Content-Type: text/html

Risk Perception in Community Context.
 
Evaluating the psychometric paradigm and its relationship
to risk amplification and reported communication channel usefulness.
 
 
 
 
 
A paper presented to the Science Interest Group at the
Annual Meeting of the Association for Education in Journalism and Mass
 
            Communication
Washington, DC
August 1995
 
 
by Craig W. Trumbo
 
 
Graduate Student
Department of Agricultural Journalism
440 Henry Mall
The University of Wisconsin
Madison, WI 53706
 
608-262-4902 (office)
608-233-9189 (home)
 
<a href="/cgi-bin/wa?LOGON=A3%3Dind9602A%26B%3D--%26E%3D0%26L%3DAEJMC%26P%3D4697548%26T%3Dtext%252Fplain" >[log in to unmask]</a>
 
 
 
 
 
 
 
 
 
 ABSTRACT
 
This project incorporates three sequential steps. First, the
 
      psychometric model of risk perception is evaluated for its validity
 
         under field conditions. Second, individuals are classified as
amplifiers
 
            or attenuators within the social amplification of risk model.
Finally,
 
            the characteristics of attenuators and amplifiers are explored, with
 
          special focus on their use of communication channels. Survey data from
 
            an on-going case study is employed in the analysis. The case study
 
        involves a mid-western community in which a controversy exists over the
 
            possibility of the existence of a cancer cluster caused by the
operation
 
            of a small reactor.
Results show that the psychometric model of risk perception has validity
 
                under the field conditions utilized in this study. Use of the
 
   psychometric model to classify individuals as risk amplifiers or risk
 
           attenuators produces a useful dichotomy that reveals differences
between
 
            the two polar groups in terms of demographics, satisfaction with
 
      institutional response to the risk, concern over individual and social
 
            levels of risk, and the evaluation of various communication channels
as
 
            having been useful in coming to a judgment about the risk.
A final model comparing the two groups suggests that, in this case, two
 
                dominant forces are in play against one another: an evaluation
of
 
       personal risk versus satisfaction with the institution managing the
 
         risk. Subordinate to these forces are the demographically based
 
     variables of education and years of residence in the community. This
 
          model also illustrates that aggregate-level observations may not be
 
         fully characteristic of underlying processes of polarization.
 
 
ACKNOWLEDGMENT
 
The author would like to thank the following individuals for their
 
            insightful comments on this project: Garrett O'Keefe, Sharon
Dunwoody,
 
            and Jack McLeod.
 
 
 1. INTRODUCTION
 
Risk has recently grown as a topic of interest within the field of
 
            communication, with two related problems generally of greatest
interest.
 
            The first involves the nature and consequences of news media
 
  representation of risk, typically with respect to technological or
 
        environmental controversies. The second area involves the difficulties
 
            encountered by scientists and other experts charged with informing
the
 
            lay public about their risk from various natural or man-made
hazards.
 
           The second area has most involved communication researchers in the
area
 
            of risk perception: understanding how the individual perceives and
makes
 
            judgments about risks faced in life.
 This investigation has the overarching goal of advancing understanding
 
                of how individuals come to a judgment about risk. This goal
includes a
 
            focus on the role of communication channels within the process of
risk
 
            judgment.
This project incorporates three sequential steps. First, the
 
      psychometric model of risk perception is evaluated for its validity
 
         under field conditions. Second, individuals are classified as
amplifiers
 
            or attenuators within the social amplification of risk model.
Finally,
 
            the characteristics of attenuators and amplifiers are explored, with
 
          special focus on their use of communication channels. These areas of
the
 
            literature will be addressed following a description of the case
study
 
            being utilized in this project.
 
2. CONTEXT: THE CASE STUDY
 
Since the Manhattan Project, the U.S. Department of Energy has operated
 
                its Ames Laboratory at Iowa State University in Ames, Iowa. The
wartime
 
            assignment of the Ames Lab was to purify enriched uranium for use in
the
 
            atomic bomb program. Since then the Ames Lab has continued to be
 
      involved in ceramics and metals development, methods of non-destructive
 
            analysis, and the development of environmental remediation
technologies.
 
            Ames Lab is also very active in the area of technology transfer.
Over the years, various activities and accidents at the Ames Lab have
 
               created a handful of waste sites in the Ames area. The Ames Lab
and the
 
            citizens of Ames have recently been in conflict over three such
 
     situations. The first involves the construction of a youth sports
 
       complex on a site contaminated in 1953 by low-level radioactive thorium.
 
            Despite being cleaned-up in 1988, community resistance stalled the
 
        city's plans to build there.[1] The second issue involves the clean-up
of a
 
            waste burial site in Ames where some 7,000 pounds of low level
 
    radioactive and other mildly hazardous materials have been interred for
 
            about 40 years. The DOE has recently concluded the remediation of
that
 
            site.[2]
The third issue involves a small research reactor that the Ames Lab and
 
                Iowa State University jointly operated from 1965 until
decommissioning
 
            in 1981. Some residents of the Ross Road neighborhood (one-half mile
 
          directly south from the former reactor site) believe that they are now
 
            part of a cancer cluster caused by the reactor. One section of the
 
        neighborhood had 13 cases. Two epidemiological studies have been done.
 
            The first was inconclusive and the second found that cancer rates in
the
 
            area were not significantly above normal.
To address concerns in the community, the Ames Lab has hosted four
 
            public forums, a community workshop, and has created an information
 
         repository on reserve at the ISU library. These issues have also
 
      received a fair amount of attention from the local newspaper.[3]
This investigation has as its primary focus the third issue: the
 
          perceived cancer cluster. This issue is being emphasized for two main
 
           reasons. First, it is of an enduring nature. It predates the other
 
        issues and will out-last them as well. Second, it is an intractable
 
         problem: there is no evidence of a cancer cluster D and even if there
 
           was such evidence it could not be causally associated with the former
 
           reactor. While the cancer cluster issue cannot be termed a "hot
button"
 
            topic, it is a issue that is widely known of in the community. The
key
 
            aspect is that due to the general recognition of the issue most
 
     individuals have had to come to some conclusion about it.
In more general terms, the situation faced by Ames residents is not
 
             uncommon. Many communities face relatively small and localized
hazard
 
           issues. Cases like Love Canal or Times Beach capture high media
 
     attention, but are atypical of the experience Americans more commonly
 
           have with environmental hazards. Understanding how individuals
perceive
 
            their situation relative to this variety of hazard and how they use
 
         information channels to form opinions is a task of some importance.
 
3. LITERATURE
 
To frame the primary questions of this investigation, three areas of
 
              literature will be briefly examined: the psychometric model of
risk
 
         perception, the social amplification of risk model, and communication
 
           channel utility.
The Psychometric Model. A considerable amount of research has been done in
 
                the area of risk perception. Dunwoody and Neuwirth prefer the
term "risk
 
            judgment" to emphasize the active information processing inherent in
the
 
            construct.[4] In any case, the psychometric model of risk perception
has
 
           grown from research that asks individuals to compare a range of
hazards
 
            based on a set of attributes, such as how well the hazard is
understood
 
            or how many people might be affected by the hazard. The research has
 
          consistently shown that people evaluate hazards not only on the
hazard's
 
            objective harm (e.g., deaths per year) but also on a range of more
 
        subjective criteria.[5]
The series of projects initiated by Fischoff and coworkers, and carried
 
                on by Slovic and associates, have shown that two important
dimensions
 
           can be seen to describe the perception of risk.[6]  [7] One dimension
of risk
 
            is termed dread. This is related to the scale of the risk and the
degree
 
            to which it impacts innocent individuals. The second dimension,
termed
 
            knowledge, involves how well a risk is understood and how observable
its
 
            consequences are.
The model has since been widely replicated and cited. The risk space
 
              model has been validated in various countries and researchers have
also
 
            expanded risk into additional dimensions using a wide variety of
 
      attributes such as the number of people affected or the voluntariness of
 
            the risk. [8]  [9] Non-human mortality and transgenerational effects
have also
 
            been examined as risk perception factors.[10] But the basic premise
of the
 
            two dimensional psychometric risk perception model has remained
 
     essentially unchanged.[11]
Some salient criticisms of the model have been addressed, including
 
             consistency among individuals in risk perception and the nature of
risk
 
            perception within a single technological domain.[12] The model held
up in
 
           these investigations, and also faired well in a reanalysis of the
 
       original data carried out by Gregory and Mendelsohn, who also found that
 
            perceived benefit plays a role in risk perception.[13]
While these studies do suggest a remarkable validity to the psychometric
 
                model, an important weakness remains in this direction of
research:
 
         risks are examined in the abstract. People can certainly be called upon
 
            to evaluate a set of risks, or attributes of a single risk, even
though
 
            they do not personally face that risk as an important aspect of
daily
 
           life. But is it safe to assume that people perceive or judge risk in
the
 
            same way when it involves a "live risk," a hazard or a risk
controversy
 
            that is part of daily life?
This question is not ignored in the literature. Risk perception studies
 
                have looked at specific risks in context. Such research has
typically
 
           been in the form of case studies.[14] While these studies and others
have
 
           great merit, they have not employed the psychometric risk perception
 
          model. It would be useful to know if the psychometric model can be
used
 
            to understand the perception of such local hazards. That is the
first
 
           issue to be addressed in this study.
Social Amplification of Risk. In response to the acute need to find some way
 
                to bring the diverse array of perspectives on risk, risk
perception, and
 
            risk communication together into an integrative framework, Kasperson
and
 
            coworkers formulated the social amplification of risk.[15]
Social amplification of risk holds that the communication and behavioral
 
                responses of individuals, groups, and institutions operating
under a
 
          risk event or controversy act as "amplification stations." It is the
 
          interaction among the risk interpretations and responses of these
 
       stations that determine the nature of the life course, or "rippling," of
 
            the risk event or controversy.
Within the concept of the amplification station exists a linkage between
 
                the macro level social functions of institutions or groups and
the micro
 
            level processes that operate within and between individuals.
Kasperson
 
            conceptualizes the micro level functioning as "individual stations
of
 
           amplification" and the macro level analog as the "social stations of
 
          amplification."
Social amplification is primarily a framework in which to utilize a
 
             variety of discrete theories and methods. Renn  describes how
social
 
          amplification might integrate the strengths and weaknesses of the
 
       social, psychological, and cultural approaches to risk.[16] The
distinctive
 
            problems of each approach D social relevance for psychometrics,
 
     complexity for sociology, and empirical validity for cultural theory D
 
            may tend to cancel out and allow for an emergent perspective.
The full application of the social amplification framework is a complex
 
                and long-term research goal. The present investigation has a
more
 
       specific focus: to develop an understanding of how individuals can be
 
           classified as being either amplifiers or attenuators within the
concept
 
            of the individual station of amplification. The existence of this
 
       dichotomy is a key micro-level prediction of the social amplification of
 
            risk model.
Since no objectively-based definition of risk exists in this case study
 
                it must be held that amplifiers and attenuators exist relative
to each
 
            other by way of their perception of risk from the threat D as
opposed to
 
            existing relative to some "true" condition. In other words, there is
no
 
            "correct" position in this controversy. Within this framework, the
 
        evaluation of individuals' perception of risk might be based on the
 
         psychometric model and this evaluation may in turn be used to group
 
         individuals as either amplifiers or attenuators.
While largely a semantic matter, placing amplifiers and attenuators
 
             under the umbrella of amplification (in its engineering guise as to
 
         either increase or decrease) can create unnecessary confusion. Rather,
 
            the idea will be recast as what it essentially is: polarization.
Channel Utility. The final aspect of this study seeks to examine the
 
           information channels people use in forming an opinion about a
perceived
 
            risk. The focus of concern will be the relative roles of mass
 
   communication, interpersonal communication, and other forms of informa
 
           tion-seeking behavior.
Chaffee presents an argument that the dichotomy of mass versus
 
        interpersonal communication has been endowed with excessive
polarity.[17]
 
           His analysis of the literature builds the case that individuals use a
 
           given channel based on how accessible the channel is and how likely
the
 
            individual believes it is that the desired information will be found
in
 
            a particular channel. In this light, it should be emphasized that
this
 
            research does not seek to determine mass or interpersonal supremacy,
but
 
            rather to assess the various ways in which individuals use both
channels
 
            in coping with a specific risk situation and how these channels
relate
 
            to other forms of information-seeking.
A handful of studies inform this question. Mazur and Hall examine how
 
               members of a New York county evaluate the risk of radon as either
a
 
         specific concern in the home or as a more diffuse national hazard.[18]
They
 
            find that neither interpersonal contact with family members nor mass
 
          media messages correlate with a specific concern about radon in the
 
         home. However, both were strongly correlated with a more general
concern
 
            about radon as a national hazard, with family influence considerably
 
          stronger than mass media influence.
McCallum and coworkers compared mass media and interpersonal channels as
 
                preferred ways of gathering information about toxic chemicals in
the
 
          local environment. They surveyed subjects in six communities around
the
 
            nation that were facing toxic chemical issues and found that mass
media
 
            channels were strongly preferred as sources of such information.
 
      Interpersonal sources were used only 12% of the time.
Following Tyler and Cook's observation that mass media impact
 
       social-level judgments more than individual-level judgments,[19] Coleman
 
          found that mass media are stronger than interpersonal channels in
 
       influencing society-level risk judgments.[20]  Mass media also had some
 
         influence on personal risk judgments, an effect which interpersonal
 
         channels did not have.
Contrasting the variety of findings in these studies goes to Chaffee's
 
                argument: interpersonal and mass communication channels have
varying
 
          roles in shaping people's perceptions and no broad generalization can
 
           hold. The role of other information-seeking behavior is less clear
from
 
            these studies. Chaffee points out that people have varying abilities
to
 
            use other information resources, such as libraries or expert
opinion.
 
            Dunwoody and Neuwirth also point out that people probably make
different
 
            use of various channels during the life span of their relationship
with
 
            a given risk. This study will attempt to address these issues,
looking
 
            at the relative usefulness of mass communication, interpersonal
 
     communication, and other forms of information-seeking in the process of
 
            risk judgment.
 
4. RESEARCH QUESTIONS
 
This study is both confirmatory and exploratory in nature. It seeks to
 
                confirm that the psychometric model describes how individuals in
this
 
           case study evaluate the given risk. It also seeks to confirm that the
 
           individuals engaged in this risk controversy can be productively seen
as
 
            being polarized into amplifiers and attenuators. Finally, it seeks
to
 
           explore the characteristics of polarization and look for important
 
        differences between the two camps. Toward these ends, this study employs
 
            a set of three research questions:
 
RQ1: Do individuals evaluate the given risk in terms of dread and
 
          knowledge as the psychometric model predicts?
 
RQ2: Based on the psychometric model, can individuals be consistently
 
              grouped as amplifiers and attenuators as polarization predicts?
 
RQ3: What differences are there between attenuators and amplifiers in
 
              terms of the demographic, risk, and channel use variables
measured?
5. METHODS
 
A mail survey instrument was developed to achieve the goals of this
 
             investigation. The instrument consists of three general segments.
First,
 
            the set of original psychometric model questions were modified
slightly
 
            to fit the specific issue at hand. While researchers have expanded
and
 
            modified this set of questions, the model is most strongly
associated
 
           with nine aspects of dread and five aspects of knowledge. These
 
     questions are shown in Table 1. The second part of the instrument
 
       consists of a set of questions seeking to ascertain what sources of
 
         information people have found to be useful in making judgments about
 
          this risk issue. Questions are also asked concerning satisfaction with
 
            attention paid to this issue by local media, Ames Lab officials,
 
      government representatives and others. Finally, a few general
 
   demographic variables are included.
The sampling unit is the non-rental household. Subjects are drawn from
 
                the northwest quadrant of Ames D the area defined by previous
 
   epidemiological studies and by stories in the Ames Daily Tribune .[21]  The
 
            area is also defined by social and geographic boundaries: the city
 
        limits to the north and west, and a large park to the south and east.
 
           The current Polk's City Directory for Ames was consulted as a
sampling
 
            frame. A random sample of 50% identified 223 households to
participate
 
            in the survey. Either principal adult member of the household could
 
         complete the questionnaire.
Mail survey procedures described by Dillman[22] were adhered to and the
 
             survey arrived in Ames on about September 23, 1994. The survey
return
 
           period was closed on November 1, 1994. At that time, 130
questionnaires
 
            were received for a return rate of 58 percent.
 
 
 
 
 
6. DISCUSSION OF THE RESULTS
 
The first task is to evaluate the psychometric model and determine if it
 
                can be used as an effective means of proceeding with the
analysis. The
 
            14 questions shown in Table I were entered into a factor analysis.
The
 
            rotated factor matrix was virtually uninterpretable. The results
 
      presented one strong factor with a mix of knowledge and dread variables
 
            and several weak factors with one or two variables each. No clear
 
       pattern was discernible.
The further evaluate the model, the communalities of the variables were
 
                examined to see if there were any very weakly associated
variables that
 
            might be appropriately excluded from the analysis (overall, the KMO
 
         statistic was adequate at .75). All variables but one had strong
 
      communality: the dread variable FATAL describing the likelihood that any
 
            illness from the risk would be fatal. This single weak variable was
 
         ejected and the factor analysis was again executed. Table II presents
 
           the results. With the exclusion of the one weak variable the rotated
 
          matrix provides a satisfying solution that conforms well to the
 
     prediction of the model.
One knowledge factor emerges made up of individual and scientific
 
           knowledge about the risk, how familiar the risk is to the individual,
 
           and how observable the consequences of the risk are to the
individual.
 
            Dread appears to be made up of 2 components. Factor 2 might be
called
 
           "pure dread" as it consists of elements that relate more clearly to
 
         fear: catastrophe, transgenerational effects, an increase in the risk,
 
            and being unable to calmly contemplate the risk. Factor 3 appears
most
 
            closely related to the idea of personal efficacy. This factor
involves
 
            the individual's ability to control exposure to the risk, ability to
 
          exercise choice in accepting the risk, and personal ability to reduce
 
           the risk. These all speak to the degree of individual agency with
respe
 
            ct to the risk. A fourth uninterpretable factor emerged that has an
 
         equal measure of both knowledge and dread.
Overall, the solution to the factor analysis supports the application of
 
                the psychometric model in this field situation, satisfying the
first
 
          research question. This is a fairly important result in itself since
the
 
            psychometric model has been most frequently applied to the
evaluation of
 
            individual perception of a range of risks considered in the
abstract.
 
           This analysis suggests, at least for the specific risk examined in
this
 
            case, that individuals may indeed evaluate risks they face through
 
        processes that can be understood in terms of knowledge and dread.
The four factor solution approaches but does not completely satisfy the
 
                proposition of a two-dimensional model of risk perception.
Questions
 
          remain: how do the four factors relate to one another, are the two
dread
 
            factors associated, and how should the fourth uninterpreted factor
be
 
           treated? To resolve these questions, factor scores were treated in a
 
          second-order factor analysis. The resulting two factor solution
grouped
 
            the dread factors together and grouped the knowledge factor with the
 
          fourth uninterpreted factor. With this result taken as evidence of
 
        association, variables for the dimensions of dread and knowledge are
 
          created by averaging the associated factor scores. Both variables are
 
           approximately normal with mean of 0 and are uncorrelated.
The second research question asks if the psychometric model might be
 
              applied to the task of separating individuals into amplifiers and
 
       attenuators. The literature on the recent concept of social
 
 amplification does not suggest what characteristics might indicate the
 
            two groups. To define the groups, the variables dread and knowledge
are
 
            plotted against each other and the scatter is divided at the mean
 
       created by the line with slope -1 running through the origin of the
 
         plane.
A discriminant analysis was run to confirm that this method of group
 
              determination is in fact consistent with the nature of the
original
 
         variables being used. The classification analysis used the full set of
 
            13 variables to predict the polarization groups. The classification
 
         matrix shows strong agreement that the 13 variables can identify two
 
          groups divided along mean responses to dread and knowledge. The
analysis
 
            correctly classifies 100% of the cases, identifying 54 amplifiers
and 40
 
            attenuators (some cases are lost due to incomplete responses). No
 
       significant differences were found between this group of 94 and the 36
 
            other survey respondents.
Research question three asks what differences might exist between the
 
               two polarization groups. Table III provides the significant
results,
 
          which can be organized in blocks: demographics, risk perception and
 
         behavior, satisfaction with institutional response to the issue, and
the
 
            reported usefulness of various information channels.
Two of the demographic variables show a significant difference between
 
                attenuators and amplifiers. Both gender and education are
significantly
 
            related to polarization. While amplifiers are slightly more likely
to be
 
            female, attenuators are very much more likely to be male (overall,
 
        respondents are 47 percent female). Attenuators also have typically
 
         completed more education, with over half having completed a graduate
 
          degree (recall that Ames is a college town). As these results suggest,
 
            further crosstabulation confirms that gender and education are
 
    significantly related with males having more education. The respondent's
 
            age is independent of group status.
Two other demographic variables approach significance and are worth
 
             discussion. It was suspected that the presence of cancer in the
 
     respondent's family might tend to be related to amplification. The data
 
            show this association, but only at a weak level of significance (p =
 
          .11). Length of residence in the area is also of interest since the
 
         reactor was removed in 1981. The data again provide weak evidence for
 
           this association (p = .098). Amplifiers tend to be newer to the area,
 
           with a mean length of residence of 15 years as opposed to 20 years
for
 
            attenuators. It may very well be a coincidence that the reactor was
 
         removed 14 years ago. Nonetheless, amplification may be intertwined
with
 
            a fear of the unknown, as some of these individuals have never
actually
 
            seen the former facility. It is also likely that generational
 
   differences may exist between the two groups with respect to
 
  environmental values and trust in government, for example
Risk perception and behavior variables provide a number of significant
 
                differences between the two groups. Taken together, these items
might be
 
            roughly conceptualized as worry. Most of these differences would be
 
         expected by virtue of the method of differentiating the groups. As
such,
 
            they also provide additional support for the validity of the two
groups.
 
            Amplifiers think about and talk about the cancer cluster issue more
 
         frequently than attenuators and also feel that both they and others are
 
            at greater risk. These differences are quite pronounced. On the
other
 
           hand, there were no significant differences found when asking how
long
 
            an individual had been aware of the issue or when asking what
actions an
 
            individual might have taken because of concern over the issue.
Further,
 
            attenuators are no different from amplifiers when it comes to
knowing
 
           other individuals in the area who have cancer.
The set of four satisfaction variables were all predictably different.
 
                Attenuators are uniformly more satisfied with attention paid to
the
 
         issue by Ames Lab, by elected officials, and by the news media. They
are
 
            also more satisfied with the results of the existing epidemiology.
The channel usefulness variables provide rather dramatic results in
 
             terms of                  non-significance. Of the 10 sources of
 
      information evaluated as having been useful in making a judgment about
 
            personal risk, only neighbors reveal a significant difference
between
 
           the two groups. Usefulness of family members differs between the
groups,
 
            but only weakly (p = .076). Amplifiers rated both neighbors and
family
 
            members as being more useful sources of information. The 8
information
 
            sources that showed no difference between the groups were the
newspaper,
 
            television, friends, physician, elected officials, Ames Lab
officials,
 
            public meetings, and the library's information repository.
Ranking of the usefulness of information channels is different for the
 
                two groups. An examination of the means of the 10 source
usefulness
 
         variables shows that both groups rated the newspaper as the most useful
 
            source of information. After that, the rankings diverge. Amplifiers'
top
 
            five are newspaper, neighbors, television, friends, and family.
 
     Attenuators' top 5 are newspaper, television, friends, neighbors and
 
          Ames Lab officials. Spearman's rho between the two rankings of 10
items
 
            is only .11 (not significant). This lack of association in the two
 
        groups provides evidence that the two groups are utilizing information
 
            sources differently.
For the next step in the analysis, all of the variables found to be
 
             significant (at p  < .1) in detecting group differences were
entered
 
          into a discriminant analysis. Table IV presents the results. The
 
      discriminant function produced by the 13 variables performed well,
 
        significantly accounting for about 40 percent of variance. Correlations
 
            between the individual discriminating variables and the discriminant
 
          function can be interpreted as an indication of the relative strength
of
 
            the variable's contribution to discriminating the groups. Further,
 
        grouping variables by the sign of the correlation can be useful (note
 
           that all variables are analyzed simultaneously and contribute to case
 
           assignment to both groups).
Variables are sorted by sign and listed by size of the correlation. It
 
                can be seen that evaluation of personal risk, risk to others,
frequency
 
            of thinking and talking about the issue, and the usefulness of
neighbors
 
            and family are grouped together. Conversely the grouping of positive
 
          correlations include education, gender, years of residence, and what
 
          might be taken as a set of variables indicating satisfaction with the
 
           institutional response to the issue.
 The sign of the correlation can sometimes be interpreted as indicating
 
                which group it is more closely associated with. In this case,
amplifiers
 
            were assigned the lower value so the set of negative correlations
are
 
           associated with amplification. Such interpretation must be approached
 
           cautiously. Here, it appears that a lack of the qualities indicated
by
 
            the negatively correlated variables indicate amplification and a
lack of
 
            the qualities indicated by the positively correlated values indicate
 
          attenuation.
 The more important results of this analysis are the manner in which the
 
                variables group and their ability to predict group membership.
Overall,
 
            the variables found to significantly differentiate between the
groups
 
           were in good agreement with the creation of the groups based on the
 
         psychometric model, with a classification rate from the discriminant
 
          analysis of 85%.
Finally, it is important to recognize that attenuators and amplifiers do
 
                not represent two homogeneous groups. There is a distribution of
risk
 
           perception within each group. To recognize this while still pressing
to
 
            examine for differences between the two groups, a continuous measure
of
 
            risk perception was created by averaging the dread and knowledge
 
      factors. This scale is then used as the dependent variable in
 
   hierarchical regressions utilizing the 13 variables which show
 
    significant differences between the groups. Variables are analyzed in
 
           blocks representing demographics, interpersonal sources, satisfaction
 
           with institutions, and worry. Results are presented in Table V.
For the analysis of the full sample, each block increments R2
 
       significantly (demographics only at p = .06). The full model achieves an
 
            adjusted R2 of .52 with 5 variables displaying significant partial
 
        coefficients: education, usefulness of neighbors, satisfaction with Ames
 
            Lab, and evaluation of both personal risk and other's risk.
The more interesting results come from a comparison of the two polar
 
              groups. For the attenuators, only the block representing
satisfaction
 
           with institutions significantly changed R2. The primary variable in
this
 
            column is satisfaction with Ames Lab. It appears possible that an
 
       interaction between gender and education prevents the demographic block
 
            from achieving significance. Inclusion of an interaction term did
not
 
           alter either the change in R2 or the adjusted R2 for any of the
blocks.
For amplifiers, the only significant block was the one involving the
 
              worry variables, although the years of residence variable in the
 
      demographic block is itself significant. For the worry block, evaluation
 
            of personal risk is the one strong element.
Stepwise regressions both focus and confirm the significant predictors
 
                of risk perception in the three groups, providing the following
models
 
            (alpha = .05, showing standardized betas):
 
Full Sample:
Risk = -2.5 + .48 (Personal Risk) + .27 (Other's Risk) - .16 (Education).
 
            Adj R2 = .54 p  < .001
 
Attenuators:
Risk = - 0.5 - .57 (Satisfaction Ames Lab) + .36 (Education).    Adj. R2
 
             = .33  p  < .001
 
Amplifiers:
Risk = -0.1 + .62 (Personal Risk) + .25 (Years Residence).     Adj. R2
 
            = .43  p  < .001
 
7. CONCLUSION
 
The analysis presented in this paper is a preliminary investigation in
 
                what will eventually be a larger case study. The primary mission
of this
 
            preliminary investigation is three-fold: to gather initial general
 
        information about the population and the controversy, to test the
 
       usefulness of the psychometric model, and to evaluate the viability of
 
            the polarization supposition made by the social amplification model.
The
 
            results of the analysis to this point suggest that each of these
goals
 
            have been met.
The secondary mission of this preliminary investigation conforms to the
 
                overarching goal of the larger case study: to understand the
role of
 
          communication in risk controversies and to add to the understanding of
 
            effective risk communication. Framing these goals in terms of risk
 
        polarization is not far from stating the problem as one of audience
 
         analysis. Much of the work done to date in risk communication has
 
       treated receivers as a somewhat homogeneous mass D evaluating, for
 
        example, the effectiveness of various forms of message construction
 
         without regard to important differences that might exist in the target
 
            audience. It is likely that risk communication could benefit greatly
by
 
            shifting some attention from message construction to audience
analysis.
With respect to the idea of audience analysis, an important lesson is
 
               demonstrated by the hierarchical regression models and their
stepwise
 
           counterparts: looking at a community's aggregate response to a risk
 
         controversy may be misleading. Different components in the community,
at
 
            least in this case, have significantly different orientations toward
the
 
            risk. These differing orientations may be most clearly seen if
 
    conceptualized and measured as polar opposites.
Utilizing such a light to examine the Ames case, it appears that two
 
              dominant forces are in play against one another: an evaluation of
 
       personal risk versus satisfaction with the institution managing the
 
         risk. Conceptually, these forces may approximate the notions of trust
 
           and outrage that have recently come to the risk perception
literature.[23]
 
            Subordinate to these forces are the demographically based variables
of
 
            education and years of residence in the community. Interpreting the
 
         stepwise regression equations for these two subordinate forces further
 
            demonstrates how aggregate characteristics may not be
representative.
 
           While attenuators tend to have more education, within their group
 
       greater education is associated with a greater perception of risk. And
 
            while amplifiers tend to have been residents of the area for fewer
 
        years, within their group those who have lived in the area longer
 
       perceive a greater level of risk.
The instrument used in this investigation provides only a rough look at
 
                the communication behaviors of this population. The main purpose
of this
 
            set of questions is to point the way for the construction of a new
 
        instrument to be applied in the near future. There are, however, two
 
          interesting results with respect to communication from the analysis so
 
            far.
First, there is apparently a dynamic involving mediated communication.
 
                Some background on the news coverage is in order. Very little
attention
 
            was paid to this issue by the television stations covering central
Iowa.
 
            Two of the three stations are located in Des Moines, some 40 miles
away.
 
            These stations pay only cursory attention to Ames. The third network
 
          affiliate is located in Ames but has what is widely considered to be
the
 
            weakest news operation. Attention from the nearest metro daily
 
    newspaper, The Des Moines Register, has been non-existent. The only
 
         significant coverage has been in the local newspaper, The Ames Daily
 
          Tribune.
It is not overly surprising that respondents would rank the local
 
           newspaper as the most useful source of information on the issue. What
is
 
            interesting, however, is that while attenuators and amplifiers alike
 
          rated the newspaper as the most useful source of information they
 
       differed significantly in their satisfaction with the attention paid to
 
            the issue by the news media overall. This result may suggest that
the
 
           two groups are processing news information differently.
More refined measures of media use are needed to support this
 
       proposition, but if respondents are basing their satisfaction in part on
 
            how the local newspaper has covered the issue (as opposed to how the
 
          other media outlets have not covered the issue) then amplifiers have
 
          rated the newspaper as most useful but were less satisfied while
 
      attenuators have rated the newspaper as most useful but have been more
 
            satisfied. This suggests the possibility that the same information
from
 
            the same source had different consequences for different audience
 
       segments. This conclusion is a stretch for these data but not an
 
      insupportable proposition.
The other interesting result from the communication variables is that
 
               the two groups did not differ in their evaluation of the
usefulness of
 
            any of the information channels except for neighbors and to a lesser
 
          degree family members. Since amplification is associated with finding
 
           neighbors and family members more useful as sources of information,
it
 
            is a fair conclusion to state that concern over risk in this case is
 
          driven by interpersonal communication to a greater degree than by
 
       mediated communication. This finding is in agreement with much of the
 
           research previously cited.
Finally, it is important to return to the notion of polarization, drawn
 
                from the social amplification model, that underlies this
analysis.
 
        Further investigation of the phenomenon of polarization might proceed
 
           along two lines, utilizing principles of social identification and
 
        models of information processing.
Social psychology has long recognized a process of group polarization.[24]
 
                In the most general terms, this is the process through which a
group can
 
            come to hold and express attitudes that are more extreme than those
held
 
            individually by its members. There are two dominant theories, social
 
          comparison and persuasive arguments. In social comparison, individuals
 
            compare their own views to the perceived average view held by the
group
 
            and then tend to shift their attitudes to the more extreme side of
the
 
            perceived group consensus. The final outcome of a group decision
 
      therefore tends to be polarized. In the persuasive arguments theory,
 
          individuals construct mental lists of arguments for and against a
 
       choice. When individuals discuss these lists in a group that tends to
 
           hold a polarized position, the arguments that are more universally
held
 
            by group members tend to be those that are more extreme in the
polarized
 
            direction. These more extreme arguments then tend to dominate both
 
        individual and group attitudes.
It is likely that some process of this nature has been in play in the
 
               Ames case. The Ames Lab, and its strong association with the
University,
 
            creates an in-group/out-group situation that may have served to
provide
 
            a polarization identity for some individuals. Further, the public
forums
 
            and coverage by the newspaper have likely provided a set of
polarized
 
           arguments around which the groups could have formed and strengthened.
Individual polarization has also been related to cognitive consistency.
 
                Chaiken and Yates found support for the hypothesis that
"thought-induced
 
            attitude polarization requires the presence of a well-developed
 
     knowledge structure."[25] They found in an experiment that individuals with
 
            a high degree of consistency tended to polarize on an issue more
readily
 
            than their low-consistency counterparts. This point of view
dovetails
 
           into an information processing perspective offered by Eagly and
Chaiken,
 
            in which individuals are hypothesized to process information in
either a
 
            systematic or heuristic manner.[26] Griffin and Dunwoody utilize
this
 
       perspective to build a model of information processing specifically for
 
            risk information.[27]
Griffin and Dunwoody consider the heuristic-systematic processing model
 
                and hypothesize on characteristics that might lead individuals
to
 
       process risk information in one way or the other. They propose a
 
      framework of variables that they organize in categories of demographics,
 
            characteristics of the hazard, individual worry, how individuals
feel
 
           that their information needs are being satisfied, and how confident
 
         individuals feel that they are able to gather needed information. If
 
          there is a relationship among heuristic-systematic processing,
cognitive
 
            consistency, and polarization D then it is also likely that the
model
 
           that Griffin and Dunwoody propose could be profitably related to
 
      polarization.
For the case of the risk controversy in Ames, it might be hypothesized
 
                that amplification is associated with (for example) systematic
 
    information processing of news information intertwined with
 
 interpersonal influence. Conversely, attenuation might be associated
 
          with more rapid "gut level" heuristic information processing of news
 
          information that is intertwined with pre-existing attitude structures
 
           related to trust in the institutions involved. Further research may
 
         approach these questions.
 
 
Table I.
Questions used for risk perception model with codewords.
Dread or knowledge questions indicated with (D) and (K).
 
 
 
For the following questions please give your own personal opinion about possible
risks
 
               to yourself from the former reactor. Circle the number you think
best locates your
 
             position on the 1 to 7 point scale.  (low values are for judgments
of low risk)
 
        Catastrophe (D) Do you think this kind of risk - that caused by small nuclear
reactors
 - has the potential to cause catastrophic death and destruction?
 
        Generations (D) Do you feel that any risk that may be posed from the former
reactor
 
               extends to future generations?
 
        Dread (D)       Is this the kind of risk you can learn to live with and calmly
deliberate
 
               about, or one that you constantly dread and worry about?
 
        Changing Risk (D)       Do you feel that your risk from the former reactor in Ames is
 
           increasing, decreasing, or staying the same?
 
        Personal Control (D)    How much control do you think you personally have over
avoiding
 
               possible risks to yourself from the former reactor?
 
        Personally Reduce Risk (D)      How easy or difficult would it be for you to reduce
any
 
              risk you might face from the reactor?
 
        Fairness of Risk (D)    Do you think the people who may have been exposed to some
risk
 
               from the reactor are the same people who may have benefited from
its operation?
 
        Fatality of Risk (D)    If you were to become ill from this risk, how likely is it
that
 
               the illness would be fatal?
 
        Personal Choice (D)     Do you think you have much choice over accepting any
possible
 
             risks from the former reactor?
 
        Harm Delay (K)  Is it more likely that any possible harm to you from the reactor
would
 
               have occurred immediately after exposure, or that it would be
delayed over time?
 
        Science Knows (K)       How knowledgeable do you think scientists are about any
possible
 
              risks from the former reactor?
 
        You Know (K)    How knowledgeable do you think you are about any possible risks
from the
 
               former reactor?
 
        Familiarity of Risk (K) Is this a new, novel kind of risk for you, or one
that's old
 
               and familiar to you?
 
        Observability of Exposure(K)    If you were exposed to a risk from the reactor,
how aware
 do you think you would be of your risk from that exposure?
 
 
 
 
 
 
 
Table II.
Factor Analysis of Psychometric Variables
 
 
 
                                Factor 1        Factor 2        Factor 3        Factor 4
        Variable        Mean    SD      Knowledge       Pure    Personal        Indeterminate
                                        Dread   Efficacy
        You Know (K)    4.6     2.0     .80
        Science Knows (K)       3.4     2.0     .77
        Observability of Exposure (K)   4.8     2.1     .56
        Familiarity of Risk (K) 4.6     2.3     .51
 
        Catastrophe (D) 2.7     1.8             .80
        Generations (D) 4.4     1.9             .69
        Changing Risk (D)       2.9     1.4             .61
        Dread (D)       2.6     1.8             .59
 
        Personal Control (D)    5.2     2.1                     .79
        Personal Choice(D)      5.1     2.0                     .78
        Personally Reduce Risk (D)      4.7     2.1                     .62
 
        Harm Delay (K)  5.9     1.6                             .75
        Fairness of Risk (D)    4.2     2.1                             .73
 
Percentage total variance       32.9    12.1     9.9        8.3
Eigenvalues        4.3     1.6   1.3        1.1
 
 
 
Factors were determined with an eigenvalue cutoff of 1.0, principle components
analysis with varimax
 
             rotation. Loadings under .5 are blanked. Four factors explain 63.2
percent of total variance. KMO
 
          statistic = .75.
 
 
 
 
 
 
Table III.
Comparison of Amplifiers and Attenuators.
 
 
        Item    Attenuators     Amplifiers
                        mean          mean         p
 
        Frequency talking about issue (1 never to 6 daily)      1.9     2.3      .004
        Frequency thinking about issue (1 never to 6 daily)     2.1     2.9      .002
        Risk to others (1 no risk to 7 high)    2.1     4.2     < .001
        Risk to self (1 no risk to 7 high)      2.1     4.3     < .001
 
        Usefulness of neighbors in making judgment (0 low to 7 high)    1.1     2.5      .003
        Usefulness of family members in making judgment (0 low to 7 high)       0.6     1.3     .076
 
        Satisfaction with epidemiology (1 not to 7 very)        4.5     3.1     .001
        Satisfaction with attention from Ames Lab (1 not to 7 very)     4.6     3.1      .001
        Satisfaction with attention from elected officials (1 not to 7 very)    3.9     2.4
<.001
        Satisfaction with attention from news media (1 not to 7 very)   4.5     3.5     .018
 
        Years as resident of northwest Ames     19.5    14.8    .098
 
 
        Item    Attenuators     Amplifiers
                      col %        col %        c2    p
 
        Respondent's gender                     4.5   .03
        female  34.2    56.6
        male    65.8    43.4
 
 
        Highest level of education completed                    18.5  <.001
        high school     16.2    22.6
        bachelor's      13.5    50.9
        graduate        70.3    26.4
 
 
        Have any members of immediate family had cancer?                        2.6   .11
        no      87.5    74.1
        yes     12.5    25.9
 
 
 
 
 
 Table IV.
Evaluation of significant differences between groups.
Saturated discriminant model using variables showing differences between groups
 
 
 
Function 1:   Canonical Correlation = .64     Wilks' Lambda = .59     c2  = 37.8
p
 
               < .001
Pooled within-groups correlations between discriminating variables and
discriminant function:
 
 
 
        .62     Personal risk
         .61    Other's risk
        .37     Frequency thinking
        .35     Frequency talking
        .31     Usefulness of neighbors
        .22     Usefulness of family
 
        -.45    Education
        -.42    Satisfaction with representatives
        -.39    Satisfaction with news coverage
        -.36    Satisfaction with Ames Lab
        -.33    Satisfaction with epidemiology
        -.22    Gender
        -.18    Years as resident of NW Ames
 
Risk Perception in Community Context
 
 
 
 
 
 
 
                                                  NO. OF           PREDICTED
GROUPS
ACTUAL GROUP                CASES                      1                       2
----------------------                ------------
-----------------------------
GROUP        1                44                37        7
AMPLIFY                 84.1%   15.9%
 
GROUP        2                 34                  5    29
ATTENUATE                               14.7%   85.3%
 
PERCENT OF CASES CORRECTLY CLASSIFIED:  84.6%
 
 
Table V.
Hierarchical Regressions: Risk on Significant Group Differences
 
 
 
 
 
FULL SAMPLE
ATTENUATORS
AMPLIFIERS
 
 
 
 
 
 
 
 
 
 
 
 
BLOCKS
beta
R2 cha
adj. R2
beta
R2 cha
adj. R2
beta
R2 cha
adj. R2
 
 
 
 
 
 
 
 
 
 
1. DEMOGRAPHIC
 
.09
.06
 
.16
.08
 
.11
.04
 
 
 
 
 
 
 
 
 
 
Gender
-.15
 
 
-.30*
 
 
.05
 
 
Education
-.20*
 
 
.36**
 
 
.20
 
 
Years Residence
-.09
 
 
.17
 
 
.29*
 
 
 
 
 
 
 
 
 
 
 
 
2. INTERPERSONAL
 
.16***
.21***
 
.00
.02
 
.05
.05
 
 
 
 
 
 
 
 
 
 
Neighbors
.27**
 
 
-.01
 
 
.22
 
 
Family Members
.10
 
 
-.01
 
 
.10
 
 
 
 
 
 
 
 
 
 
 
 
3. INSTITUTIONS
 
.12**
.30***
 
.33**
.30**
 
.02
.03
 
 
 
 
 
 
 
 
 
 
Ames Lab
-.22*
 
 
-.37**
 
 
-.04
 
 
Elected Officials
-.18
 
 
.22
 
 
.01
 
 
Epidemiology
-.06
 
 
.03
 
 
.01
 
 
News
.07
 
 
-.07
 
 
-.02
 
 
 
 
 
 
 
 
 
 
 
 
4. WORRY
 
.21***
.52***
 
.08
.30*
 
.40***
.38**
 
 
 
 
 
 
 
 
 
 
Thinking About Risk
-.02
 
 
-.07
 
 
.14
 
 
Talking About Risk
.01
 
 
-.01
 
 
.06
 
 
Personal Risk
.42***
 
 
.04
 
 
.49***
 
 
Other's Risk
.28*
 
 
.28
 
 
.01
 
 
 
 
 
 
 
 
 
 
 
 
 
* p  < .10    ** p  < .05    *** p  < .01
 
Betas are partial coefficients from regression on each block independently.
Dependnet variable is
 
             risk, as an averaged score of the variables knowledge and dread.
High values on risk equal
 
        perception of great risk.
 
 
 REFERENCES
 [1]  Armour, S. (1994, July 7). Uncertainty lingers over sports site. The Am
es Daily
 
               Tribune, p. A-1.
[2]  Armour, S. (1994, Nov. 26). Meeting p
lanned on waste removal. The Ames Daily
 
            Tribune, p. A-1.
[3]  Armour
, S. (1994, Jan. 10). Cancer fear lingers in Ames neighborhood. Westside
 
 
    residents look for answers. The Ames Daily Tribune, p. A-1; Armour, S.
(1994, Jan. 1
9).
 U of I researcher begins study of area. Will look at cancer rates in west Ames.
Th
e
 
                 Ames Daily Tribune, p. A-2; Armour, S. (1994, Jan. 21). Study
cle
ars neighborhood of
 
                 cancer link. The Ames Daily Tribune, p. A-1; Ar
mour, S. (1994, May 2). DOE reviews
 
               cancer studies. The Ames Daily
Tribune,  p. A-1; Armour, S. (1994, May 20). $4 million
 
                 for cleanup
 of toxic sites. The Ames Daily Tribune, p. A-1.
[4]  Dunwoody, S., & Neuwirth, K. (199
1). Coming to terms with the impact of
 
       communication on scientific an
d technological risk judgments. In L. Wilkins & P.
 
             Patterson (Eds.)
, Risky Business: Communicating Issues of Science, Risk, and Public
 
 
 Policy (pp. 243-286). Beverly Hills, CA: Sage.
[5]  Morgan, M. G. (1993, July). Risk a
nalysis and management. Scientific American, 269
 (1), pp. 32-41.
[6]  Fischhoff, B.,
 Slovic, P., Lichtenstein, S., Read, S., & Combs, B. (1978). How
 
              safe
 is safe enough? A psychometric study of attitudes toward technological risks
and
 
 
               benefits. Policy Sciences, 9, pp. 127-152.
[7]  Slovic, P., Fischhoff, B
., & Lichtenstein, S. (1985). Characterizing perceived
 
              risk. In R. W.
 Kates, D. Hohenemser, & J. Kasperson (Eds.) Perilous Progress:
 
          Tec
hnology as Hazard (pp. 91-123). Boulder CO: Westview.
[8]  Englander, T., Farago, K., S
lovic, P., & Fischhoff, B. (1986). A comparative
 
            analysis of risk per
ception in Hungary and the United States. Social Behavior, 1, pp.
 
                 5
5-66; Namur, E., & Sornay, C. (1988). La perception des risques sociaux
(Perception of
 
 societal risks). Villeneuve d'Ascq: Universite Charles-de-Gaulle;
Kleinhesselink, R.
 
 
                 R., & Rosa, E. A. (1991). Cognitive representation of risk
perception
s: A comparison of
 Japan and the United States. Journal of Cross-Cultural Psychology,
22, pp. 11-28;
 
               Teigen, K. H., Brun, W., & Slovic, P. (1988). Societ
al risk as seen by a Norwegian
 
               public. Journal of Behavioral Decisi
on Making, 1, pp. 111-130.; Keown, C. F. (1989).
 
                 Risk perception of
 Hong Kongese vs. Americans. Risk Analysis, 8, pp. 435-455.
[9]  Mullett, E., Duquesnoy
, C., Raiff, P., Fahrasmane, R., & Namur, E. (1993). The
 
              evaluative f
actor of risk perception. Journal of Applied Social Psychology, 23 (19),
 
 
      pp. 1594-1605; Morgan, M. G., Slovic, P., Nair, I., Geisler, D.,
MacGregor, D.,
 
 
            Fischhoff, B., Lincoln, D., & Florig, K. (1985). Powerline frequency
electric and
 
              magnetic fields: A pilot study of risk perception. Ris
k Analysis, 5, 139-149; Slovic,
 
                 P., Fischhoff, B. & Lichtenstein, S
. (1985). Regulation of risk: A psychological
 
             perspective. In R. G.
 Noll (Ed.), Regulatory Policy and the Social Sciences (pp.
 
             241-278
). Berkeley: University of California Press; Kraus, N. N., & Slovic, P. (1988).
 
 
             Taxonomic analysis of perceived risk: Modeling individual and group
percept
ions within
 
                 homogeneous hazard domains. Risk Analysis, 8, pp. 435-4
55.
[10]  Hohenemser, C. et al. (1986). Methods for analyzing and comparing
technologic
al
 
               hazards. In V. Covello, J. Menkes, & J. Mumpower (Eds.),
Contempor
ary Issues in Risk
 
                 Analysis : Vol. 1: Risk Evaluation and Managemen
t (pp. 249-274). New York: Plenum.
[11]  Kleinhesselink & Rosa (1991); Mullet et al. (1
993).
[12]  Kraus, N. N., & Slovic, P. (1988). Taxonomic analysis of perceived risk:
Mo
deling
 individual and group perceptions within homogeneous hazard domains. Risk Analys
is, 8,
 
               pp. 435-455.
[13]  Gregory, R., and Mendelsohn, R. (1993). P
erceived risk, dread, and benefits. Risk
 Analysis 13 (3), pp. 259-264.
[14]  Eyles, J
., Taylor, S. M., Baxter, J., Sider, D., & Willms, D. (1993). The social
 
 
    construction of risk in a rural community: Responses of local residents to
the 1990
 
 
                Hagersville (Ontario) tire fire. Risk Analysis, 13 (3), pp.
281-290
; Fitchen, J. M.,
 
                 Heath, J. S., & Fessenden-Raden, J. (1987). Risk
perception in community context: A
 
                case study. In B. B. Johnson & V
. T. Covello (Eds.), The Social and Cultural C
 
          onstruction of Risk (
pp. 31-54.) Dordrecht: Reidel; Edelstein, M. R. (1988).
 
         Contaminate
d Communities. Boulder, CO: Westview Press; Edelstein, M. R., & Wandersman,
 
 
         A. (1987). Community dynamics in coping with toxic contaminants. In I.
Altman &
 A.
 
               Wandersman (Eds.), Neighborhood and Community Environment. New
York: Plenum.
[15]  Kasperson, R. E., Renn, O., Slovic, P., Brown, H. S., Emel, J., Gob
le, R.,
 
           Kasperson, J. X., & Ratick, S. (1988). The social amplificati
on of risk: A conceptual
 
                 framework. Risk Analysis 8 (2), pp. 177-18
7.
[16]  Renn, O. (1992). Concepts of risk: A classification. In S. Krimsky & D.
Gouldi
ng
 
               (Eds.), Social Theories of Risk (pp. 53-82). Westport, CT:
Praeger
.
[17]  Chaffee, S. (1982). Mass media and interpersonal channels: Competitive,
 
 
        convergent, or complimentary? In G. Gumpert & R. Cathcart (Eds.),
Inter/Med
ia:
 
            Interpersonal Communication in a Media World (2nd ed., pp. 57-7
7). New York: Oxford
 
                University Press.
[18]  Mazur, A., & Hall, G.
 S. (1990). Effects of social influence and measured
 
          exposure level o
n response to radon. Sociological Inquiry, 60 (3), pp. 274-284.
[19]  Tyler, T. R. & Co
ok, F. L.  (1984). The mass media and judgments about risk:
 
             Distingui
shing impact on personal and societal level judgments. Journal of Personality
 
 
           and Social Psychology, 47 (4), pp. 693-708.
[20]  Coleman, C. (1993). The in
fluence of mass media and interpersonal communication
 
               on societal and
 personal risk judgments. Communication Research 20,  pp. 611-628.
[21]  Armour (1994,
Jan. 19).
[22]  Dillman, D. A. (1978). Mail and Telephone Surveys: The Total Design Met
hod. New
 
               York: John Wiley.
[23]  Flynn, J., Burns, W., Mertz, C. K.,
 & Slovic, P. (1992). Trust as a determinant
 
               of opposition to a high-
level radioactive waste repository: An analysis of a structural
 model. Risk Analysis 1
2 (3), pp. 417-429; Kasperson, R., Klaidman, S., Houk, V.,
 
              Silberge
ld, E., Blair, E. H., Lynn, F., Sandman, P., Alm, A., & Vacor, T. (1987). Trust
 and cr
edibility: The central issue? In J. C. Davies, V. T. Covello, & F. W. Allen
 
 
         (Eds.), Risk Communication (pp. 43-62). Washington, DC: The
Conservation Founda
tion.
[24]  Isenberg, D. J. (1986). Group polarization: A critical review and
meta-anal
ysis.
 
               Journal of Personality and Social Psychology 50 (6), pp. 1141-1
151; Mackie, D. M.
 
              (1986). Social identification effects in group p
olarization. Journal of Personality and
 Social Psychology 50 (4), pp. 720-728; Pavitt,
 C. (1994). Another view of group
 
             polarizing: The reasons for one-s
ided argumentation. Communication Research 21 (5), pp.
 625-642.
[25]  Chaiken, S., &
Yates, S. (1985). Affective-cognitive consistency and
 
     thought-induced
 attitude polarization. Journal of Personality and Social Psychology 49
 
 
     (6), pp. 1470-1481.
[26]  Eagly, A. H., & Chaiken, S. (1993). The Psychology of At
titudes. Fort Worth:
 
             Harcourt Brace Jovanovich.
[27]  Griffin, R J.,
 & Dunwoody, S. (1995). Superficial and systematic uses of
 
         informatio
n about environmental risks. Paper presented to "Communication and Our
 
 
    Environment: An Interdisciplinary Conference," Chattanooga, TN, March 30 -
April 2,
 
 
                1995.