Is It a Wall? A Tree? A Rope? Or an Elephant? --
Television News and Ads as Sources of Issue Information
Xinshu Zhao
Assistant Professor
University of North Carolina at Chapel Hill
School of Journalism and Mass Communication
Howell Hall, CB# 3365
Chapel Hill NC 27599-3365
(919) 962-1465
Glen L. Bleske
Ph.D. Student
University of North Carolina at Chapel Hill
School of Journalism and Mass Communication
Howell Hall, CB# 3365
Chapel Hill NC 27599-3365
(919) 933-0382
Steven Chaffee
Professor
Stanford University
Department of Communication
Stanford CA 94305-2050
(415) 723-4611
Paper submitted to the AEJMC annual convention,
Radio-Television Journalism Division
March 1993
The authors thank participants in JOMC 345 Seminar, Fall 1992,
taught by Prof. Phil Meyer, who helped guide this project with
their useful comments. This study is partially funded by an
award from the University Research Council, University of North
Carolina at Chapel Hill.
"It was six men of Indostan
To learning much inclined
Who went to see the Elephant
(Though all of them were blind),
That each by observation
Might satisfy his mind....
...And so these men of Indostan
Disputed loud and long,
Each in his own opinion
Exceeding stiff and strong,
Though each was partly in the right,
And all were in the wrong!"
From The Blind Men and the Elephant
John Godfrey Saxe
ABSTRACT
Is It a Wall? A Tree? A Rope? Or an Elephant? --
Television News and Ads as Sources of Issue Information
Research that has described and compared the informative roles of
televised political news and advertisements has produced
inconsistent results. In some cases, only political ads are
effective information sources, while in other cases, television
news is judged more effective. This study, a replication of
three other studies, explores the relationships between an
audience's political knowledge and its attention to televised
political news and ads. Secondary analyses of results from two
surveys from the 1992 presidential election were used. Results
from a series of heirarchical regressions indicate that
television news is informative, while ads are uninformative.
When compared with the replicated studies, it appears that
televised news is a consistent information source for voters,
while the information value of advertising varies from one
campaign to another. This result contradicts the widely accepted
generalization that televised political ads are more informative
than television news, which is not informative at all.
Is It a Wall? A Tree? A Rope? Or an Elephant? --
Television News and Ads as Sources of Issue Information
In a Hindu fable, five blind men argued about what an
elephant was like. One touched the body, and declared it a wall.
Another grabbed a leg, and claimed it a tree. Another held the
tail, so he argued it was a rope.
The 20-year inquiry regarding television news versus
television ads as sources of issue information in U.S. political
elections may be like that fable. Have researchers seen only
part of the "elephant"? Early research concluded that TV
political ads were informative and TV political news was not.
But was that a mistake? And has it been a mistake to keep
repeating the generalization?
In a series of studies, the authors have tried to answer
some of these questions. In one study, we reported that TV
election news was informative, but ads were not. In trying to
replicate the study, we found that both were informative, but
news was more informative than ads, and in another replication we
found that while both were again informative, ads were more
informative than news.
The objective of this paper is two-fold: 1) to collect as
many pieces of evidence as possible, and 2) then piece everything
together to have an overall picture of the role of television
news and ads in voter learning about the issue stands of the
candidates. We will first report new evidence: the results of
two surveys, one in a three-county urban area and the other
statewide, both conducted during the 1992 Bush-Clinton-Perot
election campaign. More importantly, we will put the pieces
together by comparing those two studies with our three previously
reported studies.
I. Ads vs. News
One of the most commonly repeated generalizations in the
research literature on political mass communication is the
conclusion of Patterson and McClure (1976) that American voters
learn issue information from television advertisements but NOT
from television news. "Network news," they wrote (p. 54) "may be
fascinating. It may be highly entertaining. But it is simply
not informative." They were equally definite about TV
commercials, both as to their ineffectiveness for projecting
candidate images and their effective power in communicating
issues. "Spot ads do not mold presidential images because voters
are not easily misled," they concluded (p. 115). "But where
image appeals fail, issue appeals work" (p. 116). The
"information gain" related to candidates' issue positions
"represents no small achievement" in their view (p. 117).
Synthesizers of the field have readily absorbed the
Patterson-McClure conclusions into textbooks and review chapters.
Kraus and Davis (1981) called it a "controversial but widely
accepted analysis" that "people learned more from television
advertising than from television news" (p. 278). Nimmo (1978,
p.385) cited Patterson and McClure and reported that
"television's political advertising, not news, is the key source
of information."
Diamond (1980, pp.61-62) noted that "these findings were
meant to be as much a criticism of television news as praise of
television commercials." Graber (1989, pp. 195), on the other
hand, argued for the apparent superiority of commercials because
of their "simplicity of content, expert eye-ear appeal, and
repetition of the message." O'Keefe and Atwood (1981, p. 339)
said with a note of surprise that "even campaign commercials
surpassed television network newscasts in providing voters with
knowledge of the candidates' issue stand" (emphasis ours).
Convinced that network news is not as informative as
televised commercials, Just, Crigler and Wallach (1990) decided
not to study television news. They instead concentrated on
commercials and televised debates in their experiment on issue
learning.
The implication is significant. There has been a debate on
whether the American public is rational, a debate considered "of
vital importance for both the theory and practice of democracy"
(Maass, 1966, Forward). Voters' issue awareness or the lack of
it has been a key idea underlying the debate: an informed
electorate implies rationality (Key, 1966; Page & Shapiro, 1992),
the lack of it implies irrationality (Lazarsfeld, 1944; Campbell
et al, 1960). Since a majority of people depend more on
television for their news than on any other media (McLeod &
McDonald, 1985), it is important to know whether voters receive
most of their information from television news, which is supposed
to inform, or from television ads, which are designed to sell.
Patterson and McClure's conclusions, therefore, deserve
careful examination. Support for their conclusions came from two
major sources: content analysis and audience survey. Based on
the content analysis of evening network newscasts and the
televised political ads, Sept. 18 - Nov. 6, 1972, Patterson and
McClure found that ads provided four times more issue information
than news did (as measured by length of time devoted to issue
discussion). This finding was replicated by Kern (1989), who
examined news and ads aired from 6 to 9 p.m. on the three major
networks during the week prior to election day 1984.
Our North Carolina Study (1990), however, argued that
content analysis of this kind, while valuable for other purposes,
does not provide the evidence needed to answer the research
question in hand. The sample of news and ads under both
studies -- prime time on three networks during the last-minute
campaign flurry -- tend to favor the ads while biased against
issue reporting in news. Horse-race reporting and political ads
are concentrated during that period. More issue reporting, news
analysis, and live interviews occur elsewhere -- in weekend, late
night, and morning shows such as 20/20, Nightline, Prime Time
Live, and Face the Nation; and on other channels such as PBS,
CNN, and Headline News; and earlier in the election year during
primaries, conventions and daily news events such as news
coverage of candidate reactions to the Los Angeles riot.
Further, mere message counts fail to take into account the
quality of the message, selective exposure (Hyman & Sheatsley,
1947), and source credibility (Hoveland, Janis, & Kelley, 1953),
all of which should play significant roles in message
effectiveness. Advertising researchers have recognized that
audience involvement level diminishes once advertisements are on
(Krugman, 1965; Webb & Ray, 1979), and viewers don't trust ads
even if they watch. It is possible, then, that one minute of
news is more effective than one minute of ads because the news is
more likely to be watched, attended to, and believed, a
possibility supported by the experimental findings (Salmon, Reid,
Pokrywczynski, & Willett, 1985) that news is more effective than
ads even when the two have identical content. On the other hand,
a counter-argument in favor of ads can be made: one minute of an
advertisement may be more informative than one minute of a news
item because ads are often carefully planned and executed with
intention to sell a well-specified point, and the same ad is
typically repeated many times.
Therefore, the major evidence has to come from audience
research. But the evidence provided by Patterson and McClure
(1976), almost the only evidence supporting the prevailing views
in synthesizing literature, is less convincing than one might
wish. Based on a sample of Onondaga County, New York (Syracuse
and environs), the empirical tests underlying the authors'
categorical generalizations consisted simply of comparing two
groups, high and low exposure, to see if the high exposure group
increased more in subjective certainty of issue awareness. For
news, the high exposure group failed to meet the test; for ads,
the high exposure group did become more certain of their issue
perceptions. The measures of supposedly competing independent
variables -- ads vs. news -- were not comparable. The dependent
variable is self-perceived knowledge rather than actual issue
knowledge. And the analysis is based on raw correlations that
represent individual differences, not controlled tests of a
causal model. Patterson and McClure's findings, therefore, may
be spurious. It may well be, for example, the result of
intellectual deficits in the audience that led to both heavy
reliance on TV for news and lack of certainty about issue
differences between Nixon and McGovern.
Findings from other audience studies contradict Patterson
and McClure's generalizations. Hofstetter, Zukin, and Buss
(1978) used regression to analyze data from a national survey
regarding the 1972 Nixon-McGovern election. They reported that,
when demographic and political variables are controlled, neither
network news nor political ads are associated with more political
information. When all the controls are dropped, however, it
appears that "network news produced almost twice the effect on
information than political advertising" (p. 569).
Our Wisconsin Study (1984) was conducted in Dane County,
Wisconsin, during the Reagan-Mondale presidential race of 1984.
The study used issue knowledge (instead of self-confidence in
issue knowledge) as the dependent variable and attention (instead
of exposure) to news and to ads as the independent variables. We
controlled for various variables including demographics, campaign
activities, and general political knowledge. We reported that
television news was informative, but television ads were not.
In a replication of the 1984 study, the Indiana Study (1988)
found that both television news and ads were associated with
higher issue awareness among respondents in Bloomington, Indiana,
during the Bush-Dukakis race, but news was more effective than
ads in predicting issue knowledge. The North Carolina Study
(1990), conducted in Orange County, North Carolina, during the
Helms-Gantt senate race, again indicated that both news and ads
were associated with issue knowledge, but this time televised
advertising appeared to be more informative than news.
In an ongoing study of the effects of political advertising
on the 1992 election, West, Kern, and Alger (1992) noted that
citizens interpret information they receive directly from
candidates (advertising, debates and press conferences) and from
mediating sources such as newspapers and television. Based on
survey data, content analysis, and focus group interviews from
four different communities during the presidential primaries,
West et al (1992) concluded that, "In contemporary campaigns,
both ads and news are vital to the process by which citizens
construct electoral meanings and interpretations" (p. 23).
West et al's recent work, the Wisconsin Study (1984), the
Indiana Study (1988), and the North Carolina Study (1990) have
not been published, and they, along with the Hofstetter, Zukin
and Buss (1978) study, have yet to make an impact on the
synthesizing literature.
Further, while these five studies have been consistently
inconsistent with Patterson and McClure's generalization, they
themselves have not offered a consistent picture about whether
television news or ads are effective or ineffective, or which one
is more effective than another. An overall picture of the
elephant has yet to be pieced together.
For that purpose, our Wisconsin, Indiana and North Carolina
studies can be used, since each used the same dependent variables
(knowledge rather than self-perceived knowledge), the same
independent variables (attention to television news and ads), and
very similar regression models. Other studies cannot be compared
with ours or with each other at the technical level because they
used different concepts and statistical models (Patterson &
McClure, 1976; West et al, 1992) or did not describe in detail
their measures and models (Hofstetter, Zukin and Buss's, 1978).
But three parts alone may not be sufficient for putting
together an elephant. This paper will add two more pieces.
Also, the problem of generalizeability needs to be addressed --
while each of our three previous studies is from a different time
and location, all of them are conducted during a two-candidate
election and are based on a sample from a college town housing a
leading state university (Madison, Wisconsin; Bloomington,
Indiana; and Chapel Hill, North Carolina).
The two surveys we will report, therefore, were conducted
during the three-way presidential election of 1992 and sampled
two very different populations. One was an urban area of three
counties including some 15 median-sized or small cities/towns
that are adjacent to each other. The more than 525,000 residents
in the area should resemble the average U.S. urban population
better than any of the three university towns we have studied.
The second was a statewide sample, including a large number of
rural residents who have been missing in our previous studies.
The two surveys, conducted three weeks apart during the same
election and using different samples, may reveal another question
the previous three studies could not explore: can the effects of
news and ads vary within the same election?
II. A Survey in a Three-County Urban Area
Methods. We performed a secondary analysis of data
collected the first week of October 1992. The sample included
360 randomly selected voting-age respondents selected by random
digit dialing method for a three-county urban area. Students
from a research methods class conducted the interviews by calling
each number at least three times and asking to speak to the
person in the household over the age of 18 who had the next
birthday.
To be as comparable as possible with the Wisconsin (1984),
Indiana (1988), and North Carolina (1990) studies, we used issue
knowledge as a dependent variable, and we employed independent
measures that referred to attention to both ads and news
specifically related to the presidential campaign. We introduced
a number of control variables to reduce the danger of accepting
spurious correlations as causal evidence. Issue Awareness.
The concept of issue awareness has earned a central position in
political behavior research in recent decades, as party
identification has declined in the American electorate. Policy
voting appears, correlatively, to be on the rise (Nie, Verba and
Petrocik, 1976). Learning how the candidates differ on major
issues of public concern and campaign debate is an obvious
necessary step if people are to live up to Key's (1966) principle
that "voters are not fools." Part of the general theory
underlying freedom of the press has been that it helps to provide
the electorate with competing viewpoints on divisive issues, so
that elections reflect the public will rather than the appeal of
particular personalities.
Eight issue questions measured the respondents' perceptions
of which candidate supported which issue statement, with each
correct answer earning 1 point. The eight questions, which probed
recall of the candidates' stands, are listed in Table 1, and the
correlations among the items are listed in Table 2. Correct
scores from the eight items were summed to create an index of
issue awareness and its distribution is shown in Table 3.
--------------------------
Tables 1, 2, 3 about here
--------------------------
The Cronbach Alpha coefficient for this knowledge variable
is 0.62, lower than the reliability coefficients, ranging 0.75 -
0.82, in the three previous studies. The bigger measurement
error is not surprising, considering that each of the three
previous studies asked respondents from a university town about
two candidates' issue stands, while this survey asked less
educated and politically less attentive respondents about three
candidates' issue stands. Also, as is shown in Table 3, there is
a compensation for the bigger measurement error -- the knowledge
measure has a much less skewed distribution (skewness = -0.436,
as compared with -0.921 to -1.122 in the three previous studies).
While our dependent variable is similar to the one used in
the Indiana Study (1988), it is, however, operationalized
somewhat differently than the dependent variables constructed for
the Wisconsin Study (1984) and the North Carolina Study (1990),
which measured issue knowledge on a Likert scale by asking each
respondent to rate how strongly EACH candidate agreed or
disagreed with a series of policy issue statements. In those two
studies, an index was created that gave plus 1 point when the
respondent correctly placed the candidate's issue stand in
relation to the other candidate, and a negative 1 point if the
respondent reversed the relationship, and 0 points if the
respondent perceived no issue differences.
In the two studies reported in this paper, we will use the
Indiana (1988) study method of awarding points for correct
answers. In reporting the results of the Indiana study (1988), we
argued that if the knowledge measures have clear face validity in
measuring the same concept, the somewhat different
operationalizations add to the quality of the evidence that the
replications and the original studies together can provide. We
believe that if certain relationships are found in each of the
five data sets (the two in this paper and the three studies we
are replicating), the relationships then would appear to be
robust against small variations in measurement instruments.
Attention to TV News and Ads. The major independent
variables of this study are attention to campaign news and
attention to campaign ads. In probing for attention to news, the
question asked, "How much attention, if any, have you paid to
news stories about the presidential campaign when you saw them on
television news?" Four response choices were given: a lot of
attention, some attention, a little attention, or no attention.
To measure attention to ads, the question was, "How much
attention, if any, have you paid to the campaign commercials on
television during the presidential campaign?"
Again, our questions are similar to the questions in the
Indiana Study (1988) but slightly different from the Wisconsin
(1984) and North Carolina (1990) measures, which asked how much
attention respondents paid to EACH candidate's news coverage on
television, and to each candidate' campaign commercials. By
using a total of four questions for the two major independent
variables, those studies have a more reliable measure. Provided
that both all our measures appear to have clear face validity in
measuring the same concepts, we hope that, again, the somewhat
different operationalizations will add to the quality of evidence
produced by all the studies together.
Following the studies we are replicating, these independent
variables refer to both news and ads about the specific political
campaigns--not to news in general vs. ads for candidates, as was
the case in the Patterson-McClure study. We see no reason to
expect other news (e.g., international events, crime and
accidents, weather and sports) to contribute to knowledge of
candidates' issue differences; the inclusion of such "noise" in
their news exposure measure may well have reduced Patterson and
McClure's chances of finding a significant correlation.
Demographics and Voting Characteristics. One of the major
criticisms against Patterson and McClure's (1976) data analysis
has been that they did not control for any extraneous variables
that may produce spurious correlations between the major
independent and the dependent variables (Wisconsin, 1984; North
Carolina, 1990). To remove as much potential spuriousness as
possible from this analysis, we followed the examples of the
Wisconsin (1984), Indiana (1988), and North Carolina (1990)
studies to develop several other variables for control and
comparison.
Education and age have been found to relate to political
communication and knowledge in prior studies and were controlled
for in the studies we are replicating. They were measured in our
study by standard self-report questions.
We will also control for gender, which was not controlled
for in the Wisconsin (1984) study, but was used in the North
Carolina (1990) and Indiana (1988) studies. Gender was coded as
one dummy variable (1 if Female, 0 otherwise). Following the
practice of the studies to be replicated, we will also control
for two dummy variables for voting orientation (Vote for Bush: 1
if planning to vote for Bush, 0 otherwise; Vote for Clinton: 1 if
planning to vote for Clinton, 0 otherwise). In similar fashion,
we also controlled for party affiliation, whom the respondent
voted for in the 1988 presidential race (coded 1 if they voted
for Dukakis, 0 otherwise), and whether respondents reported that
they were registered and planning to vote.
Because of the substantial number of missing values in
income, a dummy variable (coded 1 if the value is missing, 0
otherwise) is entered into the regression equations, and mean
score is substituted for the missing value on the original income
variable, although any constant would produce the same result.
Data Analysis. Following the techniques used in the studies
we are replicating, we tested the correlation between attention
to news/ads with the dependent measure of issue awareness. Table
4 lists the intercorrelations for all the variables, which were
entered into a series of hierarchical regressions. The results
are in Table 5.
--------------------------
Tables 4 and 5 about here
--------------------------
The first equation (Equation 1) in Table 5 is a base model
with issue awareness as the dependent variable and ten control
variables (plus one dummy variable for missing data on the Income
variable) as the independent variables. As shown in Table 5,
several of these control variables make significant contributions
to the explanation of variance even when all the others are
controlled, and together they produce a multiple R-square of
0.273. It appears, then, that Equation 1 is a rigorous basis
against which to assess any further increments to variance in
issue awareness. In effect, it accounts for most of the
individual differences that might create spurious relationships
with our suspected causal variables. Further, because the
multiple R-square of this equation (.273) is very close to its
counterpart R-squares in the Wisconsin Study (.270), the North
Carolina Study (.292), and the Indiana Study (.281), Equation 1
also provides a good basis for comparing the results of all four
studies.
Equation 2 and 3 in Table 5 each adds, alternatively, a
different attention variable onto the basis of Equation 1.
Equation 2 adds attention to news as an independent variable. It
produces a significant 2.04% increment to the variance explained
(incremental F=11.98, df=1/347, p<.001). This result indicates a
striking similarity between our data and the three studies we are
replicating, the 1984 Wisconsin study reported a 2.1% incremental
R-squared due to attention to TV news, the 1988 Indiana study
reported 2.22% and the 1990 North Carolina study reported 2.14%.
Equation 3 in Table 5 substitutes attention to ads in the
same position as attention to news was in Equation 2. It
produces a significant 1.05% increment to the variance explained
(incremental F=11.41, df=1/347, p<.001). The Wisconsin (1984)
finding regarding the ineffectiveness of TV ads on issue
awareness was not replicated in our data -- the counterpart
increment reported in 1984 study was an insignificant 0.8%. The
effectiveness of ads in this study, however, is smaller than the
Indiana Study (1988) finding of an incremental R-squared of 1.5%
and is much smaller than the North Carolina (1990) finding of an
incremental R-squared of 3.1%.
News appears to be more informative than ads in this study
(2.04 vs. 1.05% in incremental R-squared). This finding
contradicts the North Carolina study (1990), which reported that
news appeared to be less informative than ads, as indicated in
the difference in incremental R-squares (incremental R-squares
2.14% vs. 3.10%). Instead, our result is closer to the Wisconsin
study (1984), which reported that television news is more
informative than ads (incremental R-squared 2.1% vs. 0.8%).
III. A Statewide Survey
The Three-County Survey in October extended our previous
college town studies to a larger urban area. Three weeks later
we had an opportunity to extend it further to a statewide sample
from one of the ten most populous states in the nation. This
state, with its largely rural economy, conservative tradition,
relatively low per-capita income and education level, serves well
to balance our previous studies that were limited to more urban,
educated, wealthy and liberal respondents.
Methods. We performed a secondary analysis of data
collected in the last week of October 1992. The sample included
841 randomly selected voting-age respondents selected by random
digit dialing method for a whole state. The survey is semi-
annually conducted as a joint venture of a journalism school and
a social science research institute at a state university.
Students from the journalism school conduct the interviews as
part of their class assignments in reporting and research
methods. Each number is called at least three times and
interviewers ask to speak to the person in the household over the
age of 18 who had the next birthday. Data from the sample
closely matched census demographics for the state on several
indicators.
To be as comparable as possible with the Wisconsin (1984),
Indiana (1988), and North Carolina (1990) studies and Part 1 of
this paper, we used basically the same techniques and variables
as discussed in Part 1 of this paper. We will note the
differences in the following text. Although this is a secondary
analysis of data, we had some limited input into questions that
were placed in the survey. Because of space limitations on the
survey, not all of our requests were filled.
Issue Awareness. Seven issue questions measured the
respondents' perceptions of which candidate supported which issue
statement, with each correct answer earning 1 point. The seven
questions, which probed recall of the candidates' stands, are
listed in Table 6, and the correlations among the items are
listed in Table 7. The distribution of summed scores is shown in
Table 8. We have speculated that more complex three-candidate
questionnaire and less educated respondents may be responsible
for bigger measurement error detected in the three-county survey
presented earlier. If this speculation is correct, we should see
an even lower reliability score in this statewide sample that
includes a large number of even less educated rural residents.
It is indeed lower (Cronbach Alpha = 0.60, vs. 0.62 in the three-
county survey, and 0.75 to .82 in the previous three studies).
As in the three-county survey, the bigger measurement error is
compensated by a even less skewed distribution (Skewness = -
0.269, as compared to -0.436 in the three-county survey and -
0.921 to -1.122 in the three previous studies). The differences
among the five studies in terms of reliability and distribution
of the dependent measures provide an opportunity to test the
robustness of the news/ads effect, provided that we find
something consistent among five studies.
--------------------------
Tables 6, 7, 8 about here
--------------------------
Attention to TV News and Ads. Unlike Part 1, these major
independent variables are measured with two questions as in the
North Carolina (1990) study. Attention to TV campaign news was
based on a pair of questions that asked respondents, "How much
attention, if any, have you paid to televised news stories about
Bill Clinton's (the second question substituted "George Bush's")
stand on policy issues?" Four response choices were given: a lot
of attention, some attention, a little attention, or no
attention. The correlation between the two items is 0.65
(p<.001).
To measure attention to ads, the question was, "How much
attention, if any, have you paid to Bill Clinton's (in the second
question "George Bush's" was substituted) campaign commercials on
television during the presidential campaign?" The correlation
between items is 0.66 (p<.001).
While such measure, like the measurements used in all the
other four studies, has clear face validity in measuring
attention to news and ads, its specific wordings are different
from any of the other studies. Further, because of the limited
space in the questionnaire, we were unable to put in a third pair
of questions that asked about Perot. These variations and
shortcomings should provide a robustness test of news/ads effect
against changes in operationalization, when the result of this
study is compared with other four studies.
Demographics and Voting Characteristics. Among the control
variables to be used in this study, three (age, education, and
voting for whom) were controlled for in the other four studies;
three (income, gender, and likelihood to vote) were controlled
for in three of the other studies; and one (party ID) was used in
the Indiana survey and the 3-county survey. Based on our
practice of introducing a few new controls each time, we included
employment status (working full time, party time, or
unemployed/not working) and political orientation
(conservative/liberal) for controls. These two appear
appropriate considering that it was a recession year and the
state has a conservative tradition. Since there are 81 missing
cases in the income measure, a dummy variable was created for the
missing cases (1 if missing, 0 otherwise) and their values in
original income measure recoded to the mean.
Data Analysis
Table 9 lists the intercorrelations for all the variables,
which were entered into a series of hierarchical regressions,
which are summarized in Table 10.
--------------------------
Tables 9 and 10 about here
--------------------------
The first equation (Equation 1) in Table 10 is a base model
with issue awareness as the dependent variable and twelve control
variables (plus one dummy variable for missing data) as the
independent variables. As shown in Table 10, several of these
control variables make significant contributions to the
explanation of variance even when all the others are controlled,
and together they produce a multiple R-squared of .271. The
multiple R-squared is very close to its counterpart R-squared in
the Wisconsin Study (.270), the North Carolina Study (.292), the
Indiana Study (.281), and Part 1 of this study (.273).
Equation 2 and 3 in Table 10 each adds, alternatively, a
different attention variable onto the basis of Equation 1.
Equation 2 adds Attention to News as an independent variable. It
produces a significant 2.25% increment to the variance explained
(incremental F=25.513, df=1/803, p<.0001). This result indicates
a striking similarity between our data and the three studies we
are replicating, the 1984 study reported a 2.1% incremental R-
square due to attention to TV news, the 1990 study reported
2.14%, the 1988 study reported 2.22%, and Part 1 of this paper
reported 2.04%. The similar attention measures in the five
studies produced almost identical incremental R-squared, an
indication that not only TV news may have been effective in these
four elections in five places, but even the magnitudes of the
effects may have been about the same.
Equation 3 in Table 5 substitutes Attention to Ads in the
same position as Attention to News was in Equation 2. It
produces a 0.34% increment to the variance explained, and it
barely failed the conventional 0.05 test (incremental F=3.733,
df=1/803, p<.0537). Since this study has by far the largest
sample among the five studies (818 as compared with 252 to 416 in
other four studies), the p-value indicates very small, if any,
explanatory power of attention to ads. Indeed, the 0.34%
incremental R-squared due to ads is by far the smallest among the
five studies (Wisconsin: 0.8%; Indiana: 1.5%; North Carolina:
3.1%; Three County: 1.05%)
News is much more informative than ads in this study (2.3%
vs. 0.4% in incremental R-squared). This finding contradicts the
North Carolina study (1990), which reported that news appeared to
be less informative than ads, as indicated in the difference in
incremental R-squared (2.14% vs. 3.10%). Instead, our result is
more consistent with the Wisconsin study (1984), which reported
that television news is more informative than ads (incremental R-
squared 2.1% vs. 0.8%), which added an insignificant increment to
the multiple R-squared.
The difference in ad effectiveness between the three-county
survey (significant incremental increase in R-squared of 1.05%)
and the statewide survey (insignificant increase in R-squared of
0.337%) is important. We notice two possible contributors: the
sample and the time of the surveys. The three-county survey
concentrates on urban population and the statewide survey
includes a large number of rural residents. It is possible that
urban residents, having more access to more diverse sources of
information, can get more out of political advertisements than
rural residents.
Also, the two surveys were conducted three weeks apart --
the three-county survey at the first week of October and the
statewide survey at the last week of October. Because of the re-
entry of Perot immediately before the three-county survey, all
three candidates found it necessary to spend some air time
defending and re-defining their and their opponents' issue
positions. When the campaign was close to the end and changing
voters' perception about candidates issue stands became more
difficult, the candidates, particularly the underdogs, put more
emphases on non-issue themes. Bush's ads talked about "trust",
while Perot's ads tried to respond to the claims that it would be
"risky" to put him in the White House. Also, the effectiveness of
advertising campaign had to hurt by the chaos and lack of focus
in the Bush campaign during October (Wines, November 29, 1992).
IV. Putting Everything Together
Now it's time to put the parts together to look at whole
shape of the elephant. Table 11 summarizes the major findings
from the five studies. It first displays the basic information
about the surveys and elections (Part 1). Then it lists which
control variables were used in which studies and the information
about the control blocks (Part 2), including the R-squared due to
controls (Part 3). After we compare the independent and
dependent measures and the basic statistics related to them (Part
4 and 5), Part 6 displays incremental R2s due to news and ads.
Since those incremental R2s have been used in each of the five
studies as the indicators of effects, we plotted them in Figure
1.
--------------------------------
Table 11 and Figure 1 about here
--------------------------------
As is shown in Figure 1, a pattern seems to emerge: news has
not only been consistently informative, but the magnitudes of the
effects as measured by incremental R-squares have been remarkably
stable, 2.1% in 1984 in Wisconsin, 2.2% in 1988 in Indiana, 2.14%
in 1990 in North Carolina, 2.04% in Three-County Survey, and
2.25% in the Statewide Survey. As is shown in Table 11, the
basis of this pattern appears solid. First, measurements are
consistent. All the five studies measured attention to TV news
as the independent variables and issue knowledge as the dependent
variables. Second, the statistical models are comparable.
Although each study employed some different control variables,
there are clearly more overlaps than differences in the control
blocks. More importantly, the total R2s due to controls have
remained stable across five studies (27.0%, 28.1%, 29.2%, 27.3%,
27.1%).
Yet there are plenty of variations among the studies, and
the impact of news appears robust against all those variations,
including place (Wisconsin, Indiana, or North Carolina), election
(senatorial and presidential), sample (university towns, urban,
and statewide), sample size (252 to 818), time of survey (before
election or after election), specific control variables (four of
the studies have at least one unique control variable),
operationalization of the dependent measures (relative position
or matching) and the independent measures (dummy variables, 1
item, or 2 items), reliability (.60 to .82) and skewness
(-.27 to -1.12) of the dependent variables (although the two
relatively low reliability scores are clearly compensated by the
relatively better skewness scores).
Equally impressive in Figure 1 is that the effect of the ads
jumps up and down from one study to another -- an insignificant
0.8% in 1984 in Wisconsin, 1.5% in 1988 in Indiana, 3.1% in 1990
in North Carolina, 1.1% in 1992 in Three-County area, and an
insignificant 0.3% in 1992 in the Statewide survey. Obviously it
is these jumps that caused the distinct findings from five
studies regarding the relative strengths of news vs. ads.
The variations cannot be easily explained as artifact due
to variations in research instruments. The question wordings
regarding news and ads are strictly comparable within each study,
although they differ slightly among studies. Since the
performance of news appears remarkably stable despite the
differences in research instruments, it is unlikely that the
instrument variation would cause such a dramatic changes in
statistics associated with advertising. The variations, it
appears, are results of variations in advertising effectiveness
from election to election and from place to place.
The Issue of Multicollinearity. Throughout the five
studies, we have employed a consistent testing procedure --
alternately entering independent variables on top of a control
block. Since the two major independent variables, attention to
news and attention to ads, are moderately correlated with each
other in all five studies, a question arises -- should we also
enter the two variables simultaneously?
Results from such tests were summarized in Part 7 of Table
11. Attention to news was a significant predictor of knowledge
in three of the five studies, while attention to ads was a
significant predictor in one of the studies. In one of the
studies ads outperformed the news, while the news outperformed
ads in the other four times.
The problem of multicollinearity, however, makes it
difficult for us to get more out of those statistics. When the
incremental R2 of ads goes up and down (Part 6 of Table 11), the
incremental R2 of news goes down and up in the opposite direction
(Part 7 of Table 11), purely as a result of partialling out
variances. What makes the interpretation even more complicated
is that the correlation between attention to news and attention
to ads also fluctuates from study to study. And, the degree of
freedom associated with the independent variables is also
different in the Wisconsin study (because of dummy coding), which
further complicated the interpretation of the P tests presented
in Part 7 of Table 11. Apparently, entering these independent
variables at the same time is not very informative -- it tends to
obscure the real pattern that appeared so clear when we entered
the two variables alternately.
V. Discussion
In the 1980s and early 1990s, television news appears
consistently informative; it appears as informative in one
election or place as it is in another. On the other hand,
effectiveness of advertising varies from election to election and
from place to place -- it can be more or less informative than
news, or not informative at all.
While contradicting Patterson and McClure's generalization
(1976), this conclusion does not sound surprising if one
considers how news and ads are managed in the United States.
While campaign managers might have some control over structuring
a campaign message for news dissemination, the broadcasting of
that message on news shows is mediated by gatekeepers who
function under professional constraints of balance, fairness, and
the professional ideal that news shows have a responsibility to
inform the public about election issues. While media critics may
complain about campaign media blitzes, sound bites and spin
doctors who massage messages (Kern, 1989), television news shows
may offer a constant amount of issue content that helps audiences
learn about the issues. Television news --driven by the battle
for ratings-- may entertain and focus on campaign tactics or even
be dominated at times by non-issue election content, yet it makes
sense that over the long run, television news shows from CNN to
talk shows provide a great amount of easily accessible and
salient issue content.
If the informativeness of televised news tends to be
relatively stable, it makes even more sense that the
effectiveness of televised ads to inform audiences would vary
greatly. Unlike the news, which is in the hands of hundreds or
thousands of news people, ads are most often controlled by a
small group of strategists or even a single candidate or campaign
manager. A higher concentration of decision making power may
easily lead to greater variations in content, sophistication,
production, frequency, relevance, involvement and other
attributes that have been shown to contribute to audience
learning about issues. Ads should not be expected to be
informative if politicians focus on flag waving, yet they should
be able to inform if politicians load their ads with issue
content.
The sophistication of the audience may have also played a
role. The purpose of news coverage is to inform with accuracy.
Different audience members, politically sophisticated or not,
should all benefit from it. The purpose of political advertising
is to sell candidates. And, unlike commercial advertising that
is restricted in many ways by consumer protection laws, political
advertising in the United States is protected by law as a form of
free speech. Quite often candidates do use advertising to
exaggerate and misrepresent their opponents' positions. A
politically sophisticated viewer may still learn from such ads
about candidates' issue positions. A less motivated viewer may
not, and may even be misled or confused.
The influential Patterson and McClure study (1976) was not
pictured in Figure 1 because, as we explained, they did not use
the same measures and models, so their study is not technically
comparable with ours. How that study would have fit into Figure
1 is difficult to say -- we cannot go back to 1972 to ask
residents of Onondaga County new questions. But, looking at
Figure 1, we may guess what would have happened -- 1972 and
Onondaga may be one of those years and places that advertising
was particularly informative, and the measurements and
statistical models may have mis-estimated the effect of news.
We all have limited information. In that sense all of us,
including the best researchers, are blind. Like the blind men of
the Hindu fable, we must make judgements on the basis of the
limited information. So we propose, with much caution, that our
five studies may have captured the general picture of the
elephant -- the impact of TV news and ads on issue awareness of
U.S. voters in the 1980s and early 1990s. Yet we would not be
shocked if tomorrow another piece of evidence emerges that draws
a different picture. And we are very curious to see if this
elephant in any way resembles other elephants -- are our findings
generalizable to other times and places?
Other important questions should be investigated. Why and
under what conditions should television ads be informative or
uninformative? Why is television news consistently informative?
Could it be uninformative in some situations? Could it be even
more informative? To some of these questions, we have offered
some thoughts as tentative answers. Diverse research tools such
as content analysis, audience research, and historical study are
needed to test the validity of these answers and other possible
answers.
ENDNOTES
REFERENCES
Campbell, A., P.E. Converse, W.E. Miller, and D.E. Stokes (1960).
The American Voter. New York: Wiley.
Diamond, E. (1980). Good News, Bad News. Cambridge: MIT Press.
Drew, D. and D. Weaver (1991). "Voter learning in the 1988
presidential election: Did the debates and the media matter?"
Journalism Quarterly, 68, 22-37.
Graber, D. (1984). Mass Media and American Politics, 3rd. ed.
Washington: Congressional Quarterly.
Hofstetter, R., C. Zukin, and T. Buss (1978). "Political imagery
and information in an age of television." Journalism
Quarterly, 55, 562-69.
Hovland, C.I., I. Janis, and H. Kelley (1953). Communication and
Persuasion. New Haven, CT: Yale University Press.
Hyman, H.H., and P.B. Sheatsley (1947). "Some reasons why
information campaigns fail." Public Opinion Quarterly, 11,
412-423.
Just, M., A. Crigler, and L. Wallach (1990). "Thirty seconds or
thirty minutes: What viewers learn from spot advertisements
and candidate debates." Journal of Communication, 40(3), 120-
133.
Kern, M. (1989). 30-Second Politics: Political Advertising in the
Eighties. New York: Praeger.
Key, V. O. (1966). The Responsible Electorate. Cambridge: Harvard
University Press.
Kraus, S., and D. Davis (1981). "Political debates." In D. Nimmo
and K. Sanders (eds.), Handbook of Political Communication.
Beverly Hills: Sage, Ch. 10.
Krugman, H.E. (1965). "The impact of television advertising:
learning without involvement," Public Opinion Quarterly, 29,
349-356.
Lazarsfeld, P., B. Berelson, and H. Gaudet (1944). The People's
Choice. New York: Duell, Sloan and Pearce.
Maass, A. (1966). "Forward." In V.O. Key Jr., The Responsible
Electorate. Cambridge: Harvard University Press.
McLeod, J., and D. McDonald (1985). "Beyond simple exposure:
Media
orientations and their impact on political processes."
Communication Research, 12, 3-34.
Nie, N., S. Verba and J. Petrocik (1976). The Changing American
Voter. Cambridge: Harvard University Press.
Nimmo, D. (1978). Political Communication and Public Opinion in
America. Santa Monica CA: Goodyear.
O'Keefe, G., and L.E. Atwood (1981). "Communication and election
campaigns." In D. Nimmo and K. Sanders (eds.), Handbook of
Political Communication. Beverly Hills: Sage.
Paige, B., and R.Y. Shapiro (1992). The Rational Public: Fifty
Years of Trends in Americans' Policy Preferences. Chicago:
University of Chicago Press.
Patterson, T., and R. McClure (1976). The Unseeing Eye: The Myth
of Television Power in National Elections. New York: Putnam's.
Salmon, C.T., L.N. Reid, J. Pokrywczynski, and R.W. Willett
(1985). "The effectiveness of advocacy advertising relative
to news
coverage." Communication Research, 12:4, 546-567
Webb, P.H., and M.L. Ray (1979). "Effects of TV clutter." Journal
of Advertising Research, 9 (3), 7-12
West, D.M., M. Kern, and D. Alger (1992). "Political advertising
and ad watches in the 1992 presidential nominating campaign."
Paper presented to the American Political Science Association,
Chicago, IL.
Wines, M. (Nov. 29, 1992). How Bush lost: For want of a strategy,
chaos ruled. The New York Times, Sect. A, pp. 1, 11.
Table 1
Frequency Distribution of Issue Awareness Items: 3-County Survey
(n=360)
Which candidate, George Bush, Bill Clinton or Ross Perot is more
likely to favor the following statement:
Income Taxes: Taxes should be increased only for the richest
Americans?
Correct (Clinton) 76.9% Wrong 16.7% Don't know 6.4%
Gas Tax: Federal Gasoline taxes should be increased by 50 cents
to pay for building new roads and bridges?
Correct (Perot) 40.3% Wrong 44.1% Don't know 15.6%
Ozone Protection: The United States should go slow in cutting
emissions to protect the Ozone layer?
Correct (Bush) 63.9% Wrong 16.6% Don't know 19.4%
Abortion: A constitutional amendment should ban abortions except
in cases where a mother's life is in danger?
Correct (Bush) 78.9% Wrong 11.7% Don't know 9.4%
Budget deficit: People with incomes higher than $25,000 should
pay income taxes on their Social Security benefits to help cut
the budget deficit?
Correct (Perot) 30% Wrong 50.8% Don't know 19.2%
Capital gains: The capital gains tax should be cut in half?
Correct (Bush) 51.4% Wrong 30.3% Don't know 18.3%
Military spending: The defense budget has been cut as much as it
should?
Correct (Bush) 82.8% Wrong 9.7% Don't know 7.5%
National service: The government should pay college costs for
young people who are willing to repay the debt with public
service?
Correct (Clinton) 67.5% Wrong 19.4% Don't know 13.1%
Table 2
Issue Item Pearson Correlation Matrix
Three-County Survey
TAX GAS OZONE ABORT BUDGET CAPITAL
GAS .14**
OZONE .26*** .24***
ABORT .20*** .18*** .23***
BUDGET .14** .36*** .15** .08
CAPITAL .19*** .24*** .18*** .15** .15**
MILITARY .19*** .10* .21*** .25*** .07 .13*
SERVICE .14** .12* .16** .18*** .03 .08
----------------------------------------------------------------
MILITARY
SERVICE .19***
ALPHA = .62
n = 360 1-tailed significance * = <.05 ** = <.01 *** =
<.001
Table 3
Frequency Distribution of Issue Awareness Variable
(Standardized)
Value Freq. Percent
----------------------------
0 9 2.5
0.125 5 1.4
0.25 29 8.1
0.375 36 10.0
0.5 63 17.5
0.625 64 17.8
O.75 78 21.7
0.875 49 13.6
1.0 27 7.5
----------------------------
Mean: 0.615
Standard Deviation: 0.237
Skewness: -0.43631
Kurtosis: -0.26461
n=360
Note: Values are correct number of answers for the 8 issue
questions (see Table 1) with 1 point for each correct answer,
transformed to a standardized scale.
Table 4
Correlation Matrix for All Variables: Three-County Survey
1 2 3 4 5 6 7 8
1. Issue
Awareness
2. Educ. .41**
3. Repub. .02 .03
4. Democrat -.01 -.04 -.49**
5. Age -.08 -.06 .02 .06
6. Gender -.18** -.19** -.11 .13 .13
7. Income .23** .31** .16* -.15* .07 -.04
8. Will Vote .21** .20** .07 .16* .17 -.00 .15*
9. Vote for -.05 .01 .56** -.32** .08 -.08 .08 .16*
Bush
10 Vote for .20** .09 -.45** .53** .01 .04 -.11 .34**
Clinton
11 Vote for .22** .19** -.29** .41** .14* .04 -.00 .26**
Dukakis ('88)
12. Attentn .19** .03 .02 .12 .02 -.02 .03 .12*
to TV NEWS
13. Attentn .06 -.12 .07 .15* .01 .04 -.04 .11
to TV ADS
----------------------------------------------------------------
9 10 11 12
10 Vote for -.46**
Clinton
11 Vote for -.26** .49**
Dukakis
12. Attentn -.04 .14* .09
to TV NEWS
13. Attentn .08 .03 .01 .46**
to TV ADS
n = 360 1-tailed significance * = .01 ** = .001
Table 5
Issue Awareness by Communication Variables
Controlling for Demographics and Voting Characteristics
in a Three-County Area
(Hierarchical Regression)
Eq. 1 Eq. 2 Eq. 3
Dependent Variable Issue Awareness
--------------------------------------------------------------
Independent Var.
Age -.06 -.06 -.05
Gender (Female) -.10* -.09* -.10*
Education .29** .29** .31**
Income .13** .13* .13**
Income Dummy -.14** -.14** -.14**
Republican .05 .03 .04
Democrat -.09 -.11 -.11
Will Vote .08 .07 .07
Vote for Bush -.03 -.03 -.04
Vote for Clinton .15* .13 .15*
Vote for Dukakis .13* .13* .13*
in 1988
Attention News .15**
Attention Ads .11*
-----------------------------------------------------------------
Total R Squared .273** .293** .283**
Incremental R Squared .0204** .0105*
due to Communication
variables
* p<.05 ** p <.01 n=360
Note: Entries are standardized Beta weights. An empty cell
indicates that the independent variable is not entered into the
equation.
Table 6
Frequency Distribution of Issue Awareness Items: Statewide Survey
(n=845)
Which candidate, George Bush, Bill Clinton or Ross Perot is more
likely to favor the following statement:
Income Taxes: Taxes should be raised for those households who
earn more than $90,000 a year?
Correct (Perot) 13.5% (Clinton) 64.9% Wrong 13.6% Don't know
7.9%
Gas Tax: The federal budget deficit should be reduced by
imposing a 50 cent per gallon increase in gasoline tax over five
years?
Correct (Perot) 63.8% Wrong 27.8% Don't know 8.4%
Oil drilling: More areas should be opened for oil drilling?
Correct (Bush) 81% Wrong 7.5% Don't know 11.5%
Abortion: A constitutional amendment should ban abortions except
in cases where a mother's life is in danger?
Correct (Bush) 64.2% Wrong 24.2% Don't know 11.5%
Health care: The nation should have universal health care paid
for by employers?
Correct (Clinton) 63.1% Wrong 25.6% Don't know 11.2%
Capital gains: The capital gains tax should be cut in half?
Correct (Bush) 37.8% Wrong 43.3% Don't know 18.9%
National service: The government should pay college costs for
young people who are willing to repay the debt with public
service?
Correct (Clinton) 55.8% Wrong 28.5% Don't know 15.7%
Table 7
Issue Item Pearson Correlation Matrix
Statewide Survey
TAX GAS OIL ABORT CARE CAP
GAS .23
OIL .13 .18
ABORT .18 .24 .20
CARE .10 .09 .21 .20
CAP .16 .22 .16 .17 .13
SERVICE .18 .29 .19 .12 .14 .20
ALPHA = .60
n = 845
Table 8
Frequency Distribution of Issue Awareness Variable
Statewide Survey
Value Freq. Percent
----------------------------
0 30 3.6
1 48 5.7
2 83 9.8
3 158 18.6
4 158 18.7
5 169 20.0
6 119 14.0
7 80 9.5
----------------------------
Mean: 4.07
Standard Deviation: 1.811
Skewness -0.269
Kurtosis -0.572
n=845
Note: Values are correct number of answers for the 7 issue
questions (see Table 6) with 1 point for each correct answer.
Table 9
Correlation Matrix for All Variables: Statewide Survey
1 2 3 4 5 6 7 8
1. Issue
Awareness
2. Attent. .06
to TV Ads
3. Attent. .24** .50**
to TV News
4. Age -.06 -.04 .06
5. Gender -.24** .02 .01 .07
6. Educat. .39** -.02 .12** -.16** -.07
7. Work .11* -.02 .04 -.47** -.24** .17**
8. Income .31** .04 .11* -.04 -.18 .39** .21**
9. Repub. .11** .01 -.02 -.11* -.02 .04 .05 .08
10 Democ. -.09* .11* .10* .11** .14** -.05 -.08 -.11**
11 Liberal -.01 -.02 -.01 -.07 .12** -.03 .01 -.12**
12 Will Vote .29** .11** .25** .23** -.03 .28** .02 .23**
13 Vote for .11* .12** .02 .03 -.07 .06 .02 .14**
Bush
14 Vote for .10* .01 .16** .10* .11* .05 -.01 .01
Clinton
15 Vote for .07 -.05 .00 -.05 -.14** .10* .05 .05
Perot
-----------------------------------------------------------------
9 10 11 12 13 14
10 Democ. -.54**
11 Liberal -.33** .29**
12 Will Vote .07 .08* -.10*
13 Vote for .46** -.36** -.43** .30**
Bush
14 Vote for -.38** .47** .32** .27** -.45**
Clinton
15 Vote for -.03 -.13** .04 .11** -.24** -.23**
Perot
n = 817 1-tailed significance * = .01 ** = .001
Table 10
Issue Awareness by Communication Variables
Controlling for Demographics and Voting Characteristics
in a Statewide Survey
(Hierarchical Regression)
Eq. 1 Eq. 2 Eq. 3
Dependent Variable Issue Awareness
--------------------------------------------------------------
Independent Var.
Income .10** .09** .10**
Inc. dummy -.07* -.05 -.07*
Age -.02 -.02 -.01
Gender(Female) -.20*** -.20*** -.20***
Education .28*** .27*** .28***
Work -.04 -.03 -.04
Int. to Vote .13** .09* .13**
Republican .06 .06 .06
Democrat -.06 -.07 -.07
Liberal .08* .08* .08*
Clinton .15** .14** .15**
Bush .10* .10* .09
Perot .05 .05 .05
Att. TV NEWS .16***
Att. TV ADS .06
-----------------------------------------------------------------
Total R Square .271** .293** .274**
Incremental R Square .02246 .00337
due to Communication
variables
* p<.05 ** p<.01 *** p<.001 n=818
Note: Entries are standardized Beta weights. An empty cell indicates
that the independent variable is not entered into the equation.
Table 11
Effects of News and Ads on Issue Knowledge
-- Summarizing Results of Five Studies
Studies Wisc. IN NC 3-County State
---------------------------------------------------------------------
1. Sample U.Town U.Town U.Town Urban Statewide
Year 1984 1988 1990 1992 1992
Time of Survey Before Before After Before Before
Election Pres. Pres. Senate Pres. Pres.
Number of Candidates 2 2 2 3 3
Number of respondents 416 252 318 360 818
---------------------------------------------------------------------
2. Controls
Age -** - -** - -
Gender =* = = =***
Race =
Education +* +** +** +** +***
Income + + +* +**
Class +
Employment status -
Self-report knowledge +**
General knowledge +**
Party ID = = =
Pol. orientation (Lib) +*
Campaign interest +
Campaign activity +**
Intention to vote +** + + +**
Vote for/against =
Vote for whom = = =* =* =*
Vote for whom last time =*
---------------------------------------------------------------------
3. Total # of Controls 9 8 12 11 13
R2 due to controls (%) 27.0** 28.1** 29.2** 27.3** 27.1**
---------------------------------------------------------------------
4. Knowledge measure Relat. Match Relat. Match Match
Number of Questions 6 11 7 8 7
Crobach Alpha .75 .79 .82 .62 .60
Skewness -1.12 -1.04 -0.92 -0.44 -0.27
---------------------------------------------------------------------
5. Independent measures 3-dummy 1-item 2-item 1-item 2-item
Operationalization: Attn. Attn. Attn. Attn. Attn.
---------------------------------------------------------------------
6. ~R2 due to news (%) 2.1** 2.2** 2.1** 2.0** 2.2**
~R2 due to ads (%) 0.8 1.5** 3.1** 1.1** 0.3
---------------------------------------------------------------------
7. ~R2 due to news after
controlling ads (%) 1.5 1.1* 0.4 1.2** 1.9**
~R2 due to ads after
controlling news (%) 0.1 0.4 1.4* 0.2 0.0
Table 11 (continued)
* : p<.05
**: p<.01
***: p<.001
Sample.
U.Town: Sample from a University town
Urban: Sample from Urban Area
Statewide: Statewide sample
Time of Survey.
Before: Shortly before the election day
After: Immediately after the election day
Controls
+: control had a positive Beta in base equation
-: control had a negative Beta in base equation
=: direction is relative for categorical variable
empty cell: control not used in this study
Knowledge Measure.
Relat.: Relative position; 1 point if a Republican cadidate is
seen as to the right of Democrat candidate, 0 point if
two are seen as having the same position, -1 otherwise.
Match: Matching; 1 point if the right candidate(s) is named
for the right statement, 0 point otherwise.
Independent Measures.
3-dummy: Three dummy variables for attention to news and three
dummy variables for attention to ads.
1-item: One question for attention to news and one question for
attention to ads.
2-item: Two questions for attention to news and two questions
for attention to ads.
|
