with frame
indicator is that frame indicators (e.g. "talks," "sanction," "plutonium") would
have more density, conductivity, and consensus when they are in their own
frames, and consequently, they would become more "symbols" rather than "ordinary
words" (Carley & Kaufer,1993:192). For example, the word "talks" (the "political
frame" indicator for this research) would have more symbolic characteristics in
the "political frames" than in other frames.
With this method, frame salience scores for each three text sets
were calculated for 52 weeks, and the trends were presented in the figure 1, 2,
and 3.
For graphical representations of semantic structures
Contingency analysis is a method of investigating associations of
concepts with co-occurrences, which requires construction of "units". "Selection
of units" should be determined by research purposes (Osgood, 1959:61-2). A KWIC
list can be used as such a concordance, or indicators of "co-occurrences," a
listing by word of each word in the text together with its context
(Krippendorff, 1980:122). Weber (1990:44-9) explains the two ways in which KWIC
lists can be used: One is that KWIC lists draw attention to the variation or
consistency in word meaning and usage. The other is that KWIC lists provide
structured information that is helpful in determining whether the meaning of
particular words is dependent on their use in certain phrases or idioms. But the
purpose of employing KWIC lists in this study is none of these; they used as a
unit through which we could determine whether certain concepts co-occurred.
Based on the KWIC lists produced with Conc 1.71 (Thompson, 1992),
15 by 15 contingency matrices were constructed for the 13 weeks (refer to the
appendix). For each week, three tables were made: one is for obtained
(observed) contingencies; the other is for expected (chance) contingencies; the
last is for significance of contingencies (t-value). Following Osgood (1959:64),
significance of contingency was defined as "the deviation of any obtained
contingency from the expected value." For the t-value, following formula was
used:
t =
[--- Pict Graphic Goes Here ---]
,
where Ov is the observed value, Ev is the expected value, and N is
the total number of units. Co-occurrences with the t-values of less than -1 or
greater than 1 represent that they are significantly (less than 5% of chances
alone) deviated from the expected value, which means the pair of the concepts
co-occurred significantly more than chances (positive t-values), or did not
co-occur significantly less than chances (negative t-values).
The observed values were calculated by the number of occurrences
of each pair of concepts divided by the total number of units, and the expected
values by multiplying the chance of the occurrences of one concept by that of
the other concept. To measure "distances" between each pair of concepts,
following formula was used.
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
where, Oii is significance of contingency of concept i, and Oij
is significance of contingency of co-occurrence of concept i and j. Based on
these distance scores, we can represent a semantic map of the news text using
multi-dimensional scaling (MDS) as shown in figure 4. And applying frame scores
to the MDS map, we may get a semantic-networking map such as figure 5. Ideally,
it should be represented here all 3 kinds of maps for each text sets for 52
weeks period, or 156 network maps, but it would be too cumbersome to show them
with a hard copy. However, using a hypertext tools like HyperCard for Macintosh,
we can show them with a "moving slide" which will show a dynamic change of
semantic structure of the texts.[6]
Limitations of the study
The signification of this study is that it shows possibilities of
quantifying frame salience through word frequencies and their relationships.
However, it has certain limitations. First of all, a crucial weak point of this
study is the "categorizing dictionary." we used only nouns. To dig up the
semantic structure more thoroughly, we should be using more refined tools,
possibly considering verbs, adverbs, and adjectives, and that would require some
theoretical framework for syntactical structure.
One of the practical problems is that the data clearing process
with a word processor program, Microsoft Word, is too much time consuming. For
an extended study, we should have written a computer program that converts news
texts into SPSS readable texts automatically.
Problem of unitizing must also be reconsidered. For example, if we
can categorize considering syntactical structure, then we believe we should be
using a sentence as a unit.
Note
[1] ) Refer articles on the special symposium topic, Journal of
Communication, 1996, 46 (1).
[2] ) Watt, Mazza, and Snyder (1993:423).
[3] ) The search terms were: (1) For news: NORTH KOREA! W/20 NUCLEAR!
AND PUBLICATION (NEW YORK TIMES OR WASHINGTON POST OR NEWSDAY OR LOS ANGELES
TIMES OR US TODAY OR CHICAGO TRIBUNE) AND DATE (AFT #/#/199# AND BEF #/#/199#);
(2) For other texts: NORTH KOREA! W/20 NUCLEAR! AND DATE (AFT #/#/199# AND BEF
#/#/199#).
[4] ) The dictionary is as following:
US: US, U.S. America-, United States
SK: South Korea-, S. Korea-, Seoul-, Republic of Korea, ROK
NK: North Korea-, N. Korea-, Pyongyang, DPRK
NUCLEAR: nuclear, nuke
WAR: attack-, war, strike-, military, bellicose, invade-, invasion,
retaliat-, combat-, troops,
CRISIS: threat-, danger-, crisis, tension, fear-, risk
WEAPON: arms, arsenal, bomb-, weapon-,missile-,
SANCTION: sanction-, international sanction-, embargo
TECH: plutonium, light water-, reactor-, fuel rod-, technology
INSPECTION: inspect-, international inspect-,
TALKS: discuss-; talk-, negotiat-, agree-, diplomatic-, diplomacy
IAEANPT: IAEA, International Atomic EnergyAgenc-, NPT-,
Non-Proliferation-,
INTUN: international-, UN, U.N., United Nations, Security Council-
JAPAN: Japan-, Tokyo-
CHINA: China-, Chinese-, Beijing
[5] ) Here is an example for computing prominence scores. At the
second week of June 1994 (36th week), in the news texts, the "war" concept,
which would be later identified as the "coercive frame indicator," had
significant and positive correlations with four concepts of "INTERNATIONAL,"
"JAPAN," "CHINA" and "TALKS." Thus, the density score for the war concept was 4.
The rate of appearance of the war concept (.41) was the highest among the key
concepts, followed by INTERNATIONAL (.17), TALKS (.15), JAPAN (.07), and CHINA
(.06). Since the average rate of appearance of the all concepts was .08, only
INTERNATIONAL and TALKS were counted as large concept. As the war concept had
two "two-way links" (with INTERNATIONAL and TALKS) and two "in-links" (from
CHINA and JAPAN), that is, 4 in-links and 2 out-links, its conductivity was 8.
The consensus was the mean of the correlation coefficients with other concepts:
with INTERNATIONAL=.29, with TALKS=.11, with JAPAN=.14, with CHINA=.15.
Therefore, the consensus score was .29 + .11 + .14 + .15 = .69 Consequently, the
prominence score for the war concept
{(4+8)*.69}+(.41*16) =14.84.
[6] ) With that presentation, the structures of texts were slowing
changing (growing in certain parts and dying in another) as if they were alive.
<Table 1> Occurrence Rates of Key Concepts in the Newspapers (%)
= (Number of stories that contain the concept / No. of all
stories of a sample)*100
Variable Mean Std Dev Minimum Maximum Valid N
________________________________________________________________________
TALKS 48.45 14.62 28.60 80.00 52
IAEANPT 18.90 17.18 .00 50.00 52
INT'L 42.09 23.39 10.00 76.70 52
JAPAN 18.38 7.97 3.30 30.00 52
CHINA 11.84 10.50 .00 30.00 52
N. Korea 97.07 2.40 93.30 100.00 52
S. Korea 34.62 18.08 3.30 73.30 52
NUCLEAR 93.33 6.11 80.00 100.00 52
U.S. 55.64 8.91 40.00 66.70 52
WAR 31.33 12.79 52.30 51.40 52
WEAPON 36.72 23.98 3.30 76.70 52
SANCTION 23.95 22.86 .00 63.30 52
TECHNOLOGY 22.42 20.40 4.20 70.00 52
CRISIS 41.65 9.55 23.30 56.70 52
INSPECTION 20.25 18.73 .00 56.70 52
________________________________________________________________________
<Table 2> Occurrence Rates of Key Concepts in the Briefings and
Hearings (%)
=(Appreance frequencies of the concepts within a sample / No. of
all words of the sample)*100
Concepts Mean Std Dev Minimum Maximum Valid N
________________________________________________________________________
TALKS 1.07 .37 .63 1.90 52
IAEANPT .34 .29 .00 .80 52
INT'L .23 .52 .03 .53 52
JAPAN .14 .12 .00 .37 52
CHINA .30 .30 .00 1.00 52
N. KOREA 1.76 .48 1.17 2.73 52
S. KOREA .32 .19 .52 .70 52
NUKE 1.22 .34 .60 1.73 52
US .63 .34 .33 1.50 52
WAR .30 .22 .03 .80 52
WEAPON .56 .36 .10 1.43 52
SANCTION .20 .20 .00 .63 52
TECHNOLOGY .39 .27 .00 .87 52
CRISIS .22 .15 .00 .50 52
INSPECTION .20 .52 .03 .45 52
________________________________________________________________________
<Table 3> FACTOR MATRIX OF THE CONCEPTS IN THE NEWSPAPERS
______________________________________________________
FACTOR 1 FACTOR 2 FACTOR 3
(INT'L) (WAR) (TALK)
_______________________________________________________________________
TALKS -.53105 .15585 .72026
IAEANPT .88262 -.39097 .03933
INT .93087 -.11545 .04776
JAPAN .02072 .70154 -.61199
CHINA .77403 .26546 -.49036
NK .18856 .66562 .46936
SK -.75282 .14334 -.53383
NUCLEAR .66710 .52048 .30829
US -.09659 .75988 .15978
WAR .05774 .79781 -.21667
WEAPON .38112 .50616 -.17452
SANCT .78947 .35146 .03551
TECH .72239 -.61386 .07443
CRISIS -.06421 .57221 .42899
INSPECT .83844 -.01474 .04736
____________________________________________________________________________
* From each factor, top 4 concepts were selected (bold with
underlines).
FACTOR EIGENVALUE PCT OF VARICANCES
1 5.58 37.2
2 3.79 25.2
3 2.03 13.6
<Table 4> FACTOR MATRIX OF THE CONCEPTS IN THE BRIEFINGS AND
HEARINGS
______________________________________________
FACTOR 1 FACTOR 2 FACTOR 3
__________________________________________________________________________
TALKS -.25495 -.24273 .54217
IAEANPT .57829 -.03150 -.32661
INT'L .31106 -.44376 -.27778
JAPAN .49729 .51906 .15125
CHINA .34720 .82535 -.10045
N.Korea .71021 -.02752 -.32383
S. Korea .19878 .12248 .82097
NUKE .84319 -.19249 .05479
US .25733 .68645 -.51008
WAR .75895 .29238 .36313
WEAPON .72788 -.24129 .48750
SANCTION .35102 -.74991 -.22176
TECH .65368 -.54303 .11029
CRISIS .61959 .39663 .24143
INSPECT .75316 -.19022 -.33097
____________________________________________________________________________
* From each factor, top 4 concepts were selected (bold with
underlines).
FACTOR EIGENVALUE PCT OF VAR
1 4.79257 32.0
2 2.92592 19.5
3 2.15149 14.3
<Table 5> Correlation Coefficients
____________________________________
OPWAR OPCLIN NEWS NF1 NF2 NF3
_______________________________________________________________________________
OPWAR 1.0000 -.8584** .7843** .1009 .1978 .7253**
OPCLIN -.8584** 1.0000 -.7808** -.3231 -.2432 -.7633**
NEWS .7843** -.7808** 1.0000 .4784 .2395 .5475
NF1 .1009 -.3231 .4784 1.0000 -.0212 - .1250
NF2 .1978 -.2432 .2395 -.0212 1.0000 .2728
NF3 .7253** -.7633** .5475 -.1250 .2728 1.0000
BF1 -.1951 -.0711 .0714 .2034 .6572* -.0660
BF2 -.5715* .3324 -.5769* -.0335 -.2265 -.3536
BF3 -.1847 .0620 -.1058 -.3568 .6699* .1535
BF1 BF2 BF3
____________________________________________________________________________
OPWAR -.1951 -.5715* -.1847
OPCLIN -.0711 .3324 .0620
NEWS .0714 -.5769* -.1058
NF1 .2034 -.0335 -.3568
NF2 .6572* -.2265 .6699*
NF3 -.0660 -.3536 .1535
BF1 1.0000 .2677 .6817*
BF2 .2677 1.0000 .0120
BF3 .6817* .0120 1.0000
____________________________________________________________________________
Note: * = (P<.05) ** = (P< .01) ---- 2 Tailed Test
OPWAR: Supports for US's military action to solve the North
Korean problem
OPCLIN: Supports for Clinton's job handling for the North Korean
problem
NEWS: No. of news stories that contin the search words (North
Korea and Nuclear) within the headlines or leads
NF1= IAEANPT+INT'L+SANCTION+INSPECT
NF2= JAPAN+NK+US+WAR
NF3=TALKS+NK+NUCLEAR+CRISIS
BF1=BNUKE+BWAR+BWEAPON+BINSPEC
BF2=BJAPAN+BCHINA+BUS+BCRISIS
BF3=BSK+BTALKS+BWAR+BWEAPON
(NF1 through NF3 are concepts in newspapers and BF1 through BF3
are in briefings and hearings)
<Figure 1>
[--- Pict Graphic Goes Here ---]
<Figure 2>
[--- Pict Graphic Goes Here ---]
<Figure 3>
[--- Pict Graphic Goes Here ---]
<Figure 4>
[--- Pict Graphic Goes Here ---]
<Figure 5>
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
Network map of the concepts for the overall period
(Numbers = average of frame salience scores)
References
Baumgartner, F., & Jones, B. (1993). Agendas and instability in
American politics. Chicago: The University of Chicago Press.
Carley, K., & Kaufer, D. (1993). Semantic Connectivity: An Approach
for Analyzing Symbols in Semantic Networks. Computer Content Analysis and
Mathematical Modeling. New York: Greenwood Press.
Cook, F. L. et al. (1983). Media and agenda setting: Effects on the
public, interest group leaders, policy makers, and policy. Public Opinion
Quarterly, 47 , 16-36.
Fan, D. (1988). Predictions of Public Opinion from the Mass Media:
Funkhouser, R. (1973). The issues of the sixties: An exploratory study
in the dynamics of public opinion. Public Opinion Quarterly, pp. 62-75.
Gamson, W. (1992). Talking Politics. Cambridge: Cambridge Univ.
Press.
Goffman, E. (1974). Frame analysis: An essay on the organization of
experience. New York: Harper & Row.
Iyengar, S., & Simon, A. (1993). News coverage of the Gulf crisis and
public opinion: A study of agenda-setting, priming, and framing. Communication
Research, Vol.20 (June), 365-383.
Kaufer, D., & Carley, K. (1993). Communication at a Distance: The
Influence of Print on Sociocultural Organization and Change. Hillsdale, NJ:
Lawrence Erlbaum Associates, Publishers.
Krippendorff, K. (1969). Models of message: Three prototypes. In G.
Gerbner & others (Eds.), The analysis of communication content: Developments in
scientific theories and computer techniques. New York: John Wiley & Sons.
Krippendorff, K. (1980). Content analysis: An introduction to its
methodolgy. Beverly Hills: Sage.
Lasorsa, D. (1992). Policymakers and the third-person effect. In D.
Kennamer (Eds.), Public opinion, the press, and public policy. Westport,
Connecticut: Praeger.
McCombs, M., & Shaw, D. (1972). The agenda-setting function of mass
media. Public Opinion Quarterly, Vol.36, pp. 176-185.
McCombs, M., & Shaw, D. (1993). The Evolution of Agenda-Setting
Research: Twenty-Five Years in the Marketplace of Ideas. Journal of
Communication, Vol.43 (2), 58-67.
Miller, M. (1993). VBPro: A program for qualitative and quantitative
analysis of verbatim text. Knoxville,TN:
Mohler, P., & Zuell, C. (1990). TEXTPACK PC. Mannheim: ZUMA.
Osgood, C. (1959). The representational model and relevant research
methods. In Ithiel. D. S. Pool (Eds.), Trends in content analysis. Urbana, IL:
University of Illinois Press.
Rand, D. (1995). Concorder 2.01. Montreal: Les Publication CRM,
Universite de Montreal.
Roberts, C., & Popping, R. (1993). Computer-supported Content
Analysis: Some Recent Developments. Social Science Computer Review, 11:3 (Fall),
pp. 283-291.
Stone, P. et al. (1962). The General Inquirer: A computer system for
content analysis and retrieval based on the sentence as a unit of information.
Behavioral Science, 7 , 484-494.
Tankard, J.,Hendrickson, L., & Lee, D.-G. (1994). Using Lexis/Nexis
and other databases for content analysis: Opportunities and risks. Presented at
the annual meeting of AEJMC, Atlanta, Georgia.
Thompson, J. (1992). Conc 1.71: A concordance generator. Dallas, TX:
Summer Institute of Linguistics.
Watt, J.,Mazza, M., & Snyder, L. (1993). Agenda-setting effects of
television news coverage and the effects decay curve. Communication Research,
Vol.20 (No.3), pp. 408-435.
Weber, R. (1990). Basic content analysis (2nd ed.). Newbury Park:
Sage.
Zaller, J. (1994). Politics as usual: The rise and fall of candidate
Perot. Presented at the annual meeting of NES, Philadelphia.
Zhu, J. H. (1992). Issue competition and attention distraction: A
zero-sum theory of agenda-setting. Journalism Quarterly, Vol.69, NO.4.
pp.825-836.
Zhu, J. H.,Watt, J.,Snyder, L.,Yan, J., & Jiang, Y. (1993). Public
issue priority formation: Media agenda-setting and social interaction. Journal
of Communication, Vol.43 (1), pp. 8-29.
|
