Communication Research, June 1997 v24 n3 p197(37)

               The role of political sophistication in learning from news;
           measuring schema development.
               June Woong Rhee; Joseph N. Cappella.

          Full Text: COPYRIGHT 1997 Sage Publications Inc. Measuring Schema Development

          Accounting for learning from news has focused on media attention, media gratification, content-specific exposure,
          and prior knowledge. One measure of prior knowledge is political sophistication, measuring components of both
          knowledge and ideology. Two studies evaluated the reliability, concurrent validity, and comparative predictive
          validity of political sophistication in comparison to those of exposure, attention, and civics knowledge. Study 1
          focused on the 1993-1994 health care reform debate. Those higher in political sophistication learned more from
          broadcast and print news, and had more differentiated constructs and higher quality arguments about health policy
          issues in open-ended essays. These findings were replicated in Study 2, which was conducted within a simulated
          mayoral election campaign. The results suggest that political sophisticates do not simply consume more news but
          process it differently than their less-sophisticated counterparts.

          Political knowledge is for the most part obtained vicariously. National political campaigns and policy debates are
          conducted through the media with the public's direct experience of the events rather limited (but see Kinder &
          Sears, 1981; Stoker, 1992). The news media are the primary carriers of political information to the polity,
          sometimes in less-filtered forms such as live telecast on C-SPAN, but more likely in highly interpreted forms such
          as talk radio shows or a New York Times cover story. An informed public is one capable of casting its votes
          based on knowledge of candidates' positions and performance and registering its opinion on matters of public
          policy from a position of knowledgeable evaluation. The processes by which the public becomes informed are
          central to our understanding of both the successes and failures of massive information campaigns aimed at raising
          the public's level of knowledge about political affairs. In the research reported here, a measure of prior political
          knowledge called political sophistication is proposed and evaluated as a predictor of learning from news. Previous
          research has shown various versions of prior knowledge to be consistently related to learning in both experimental
          and survey contexts. We pit our measure against other predictors of learning to evaluate its effectiveness, asking in
          the end what distinguishes the way political sophisticates process news.

          Learning From the News Media

          Popular perception suggests that the media should have large effects on learning and attitude change (e.g.,
          Davison, 1983). When topics are high in personal relevance and when prior knowledge on a topic is limited,
          learning from news should be substantial. The health care reform debate of 1993-1994 is just such a policy issue.
          But there are many reasons why mass*e media effects do not materialize even in the face of high relevance and
          little prior knowledge (McGuire, 1986).

          Simple Media Exposure and Learning

          Robinson and Davis (1990) reviewed large-scale, national surveys of public knowledge about candidates and
          public affairs issues completed in the period 1967 to 1990. Their review concluded that the most important
          predictor of accurate recall is education, followed by political interest. Newspaper reading was a less powerful but
          consistent predictor whereas viewing television news had a very small to negligible effect on learning. These
          conclusions are not surprising. Exposure to news may be a necessary condition for learning political information,
          but it is certainly not sufficient.

          Some researchers have argued that simple exposure measures are blunt instruments for understanding the effects
          of media on learning. J. M. McLeod and McDonald (1985) distinguished between general media exposure and
          specific exposure to public affairs content in both broadcast and print media. They also assessed attention to the
          media. In addition to these refined measures of media exposure, the authors also argued that how people use the
          media may have important mediating effects on learning. If surveillance and utility are two of the typical
          gratifications sought from the media, then perhaps learning would be elevated when these needs are activated.

          J. M. McLeod and McDonald found that simple, generic measures of exposure are not effective predictors of
          learning of specific content. Rather, when measures of exposure to specific content were obtained, specific types
          of learning could be predicted. For example, exposure to news about foreign affairs during a presidential campaign
          was more likely to predict knowledge about candidates' positions on foreign affairs than was general exposure to
          news. Also, attention to the media and gratifications sought in processing information were important predictors in
          achieving even modest amounts of explained variance. These conclusions suggest that much more refined
          measures of exposure to news, including at least specific forms of exposure and specific gratifications, are
          necessary to explain learning effects. Other modifications to the measurement of simple exposure have also been
          proposed and tested with some success. Chaffee and Schleuder (1986) suggested the need for general and
          specific measures of attention to various media in addition to exposure measures. They found that media attention
          added significantly to variance explained in their learning scores.

          The claim that learning from television news is minimal has often been challenged. Chaffee, Zhao, and Leshner
          (1994) found that newspaper reading and television news consumption are related to a voter's knowledge of
          issues and to knowledge about the candidate's personal characteristics. The strongest relationship was to
          knowledge about the candidate's character where variance explained was 10.5%. Candidate personal knowledge
          consisted of information such as who disliked broccoli and who was a Rhodes scholar. The authors concluded
          that "this study adds to the documentation of television's emergence as a principal medium of campaign
          communication" (p. 317). Contrary to detractors, television news consumption "stands out particularly in relation
          to knowledge about issue differences between the candidates" (p. 318). Chaffee et al.'s data indicated that
          learning was related to simple exposure and attention measures. On the negative side, the strongest relationships to
          straight exposure were for a 10-item candidate personal knowledge. Television news consumption was related to
          a 10-item measure of candidate differentiation on issues, but all the exposure measures accounted for 1.8% of the
          variance. Both were also related to a 20-item measure of party issue information, although again the variance
          accounted for was only 3.7%.

          The consensus across a range of research is that simple measures of exposure to news on television, in print, and
          on radio do not account for much variance in the public's knowledge about campaigns or public policy
          debatesunter, 1991; Robinson, 1986; Robinson & Davis, 1986; Robinson & Levy, 1986). This consensus cannot
          be interpreted to mean that the public does not learn from news at all. Rather, the what and when of learning from
          news depends on a complex of factors related to education, prior knowledge, attention to issues and to the media
          themselves, audience motivations, and specificity and reliability of measurement. The research presented here
          focuses on a measure of political knowledge that we call political sophistication. We explore the reliability and
          predictive and concurrent validity of the measure and pit it against alternative ways of predicting learning from
          controlled exposure to news.

          Background Knowledge as an Indicator of Schematic Structure

          Media exposure, attention, and gratifications are at best surrogates for the measure that would be a direct measure
          of reception of information (Zaller, 1992). Reception of information assumes that a person has been exposed to
          the information, attended to it, comprehended it, and can use it effectively in making political judgments. Consider
          an analogy. Teachers evaluate students in terms of their ability to comprehend, recall, and use the information
          supplied in lectures and readings. They do not evaluate students' performance solely in terms of attendance,
          attentiveness, and attitude toward the subject matter. Although these factors are undoubtedly correlated with more
          direct measures of performance, the direct measures of mastery of course materials are certainly more valid
          indicators than the indirect measures.

          This is precisely what Zaller has proposed at both the theoretical and measurement levels. Reception is measured
          by asking fairly general questions about the positions of various political actors and groups on different issues. For
          example, one question might ask: "On the issue of abortion some people are primarily pro-choice and some are
          primarily pro-life. Where does Ross Perot stand on this issue?" To correctly answer this question, a person must
          have received information about Perot and his position on the choice-life issue; that is, there had to be exposure,
          attention, comprehension, and recall. What Zaller (1987, 1992) has proposed is that the family of measures of
          domain-general political knowledge will be better predictors of specific forms of political learning and attitude
          change than will measures of media exposure to political information (and its surrogates).

          Other researchers have suggested similar measures, calling them political sophistication (Crone, 1993; Luskin,
          1987), political expertise (Fiske, Lau, & Smith, 1990), public affairs knowledge (D. M. McLeod & Perse, 1994;
          Price & Zaller, 1993), awareness (Zaller, 1987), and political schema (Kuklinski, Luskin, & Bolland, 1991). We
          will use the phrase political sophistication. Public affairs knowledge is often not background knowledge but factual
          knowledge specific to various research topics. Expertise suggests a special class of individuals whose
          sophistication may be very high indeed. Political schema carries other structural baggage that is not directly
          measured by simple knowledge scores.

          Crone (1993) has carried out a careful analysis of the concept of political sophistication as used in political science
          and political psychology. She isolated two approaches: sophistication as "overarching ideological viewpoints" (p.
          1) and as the construction of political opinion, knowledge, and information. The former is fundamentally due to the
          work of Converse (1964). Crone maintained that Converse's view implicitly mixes ideology and knowledge.
          Although he does not make the connection explicitly himself, the operational definitions employed seem to require
          knowledge about the ideological positions of parties, groups, and candidates.

          The second view is based on the cognitive perspective from psychology in which information and its organization
          is central to the definition of those who are sophisticated. In a special issue of the journal Social Cognition,
          Krosnick (1990) brought together a number of researchers actively studying the structure of political knowledge.
          Their conceptual and operational approaches to political sophistication varied, but all focused on a central
          theme--those higher in political sophistication had more complete and more accurate knowledge about politics,
          government, and civics than those who were lower.

          Krosnick (1990) included in his measure of sophistication interest in polities, attention to the mass media, and
          political participation as well as political knowledge. Fiske et al. (1990) included political activity, mass media
          exposure, and "political self-schemata" (related to political interest) in addition to political knowledge. Judd and
          Downing (1990) combined political involvement and knowledge into a single scale, as did McGraw and Pinney
          (1990) with behavior, knowledge, interest, and media use.

          Only Zaller (1990) and Kinder and Sanders (1990) used pure knowledge measures. There are a number of
          reasons for adopting the position of these authors according to Crone. First, knowledge is the only measure that is
          common to all those who have conceptualized this variable. Even Converse's ideological tilt can be seen as
          involving knowledge in a central way. Second, knowledge is conceptually distinct from ideology, because one can
          be ideologically driven and still possess low levels of knowledge in terms of breadth and organization. Measures
          like interest in politics, mass media exposure to political information, and involvement in the political process are
          conceptually distinct. Although they may be related to political sophistication, they should remain conceptually
          distinct. Third, political knowledge is a good predictor of the outcome that it should predict, including learning
          (Price & Zaller, 1993), reaction times to answer political questions (Fiske et al., 1990), and attitude consistency
          (Judd & Downing, 1990; Wyckoff, 1987). Crone concluded her theoretical and empirical assessment of political
          sophistication with the notion that the two worlds of political sophistication--ideology and information--are not
          independent but related. Conceptual and operational practice then should incorporate into political sophistication
          both ideological and informational components.

          Part of the theoretical value of the political sophistication construct is found in its explanatory role in models of
          change in public opinion. Zaller (1987) reformulated the models of McGuire (1968) and Converse (1962) in a
          way that explicitly showcases the role of political sophistication in political belief change. The McGuire-Converse
          model holds that the probability of change is a function of the product of two other probabilities: the probability of
          exposure to a persuasive message and the probability of acceptance of the message given exposure. In turn, the
          probability of exposure is a monotonically increasing function of political sophistication as the more knowledgeable
          audiences are more likely to be exposed than the less knowledgeable audiences. The probability of accepting a
          message is a monotonically decreasing function of political sophistication because more knowledgeable audiences
          are presumably more difficult to persuade (Petty & Cacioppo, 1986). The result is a family of (often) nonlinear
          change curves as a function of political sophistication. Zaller (1992) tested this theory in a wide variety of political
          contexts. He found that a relatively simple measure both operationally and theoretically is the driving engine of a
          theory of political belief formation and change. Important, the measure is a surrogate for the media's effects and a
          surrogate for the audience's structuring of general political knowledge.

          Price and Zaller (1993) specifically tested the relative efficacy of political sophistication versus newspaper and
          television exposure, and interpersonal discussion in predicting people's knowledge of specific news stories
          appearing in the national news agenda. Across 16 news items, political sophistication was a strong, significant
          predictor of learning even in the presence of education, whereas newspaper exposure was not significant and
          television news exposure had small effects, significant in about half the cases. The authors argued that even very
          simple measures of political knowledge are better predictors of learning from the media than are most any
          combination of exposure measures. The authors avoided concluding that television, print news exposure, and
          education have no predictive power, but they did conclude that "their incremental predictive power is relatively
          weak, and preexisting levels of general political knowledge clearly offers us the most reliable and parsimonious
          way of predicting individual differences in likelihood of news reception" (p. 153).

          Although the authors did not imply media exposure measures are useless in studies of learning and opinion change,
          they did invite students of political communication to "abandon their normal reliance on self-reported levels of
          news media exposure and look instead to prior political knowledge [political sophistication] as a preferred general
          indicator of news reception" (p. 160). It is our view that such advice is premature, not because Price and Zaller's
          findings are flawed in any obvious way, but because replication and extension of their analyses are necessary
          before the long and venerable traditions of using media exposure measures are replaced. Also, political
          sophistication is presumed to be related to exposure to political information and to its processing. Exploration of
          this very assumption is the primary purpose of the present research.

          Research Questions and Hypotheses

          Our research was conducted in the service of three goals: (a) to replicate the findings on the relationship between
          political sophistication and learning in other contexts, (b) to explore the conceptual underpinnings of the political
          sophistication construct empirically and theoretically, and (c) to begin explaining how political sophistication comes
          about.

          The Relationship Between Prior Knowledge and Learning

          If political sophistication is to be a good measure of reception, then its relationship to learning should be robust.
          Numerous studies and a variety of researchers have noted that those with higher levels of general knowledge also
          learn more from exposure to information from the media. When education is treated as a surrogate measure of
          prior knowledge, it predicts learning in a variety of contexts, often more strongly than any other factor (Chaffee et
          al., 1994; Robinson & Davis, 1990).

          Education, when coupled with levels of political interest, may be an indicator of prior knowledge as Robinson
          (1986) suggested, but it is not a direct measure of prior knowledge. When prior knowledge is measured directly,
          subsequent learning is predicted strongly. Robinson and Davis (1986) showed in two studies of learning in the
          United States and England that those with high levels of prior knowledge comprehend and recall more from a
          single evening's newscast than those with low prior levels of knowledge. Learning from news is "not a simple
          by-product of the educational process but is influenced by some personal investment of interest and ability as well"
          (p. 129).

          Neuman, Just, and Crigler (1992) conducted a series of experiments on learning from various media. Part of their
          concern was with the audience's level of ability to process information. They wanted to differentiate participants'
          ability level but not confound general cognitive abilities with simple recall. Rather, they tried to keep cognitive
          ability distinct from recall by using vocabulary and inference skills as their measure of ability. In one experiment,
          participants were presented with a story from network TV news, a news magazine, or a newspaper. They found
          that knowledge about news stories that the audience brought into the experiment was correlated with cognitive
          skill. And, after exposure to the news stories, the percentage correct on tests of knowledge was linearly related to
          education, cognitive skill, and previous levels of attention to the story.

          Those with low cognitive skill and low attention to issues were lowest in knowledge prior to exposure to news and
          had lowest recall scores after exposure to news; those highest in each of these factors were highest in preexposure
          levels of knowledge and highest in postexposure recall. Those high in verbal and cognitive skills not only have
          substantial levels of knowledge to carry with them as they approach news but also retain more either as a result of
          previous knowledge, skill at retention and retrieval, or both. The authors argued that those who are highly skilled
          are simply more efficient than the less skilled. They can scan much more material than those who are low in skill
          and retrieve what is useful and relevant.

          Price and Zaller (1993) used a measure of factual political knowledge to predict learning from news. Factual
          political knowledge was based on a person's ability to identify jobs held by various figures (such as George
          Schultz, Margaret Thatcher, and Ted Kennedy) and questions about which party is more conservative, in power
          in the House and Senate, and so on. The authors argued that even very simple measures of political knowledge
          are better predictors of learning from the media than are most any combination of exposure measures. They did
          not argue that political knowledge is expertise but only that previous knowledge is important for subsequent
          knowledge perhaps because of its relation to exposure and perhaps because of its relation to reception. Price and
          Zaller (1993) found that their measure of political knowledge predicted learning from news better than any
          measure of exposure to print or broadcast news.

          In sum, both experimental and survey evidence indicated that previous knowledge of topics facilitates recall of
          information from the news even when the previous knowledge is not topically related to the news items. Also,
          prior knowledge predicts learning of news better than does simple exposure to print or broadcast news. What is
          not clear is why political sophistication should function so effectively in predicting reception of information from
          news.

          Explanations of the Sophistication-Learning Relationship

          One answer to this question has been that political sophistication is a surrogate for receptivity and is therefore an
          indicator of successful comprehension and recall of political information. Actually, the measure of sophistication is
          itself a measure of accurate recall of factual information, but it is assumed to represent intervening subprocesses
          such as attention and comprehension. This explanation is unsatisfying theoretically because political sophistication
          may simply be a more abstract indicator of learning and related to specific learning in much the same way that
          general intelligence is related to more specific forms of intelligence.

          Some researchers have argued that political sophistication is an indicator of the degree of schema development
          (Fiske et al., 1990; Hsu & Price, 1993) and all that such development entails (Markus & Zajonc, 1985). If it is,
          then those high in political sophistication should have more developed political schemes with positive
          consequences for political learning (Fiske et al., 1990) such as a greater number of thoughts in response to
          persuasive messages (Hsu & Price, 1993), faster response times to political questions, and more accurate
          inferences from political messages.

          However, the measure of political sophistication is only a measure of general factual knowledge and not anything
          like the organized structures of knowledge and procedure presumed to be the core of the concept of schema. As
          Kuklinski et al. (1991) have pointed out, the schema concept is much broader and deeper than its measurement
          has been. If we are to even entertain the idea that political sophistication measured through general questions
          about political issues is an indicator of the level of development of political schemes, then additional evidence of
          schematicity is necessary. One goal of the research reported here is to ask whether the construct of political
          sophistication is an indicator of the level of development of political schemes.

          There are many ways to index a cognitive schema, but, as a practical matter, none is definitive. Schemas are
          organized knowledge structures in which the content can be social, personal, role, self, analogic, temporal,
          technical, political, and so on (Fiske & Taylor, 1991) and the organization hierarchical, associational, or a
          combination. Schemas are basically unobserved constructs that, if they are to be accepted as viable accounts of
          human cognition and behavior, have to be useful shortcuts for explaining observed outcomes. As with any
          unobserved construct, researchers use schemes conceptually to make sense of observed outcomes while trying to
          gather increasingly direct evidence of the existence of the schema itself. In our studies, two indicators of schematic
          organization are employed: construct differentiation in open-ended narratives and argumentative depth in
          open-ended questions about social issues.

          Response times can be used to assess schematic organization (Luce, 1986). To do so requires knowing the
          schematic structure of knowledge in advance or establishing it experimentally. In either case, one must have an
          operational definition of schema in advance of employing the response time measure that is independent of the
          response time measure itself. In short, measuring schemes is a boot-strapping process in which softer measures
          such as cognitive differentiation and integration must serve as direct measures in some studies and validators of
          harder measures in other studies. Our research design did not allow the use of response time measure of
          schematic operation.

          Construct differentiation has been widely used as a measure of the complexity of a person's knowledge about a
          particular topic. In its various guises, construct differentiation has been employed in studies of interpersonal
          communication (Burleson & Waltman, 1988), information processing (Schroder, Driver, & Streufert, 1967), and
          political reasoning (Tetlock, 1985). If political sophistication is an indicator of the level of schematic development,
          then those higher in sophistication should also have higher differentiation scores. Presumably, the greater the
          construct differentiation, the more different ways a person has to categorize and differentiate issues on a topic.
          Construct differentiation indexes the dimensionality of a person's schema and is, therefore, a necessary condition
          for showing that political sophistication is an indicator of schema development. But differentiation shows neither
          the organization of knowledge in the schema nor the person's ability to use the knowledge in that most important
          of political activities--reasoned argument. If political sophistication is an indicator of schema development, then
          those higher in sophistication should exhibit more highly developed reasoning in their political arguments.

          Our measure of political argument depth is derived from the work of Kuhn (1991). Kuhn studied arguments that
          people made in interviews about such topics as recidivism and failure in school. She coded the number of different
          reasons that people cited for a particular claim, their use of evidence, their ability to conjure relevant
          counterarguments to their stated positions (e.g., from someone who might disagree), and their ability to refute the
          hypothetical counters. People who show mastery of all of these components of an argumentative exchange might
          be said to have a well-organized knowledge structure (on the topic) and to be capable of using that knowledge in
          dialogic arguments. Based on Kuhn's research and theorizing about argument, a measure of argumentative depth
          was created (Woodard, 1995). We assumed that those high in political sophistication would also show more
          argumentative depth and that this depth is indicative of more integrated schemes and greater ability to use the
          knowledge stored there.

          When media exposure on a topic is controlled, does political sophistication still explain learning on the topic over
          and above the learning due to exposure? It is assumed that political sophistication is a measure of receptivity and
          that receptivity is exposure plus attention plus comprehension plus recall. So when exposure is controlled, and if
          political sophistication still results in learning, then the results must be attributed to attention, comprehension, and
          recall and not to simple exposure. This would mean that those who differ in political sophistication differ not just in
          exposure but how, cognitively, they are processing the information to which they are exposed (Graber, 1988;
          Woodall, 1986).

          Antecedents of Political Sophistication

          The final research question raised in this article concerns the antecedents of political sophistication and of learning.
          Although political sophistication may be a predictor of learning, it is almost certainly not independent of media
          exposure to political news. On the contrary, news exposure is a necessary condition for learning and activates the
          knowledge structures of sophisticates. What is key is a precise understanding of the relationship between
          measures of exposure and measures of political sophistication. Because those high in political sophistication are
          purported to have high exposure and to process the information to which they are exposed more completely
          (through attention, comprehension, and recall), we hypothesize that two broad classes of predictors must be
          considered when accounting for sophistication scores: exposure levels (obviously) and the uses to which the
          information is likely to be put.

          The extent to which people pay attention to political events and issues and the way they use the news should differ
          for those who differ in political sophistication. Specifically, we expect the effects of media news exposure on
          political sophistication to be mediated by the extent to which people pay attention to news and how politically
          involved they are. As suggested by J. M. McLeod and McDonald (1985) and Chaffee and Schleuder (1986), the
          level of attention represents an important way that media news can matter to people who are more and less
          politically sophisticated.

          In sum, four questions are raised in the research reported here: Does political sophistication, measured as
          knowledge and ideology, predict learning from controlled sources of news? Does political sophistication predict
          learning better than, or worse than, other predictors including education, media exposure, and other measures of
          knowledge? Do people high in political sophistication exhibit more developed cognitive schemes as evidenced in
          more elaborated, differentiated, and deeper reasoning about issues? Is more elevated political sophistication the
          result of demographic, attentional, media use, and involvement differences?

          Study 1

          From late summer 1993 to early fall 1994, the Clinton administration pushed vigorously for change in the health
          delivery system in the United States. Media attention to the need for and prospects of health care reform was
          intense. In the period from January 15, 1994, to July 25 of the same year, we catalogued 2,066 newspaper items
          about health care reform from 10 national and local newspapers and 678 items from 9 sources of broadcast news.
          During the same period, national public opinion polls showed that the percentage of persons who knew what a
          "health alliance or consumer purchasing cooperative" was changed from 22% in early October 1993 to 36% in
          late June 1994.(2) Despite massive exposure, the public's knowledge and confidence in their knowledge about the
          issues surrounding health care remained at low levels.

          Method

          Data to evaluate our hypotheses came from a field experiment on the effects of print news formats on learning and
          public cynicism during the health care reform debate. The experiment was conducted during March 1994 and
          some of its results are discussed in Cappella and Jamieson (1994).

          SUBJECTS

          Approximately 350 people from six media markets (New York; Philadelphia; Washington, D.C.; Dallas; Los
          Angeles; and Chicago) were recruited to participate in the field experiment through notices posted in fraternal,
          social, and religious groups, work, and other settings. The sample was not random and participants were recruited
          for pay. Space limitations prohibit a detailed discussion of the demographic characteristics of the sample. Our
          participants were more highly educated (58% college degree or higher), more female (60%), and more
          Democratic (47%) than national samples. With regard to age and race, they were roughly comparable to national
          norms.

          STIMULI

          The types of news coverage tested were chosen to reflect the kinds of coverage that journalists had been giving to
          health care reform from September 1993 through March 1994. These are discussed in detail elsewhere (Cappella
          & Jamieson, 1994). Suffice it to say that four unique types of media coverage were used. Six combinations of
          print news stories were given to the six experimental groups.

          DESIGN AND PROCEDURES

          A research assistant was on-site for a full week in each of the media markets. He or she administered a pretest
          questionnaire to each of the participants before any news articles were read. Everyone received 15 news articles,
          3 per day, to read. One factual article on the basic issues in health care reform was common to all groups
          including the control. The control received 14 other news articles on current affairs but not on the health care
          debate.

          The design was a 1 x 7 factorial with posttest-only measures for all variables except those for attitudes that were
          obtained both in the pretests and posttests. Random assignment to condition was successful for various
          demographic variables, as well as media consumption, attitude, political sophistication, and cynicism variables.
          Participants reported following instructions about reading news articles (or were dropped from analysis). They
          also found the news articles realistic, like the ones they normally find in newspapers, but they reported reading
          them somewhat more closely than they normally would have. At the end of the week, all participants met together
          to watch an edited, 20-minute debate on health care that had previously appeared on C-SPAN. They filled out
          the final questionnaire on attitudes, learning, cynicism, and other items, and were debriefed.

          Measures

          POLITICAL SOPHISTICATION

          Political sophistication was measured with a series of questions on general political knowledge. Specifically,
          participants were asked to place Bill Clinton, most conservatives, and most liberals on a continuum that most
          closely described their positions on various issues. For example,

          Some people say that abortion should not be permitted because it is

          taking the life of an unborn child. Others say that abortion should be

          permitted in the first three months of pregnancy as a matter of personal

          choice.

          What does Bill Clinton say about life and choice issues?

          Pro-life:--:--:--:--:--:--:--:Pro-choice

          Other issues included preference in hiring for minorities, NAFTA and jobs, environment and job loss, health costs
          and price controls, and health reform and government involvement.

          Scoring of the answers was done by comparing a person's positioning of Clinton, most liberals, and most
          conservatives relative to one another. For example, if a person indicated that the position of most liberals on
          NAFTA was more toward the "lose jobs" end of the continuum whereas that of most conservatives was toward
          the "gain jobs" end, they would be given 1 point correct but not if the order was reversed. Thus, the positioning of
          Clinton relative to conservatives, Clinton to liberals, and liberals to conservatives was compared for correct
          positioning relative to one another. Being on the "correct" side of the scale was not evaluated; only the relative
          positions were evaluated. In those cases in which the relative positions did not yield a clear answer in terms of
          acknowledged public positions, the comparison was ignored. For example, Clinton is not clearly more liberal nor
          clearly more conservative than most liberals on the issue of pro-choice versus pro-life. So this comparison does
          not enter the index of political sophistication.

          This measure provides a number of advantages as an indicator of political sophistication. First, it allows people to
          use the scales in their own way and only makes comparisons within persons rather than evaluating the person's
          score to some external standard. Second, the measure allows both knowledge and ideology to be factored
          simultaneously. As Crone (1993) has argued, political sophistication should reflect both ideological knowledge
          and knowledge not readily derived from ideological heuristics. By comparing ideologically affiliated groups to one
          another on a continuum defined by an issue (and not an ideology), the measure has features of ideological and
          nonideological, substantive knowledge as components. Finally, the measure does not sound like a test of
          knowledge, so respondents may be less anxious about it than recall or recognition tests.

          The scale consisted of 11 comparisons with a range of 0 to 11 (M = 8.50, SD = 2.4), and an internal reliability of
          .79.(3)

          CIVICS KNOWLEDGE

          In contrast to political sophistication, which tries to measure current political knowledge, civics knowledge
          assesses background information about the functioning of government. Five questions were used to measure civics
          knowledge, including the size of the majority necessary to override a presidential veto, the number of full terms of
          office that can be served by the president, the term of office of a United States senator, the party currently in
          power in the House, and the names of the two senators from the person's state. The range on this variable was 0
          to 1.25 (M = 1.06, SD = .26). Cronbach's alpha for this measure was .51.

          POLITICAL INVOLVEMENT

          Political involvement was assessed by a 7-item yes or no scale asking people if they had ever written a letter to a
          newspaper, joined a political organization, written a letter to a politician, signed a petition on a political issue,
          participated in a march or demonstration, voted in the most recent presidential election, and voted in the most
          recent senate election. This measure has a range of 0 to 7 (M = 3.8, SD = 1.9) and an internal reliability of .72.

          ATTENTION TO NATIONAL ISSUES

          Attention to national issues was measured by four questions asking how closely (scored 1 to 4) the person had
          been following the Whitewater allegations, NAFTA, health care reform, and discussions about the national debt.
          The range on this variable was 1 to 4 (M = 2.5, SD = .69). This measure was found to be reliable with
          Cronbach's alpha of .80.

          MEDIA NEWS EXPOSURE

          Nine questions were asked about exposure to media for news. People were queried about their use of national
          TV news programs, local TV news, TV interview news, newspapers, news magazines (like Newsweek), radio
          news, radio talk and call-in, TV talk shows, and conversations with friends and coworkers. All questions were of
          the form "How often do you listen to radio call-in and radio talk shows for your news?" with weekly or daily
          measures used as appropriate. Four indexes were generated through factor analysis: TV news (3 items with an
          internal reliability of .52), newspapers (2 items with a correlation of .31), radio news (2 items with a correlation of
          .47), and a single item for conversation with friends.(4)

          LEARNING

          At the final meeting of the participants in each research site, people answered questions to test their recall of
          information from the C-SPAN debate and from a factual background newspaper article about health care.
          Everyone in the study had exposure to both sources. Three types of questions were used to measure recall: factual
          recall from the print article, recall of strategic information from the videotaped debate, and issue-based recall from
          the debate.

          One cluster of items consisted of seven issue-oriented questions from the C-SPAN debate. For example, these
          included questions such as:

          Who argued that providing Americans with free health services would

          not work because people would not "show up" to get these services?

          Who complained that health care costs only slow down when there is

          a threat of government price controls?

          Although about 27% of the participants got fewer than half the issue-based answers correct, there was a sharp
          skew toward correct answers with one quarter of the participants getting more than 85% of the answers correct.

          A second group from the C-SPAN debate consisted of five strategy-oriented questions. Strategy-based questions
          included:

          Who leaned back in his chair and looked overconfident?

          Who referred to the other people on the panel by their first names?

          Sixty-one percent of the participants got 40% to 60% of the strategy answers correct. The distribution of correct
          answers was approximately normal.

          The third cluster from the newspaper article consisted of five questions. Questions testing knowledge gleaned from
          the factual newspaper article included:

          Do the health care plans being debated now agree on anything?

          Which of the plans set government limits on national health spending?

          Each of the questions had an appropriate set of alternatives from which to choose. Sixty percent got 40% to 60%
          of the answers correct. The distribution of correct answers was approximately normal.

          It should be emphasized that the correct answers from the C-SPAN debate required watching and attending to
          the debate. They could not be answered solely from one's previous exposure to information about health care
          reform (although it was our hypothesis that exposure to certain kinds of coverage would enhance or depress
          people's ability to handle strategic and issue-based information). Correct answers based on questions from the
          newspaper article could be enhanced or depressed as a result of reception of outside information about the health
          care reform debate.

          ARGUMENTATIVE DEPTH

          To assess argumentative depth, participants were asked the following series of open-ended questions:

          In the debate about health care, almost everyone agrees that between

          35 and 45 million Americans have no health coverage.

          Why do you think these people do not have health care coverage?

          What other reasons come to mind as to why people don't have health

          care coverage?

          If someone were to disagree with your reasons about why people don't

          have coverage what might that person say to try to convince you that

          your reasons were wrong?

          What reasons could you give to convince your friend that you were right

          in the first place?

          Based on Kuhn's (1991) work with naturally occurring arguments, a hierarchical coding system was developed
          that reflected whether (relevant) reasons were provided at all; the number of (relevant) reasons provided; the
          number of claims with coherent reasoning; whether a relevant counter was provided; or whether a relevant rebuttal
          to the counter was provided.(5) A single score representing argumentative depth summed the five components.
          The subcomponents of argumentative depth together produce an internal reliability of .72 for the entire scale,
          suggesting they are each measuring a component of the same process. Reliability of coding was assessed at each
          of the five levels of the coding system and overall. The range across levels was .56 to 1.00 on Krippendorff's
          alpha (Krippendoff, 1980) and 67% percent to 100% on agreement.

          CONSTRUCT DIFFERENTIATION

          To assess the level of complexity of people's knowledge about the health care debate, participants were asked to
          answer the following open-ended question at the conclusion of the field experiment:

          Pretend that a good friend writes you a letter asking what you know

          about the health care debate in the United States. Your friend has been

          out of the country and knows nothing except that a health care policy

          is currently being debated. Write a letter to your friend telling him or

          her as much as you can about the alternative health proposals, the

          persons sponsoring them, and their likely effect on you if they are

          passed. In responding to your friend draw on everything you know

          about the debate including information gotten from other sources and

          from what you read and watched in the past week as a part of this study.

          To evaluate the complexity of these essays, a coding system was developed based on the methodologies of
          Tetlock (fetlock & Hannuum, 1984) and of Burleson (Burleson & Waltman, 1988) that evaluated the essays for
          the number of unique constructs related to the health care debate.(6) Fifty-seven of 340 essays were coded for
          construct differentiation. The reliability of the coding was high, being in the range of .90. However, the correlation
          between construct differentiation and simple word counts was so high (r = .95) that coding was abandoned in
          favor of simple word counts. The two measures share 90% of their variance. Word counts ranged from 0 to 686
          (M = 161, SD = 109).

          Results

          CORRELATIONAL ANALYSIS

          Zero-order correlations among political sophistication, civics knowledge, involvement, attention, media variables,
          and demographics are reported at the bottom diagonal of Table 1. Political sophistication was modestly and
          significantly associated with civics knowledge, political involvement, and attention to national issues. In relation to
          media news exposure variables, political sophistication showed a weak positive association with newspaper
          exposure (r = .11, p < .05), whereas it was negatively linked with television news exposure (r = -.21,p < .001).
          Civics knowledge and political involvement were significantly associated with newspaper exposure but not with
          television exposure. Attention to national issues was closely related to all the media exposure variables. In addition
          to the modest relationship with education, political sophistication is significantly associated with other political
          orientation variables. This suggests that, for the most part, the political sophistication construct is empirically sound
          in the sense of concurrent validity

          Table 1
          Pearson Correlations among Variables (Studies 1 and 2)

                                             Political           Civics
                                           Sophistication       Knowledge
          Variable                               1                  2

          Political sophistication              --               .37(***)
          Civics knowledge                      .30(***)         --
          Involvement                           .40(***)         .40(***)
          Attention                             .26(***)         .31(***)
          Print                                 .11(*)           .30(***)
          TV                                   -.21(***)         .03
          Radio                                -.08              .08
          Talk                                 -.02              .02
          Education                             .35(***)         .32(***)
          Age                                   .01              .18(**)
          Female                               -.03             -.12(*)

                                      Involvement    Attention     Print
          Variable                         3             4           5

          Political sophistication     .25(***)       .24(***)      .12
          Civics knowledge             .39(***)       .31(***)      .27(***)
          Involvement                  --             .32(***)      .15(*)
          Attention                    .42(***)       --            .33(***)
          Print                        .31(***)       .36(***)      --
          TV                          -.01            .29(***)      .21(***)
          Radio                        .01            .21(***)      .06
          Talk                         .09            .25(***)      .25(***)
          Education                    .42(***)       .16(**)       .16(**)
          Age                          .18(**)        .25(***)      .25(***)
          Female                      -.06           -.16(**)      -.16(**)

                                        TV           Radio         Talk
          Variable                       6             7            8

          Political sophistication     -.02          .13(*)         --
          Civics knowledge              .13(*)       .17(*)         --
          Involvement                   .09          .05            --
          Attention                     .42(***)     .10            --
          Print                         .33(***)     .18(**)        --
          TV                            --           .08            --
          Radio                         .22(***)     --             --
          Talk                          .17(**)      .14(*)         --
          Education                     .22(***)    -.06            .06
          Age                           .17(**)      .13(*)        -.15(**)
          Female                       -.23(***)    -.07           -.03

                                      Education      Age        Female
          Variable                       9           10           11

          Political sophistication      .29(***)    -.01           -.03
          Civics knowledge              .22(***)     .24(***)      -.07
          Involvement                   .31(***)     .28(***)       .13
          Attention                     .14(*)       .19(**)       -.05
          Print                         .10          .26(***)      -.04
          TV                           -.11          .40(***)       .01
          Radio                        -.01          .05           -.07
          Talk                          --           --             --
          Education                     --           .01            .08
          Age                           .11(*)       --             .02
          Female                       -.09          .02            --

          Note. The bottom diagonal is for Study 1 (N = 325). The upper diagonal is for Study 2 (N = 243). (*) p <.05.
          (**) p <.01. (***) p <.001.

          PREDICTING OUTCOME MEASURES

          WITH POLITICAL SOPHISTICATION

          In an effort to examine the predictive power of political sophistication, three conceptually distinct outcome
          measures were assessed: (a) learning measured by three different types of recall items (factual, issue, and strategy
          recall); (b) use of knowledge in argumentation measured by argumentative quality; and (c) construct differentiation
          measured by the number of words in the essay question. Multiple regressions were run to evaluate the relative
          importance of political sophistication in predicting these outcome measures as compared to other predictors,
          including media exposure variables, attention, involvement, and civics knowledge. Education and experimental
          condition (1 = experimental condition, 0 = control one) were also entered in the regression equation to control for
          the effects of added exposure.

          The first three regression equations in Table 2 indicate that political sophistication is a significant predictor of each
          of the three types of learning (two from broadcast and one from print). Even after controlling for the effects of
          education, experimental condition, and news exposure, people with greater political sophistication were more
          accurate in their recall of information. The effect size for factual recall questions about the newspaper articles (p =
          .24, p < .001) was about the same with that of issue questions about the videotaped debate (p = .26, p < .001).
          But it was relatively small for strategy-type questions about the videotaped debate (p = .17, p < .01). Except for
          the negative effect of television exposure on issue recall (and the borderline effect of radio exposure on strategy
          recall), media exposure variables did not approach significance in predicting learning measures. The results suggest
          that the presence of an effect of political sophistication on learning must be due to differences in the way the high
          sophisticates process information in comparison to the lows.

          Table 2
          Predicting Recall, Argument Quality, and
          Construct Differentiation (Study 1)

                                                  Outcome Measures

                                      Factual          Issue        Strategy
          Predictor                    Recall          Recall        Recall

          Education                     .14(*)           .06           .02
          Experimental condition        .05              .01          -.00
          Political sophistication      .24(***)         .26(***)      .17(**)
          Civics knowledge             -.01              .08           .02
          Involvement                  -.01              .13(*)        .00
          Attention                     .12(+)           .06           .02
          Print                        -.02             -.07          -.04
          Television                   -.03             -.15(**)      -.04
          Radio                        -.04              .07           .10(+)
          Talk                          .01              .03           .07
          [R.sup.2]                     .15              .20           .05
          F                            5.86(***)        7.88(***)     1.82(+)
          N                          332              332               332

                                       Argument         Construct
          Predictor                     Quality        Differentiation

          Education                      .10(*)           .24(***)
          Experimental condition         .06              .00
          Political sophistication       .19(**)          .24(***)
          Civics knowledge               .12(+)           .02
          Involvement                   -.06              .10
          Attention                      .10              .04
          Print                          .00             -.09
          Television                    -.03             -.14(*)
          Radio                         -.04             -.06
          Talk                           .08              .05
          [R.sup.2]                      .14              .24
          F                             5.28(**)        10.31(***)
          N                           328              332

          Note. Cell entries are beta weights from ordinary least squares regression.

          (+) p < 10.

          (*) p < .05.

          (**) p < .01.

          (***) p < .001.

          Columns 4 and 5 in Table 2 present the results on the two indicators of schematicity: argumentative depth and
          construct differentiation. In both cases, political sophistication was strongly related to these indicators of
          schematicity over and above the effects due simply to education. People high in political sophistication wrote
          essays about the health care reform debate that had more constructs than those who were low ([Beta] = .19,p <
          .01). Similarly, sophisticates crafted arguments about one of the fundamental problems in the health care system
          that were of greater depth than those less sophisticated ([Beta] = .24, p < .001). Together, these two findings
          suggest that sophisticates may have more developed schemes reflected in the complexity of constructs employed,
          their integration, and ready availability for use in crafting arguments.

          In all five regression analyses, the tolerance indexes for all the predictors were above .65, which suggested that
          there was no serious problem of multicollinearity. Therefore, the beta weights for the predictors can be interpreted
          as the relative importance in predicting the outcome measures. The beta weights for political sophistication ranged
          from .17 to .26. They were larger than those of all the other predictors. This indicates that political sophistication is
          a better predictor of learning, argument quality, and cognitive differentiation than civics knowledge, involvement,
          attention, and news exposure variables.

          STRUCTURE OF POLITICAL SOPHISTICATION

          Political sophistication shows good predictive validity in terms of learning and schematicity. In this section, we ask
          what accounts for variation in political sophistication. If it is a measure of reception, then it should be related to
          exposure measures and to other measures of how the news media are used. Table 3 shows the relative amount of
          explained variance in sophistication by three classes of predictors: demographics, political orientations (knowledge
          of government, involvement, attention to current issues), and media exposure variables. Four regressions are
          presented: demographic predictors alone, political orientations alone, news exposure measures alone, and all three
          groups at once.

          Table 3
          Predicting Political Sophistication (Study 1)

                                              Equations

          Predictor               1              2            3

          Education              .36***
          Age                   -.02
          Female                -.03
          Civics knowledge                    .14(*)
          Involvement                         .32(***)
          Attention                           .07
          Print                                             .13(*)
          Television                                       -.22(***)
          Radio                                            -.03
          Talk                                              .06
          Adj. [R.sup.2]         .12          .18           .04
          F                    15.7(***)    24.9(***)      4.8***
          N                   327          327           329

                               Equations

          Predictor               4

          Education               .19(***)
          Age                    -.07
          Female                  .05
          Civics knowledge        .11(*)
          Involvement             .22(***)
          Attention               .23(***)
          Print                   -.03
          Television              -.22(***)
          Radio                   -.03
          Talk                    -.07
          Adj. [R.sup.2]           .27
          F                      12.7(***)
          N                     325

          Note. Cell entries are beta weights from ordinary least squares regression.

          (*) p < .05.

          (**) p < .001.

          In Equation 1, demographics accounted for about 12% of the total variance in political sophistication with one
          significant predictor--education ([Beta] = .36, p < .001). Originally, party affiliation was included in Equation 1.
          However, partisanship did not predict political sophistication at all. Equation 2 reveals that civics knowledge,
          involvement, and attention explained 18% of the variance in political sophistication. Among these, civics
          knowledge ([Beta] = .14, p < .05) and involvement ([Beta] = .32, p < .001) were significant predictors. As
          shown in Equation 3, a mere 4.4% of the variance in sophistication was accounted for by the media news
          exposure variables. Newspaper exposure was significant with a beta of .13 (p < .05), and television exposure was
          negatively related to political sophistication with a beta of -.22 (p < .001). Equation 4 indicates that all the
          explanatory variables accounted for about 27% of the variance in political sophistication. No evident problem of
          multicollinearity was found in Equation 4. That is, the level of tolerance for the predictors was at least above .63
          and the variance inflation factors (VIF) were all below 2.0. This allows us to interpret the beta weights in Equation
          4 as the relative importance in predicting political sophistication. Education ([Beta] = .19, p < .001), political
          involvement ([Beta] = .22,p < .001), attention to national issues ([Beta] = .23, p < .001), and (lack of) exposure
          to television ([Beta] = -.22, p < .001) were the most important predictors. That is, the politically sophisticated are
          more likely to be educated, be engaged in political activities, and closely follow national issues, although less likely
          to watch television news.

          To summarize, the low explained variance due to media exposure, coupled with the substantial effect sizes due to
          political orientations, supports the claim that political sophistication is less the result of exposure to news and more
          the result of how news is processed. Those with higher attention to national issues and greater political involvement
          are more likely to use the news media productively than are those comparably lower. The political news has
          greater utility for those following national issues or those involved in affecting those issues.

          Study 2

          Although the results of Study 1 speak to a number of important issues concerning the predictive validity of political
          sophistication within the context of the 1994 health care reform debate, one might raise a question as to whether
          the findings in Study 1 will hold in other political contexts. Study 2 aimed to replicate the findings of the previous
          study using the same research variables within a different political context--an election campaign. In particular, to
          keep consistency in the framework of hypothesis testing, key research variables--such as political sophistication,
          media exposure, and civics knowledge--were constructed in much the same way as Study 1. One exception was
          the measure of attention. In Study 1, attention was measured by how closely people followed major national
          issues; this time, in Study 2, attention was specifically measured by the extent to which people paid attention to the
          1992 presidential election campaign and to the news coverage of it. In addition, to retest the validity of political
          sophistication as a measure of schematicity, a new outcome measure, called the issue-elaboration index, was
          employed. This measure concerned the extent to which people attributed campaign issues to political agents (i.e.,
          major candidates) and particular claims made by them. Because this measure presumably captures the cognitive
          ability to sort out different campaign issues and to make connections among social agents, sources, and claims, we
          believe that it taps into advanced uses of political knowledge.

          Method

          The data came from a research project on news in a simulated political campaign (Jamieson & Cappella, 1993)
          examining the effects of news reporting frames on learning, cynicism, and voting intention. This research project
          was designed as a field experiment with a sample of 276 adults recruited in seven states. The sample
          characteristics and the nature of study design were similar to those of Study 1. The sample was representative in
          terms of gender, age, and race as compared to the estimates of U.S. voter population in the National Election
          Survey. The sample overrepresented the most highly educated. The experimental materials were news stories
          about the 1991 Philadelphia mayoral election campaign. In practice, two different formats of news
          coverage--issue versus strategy-oriented reporting--were created for print and broadcast news. The results of the
          stimulus-evaluation tests and the manipulation check were reported in the experimenters' report (Cappella &
          Jamieson, 1994). It was found that the news stories were generally consistent with what the experimenters had
          assumed about the articles. Before the experimental treatment, participants answered the questions about
          demographics, political orientations, and political sophistication. Participants were randomly assigned to the
          experimental setting and were asked to read or watch the news materials. After 5 days of exposure to the
          materials, the items measuring learning and schematicity were administrated as part of posttests.

          Measures

          POLITICAL SOPHISTICATION

          In constructing the measure of political sophistication, the same method used in Study 1 was employed. Five issue
          items (hiring for minorities, defense spending, abortion, environment, and Medicare) about three political agents'
          ideological stances (Bill Clinton, most conservative, and most liberals) rendered 11 discernible comparisons. This
          scale ranged from O to 11 (M = 8.26, SD = 2.89) and had an internal reliability of .84.

          CIVICS KNOWLEDGE

          The same 5-item scale used in Study 1 was used. The number of correct answers to the five questions was
          counted. Because one of the questions had two correct answers, this measure ranged from O to 6 (M = 4.60, SD
          = 1.46). An internal reliability of this measure was .57.

          POLITICAL INVOLVEMENT

          Among the seven items used in Study 1, six of them were employed here except for the one about participation in
          a march or demonstration. This measure ranged from 0 to 6 (M = 3.15, SD = 1.53) with an internal reliability of
          .64.

          ATTENTION TO ELECTION CAMPAIGN

          Media attention was measured by summing three 5-point scales measuring the extent of attention that participants
          paid to the 1992 presidential election campaign during the last 2 months of the campaign. The measure was
          constructed based on the average value of three items, including the extent of paying attention to newspapers, the
          extent of paying attention to television programs, and the extent of paying attention to the political campaigns. This
          measure had a reliability coefficient of .81 with a mean of 3.55 (SD = .90).

          MEDIA NEWS EXPOSURE

          Television news exposure consisted of three items about network news, local news, and television interviews.
          Because the items were scaled slightly differently, they were standardized and summed. This measure was found
          to be internally reliable with Cronbach's alpha of .78. Print news exposure was measured by two standardized
          interval scales about newspaper and news magazine consumption (r = .38). Radio exposure consisted of two
          items about radio news and radio talk show (r = .59).

          LEARNING

          Two sets of recall items were employed to measure recall: issue recall and strategy recall. Participants (except for
          those in the control condition) received either issue or strategy news stories about the simulated mayoral election
          campaign. Fifteen issue recall items and eight strategy recall items were asked. For example, one issue recall
          question was "Which candidate for mayor of Philadelphia favored privatization of park maintenance and trash
          collection?" One strategy recall was "Which candidate had the most to lose in the debates?" The number of
          correct answers to each type of question was counted and divided by the number of valid answers. Issue recall
          had a mean of .57 (SD = .21) whereas strategy recall had a mean of .69 (SD = .22).

          ISSUE ELABORATION INDEX

          An index of issue elaboration was based on answers to an open-ended question. During the posttest, participants
          were asked to write "the most important things that come to mind" from the news materials that they had watched
          and read about the Philadelphia mayoral candidates, the issues, or the campaign. The answers to this question
          were coded by two primary coders following a set of predetermined rules. The coding rules were constructed to
          capture a conceptual unit called an event. An event conveys a thought that has information about an agent, an
      &