Award R000239333

Election Learning and Persuasion: An Experimental Study

Executive Summary of the Research

Aims

There has been a long-standing debate about the extent to which media messages
affect the political knowledge and preferences of the general public. This research
project sought to assess the effects of media messages on the opinions of UK voters
during a period of generally heightened political awareness – the 2001 general
election campaign. The project sought to answer three main empirical questions:

1. What impact, if any, do different types of media (television, press, the

internet) have on political learning for different types of voters?

2. What influence, if any, do Party Election Broadcasts (PEBs) have on voters’

party preferences?
3. What impact, if any, does exposure to ‘positive’ or ‘negative’ television news
about political parties have on voters’ party preferences?

Approach

We adopted an experimental approach because of the difficulty of assessing the

impact of media messages on people’s knowledge and opinions using more traditional
methods. For example, with survey-based or focus group research designs, it is rarely
clear precisely which political messages people have been exposed to during any
given time period. Thus it is difficult to show that any expressed political views, or
even changes in views over time, are the result of a particular set of media exposures
– rather than the result of the myriad other influences and messages to which
individual respondents might be exposed over time.

Experimental methods have the huge advantage that respondents’ views can be
carefully assessed at the outset of the experiment. Respondents can then be exposed
to a specific, known, media stimulus. And their views, and thus any changes in those
views, can then be systematically measured again immediately after the exposure. By
the careful use of control groups – who are not exposed to the ‘test’ message – the
analysis can determine the extent to which exposure to a particular media message
produces a change in people’s knowledge and opinions. Moreover, because
respondents’ views are solicited at the pre-exposure stage, statistical controls can be
applied for a range of prior demographic and attitudinal characteristics. Thus it is
possible, for example, to assess the extent to which people’s prior political knowledge
or their prior political preferences affect the way that they react to particular media
messages.

1
The Experimental Method

The research involved an experimental study of the way in which voters acquire new
information and change their political preferences as a result of exposure to different
communication media. It built on an earlier study conducted by Sanders and Norris
during the1997 UK general election campaign. A representative sample (N=919) of the
Greater London population was subjected to a series of ‘media exposure experiments’
during the 2001 general election campaign. A pre-exposure test determined subjects’
political knowledge and electoral preferences. Different groups of subjects were exposed
to different tightly controlled media messages presented in either television, newspaper
or web-page format. Post-exposure tests then measured how far each subject’s political
knowledge and/or views changed as a result of this exposure. Additionally, some
subjects were contacted by telephone in the three weeks after taking part in order to
ascertain whether any change in knowledge and/or views had persisted.

Main Findings

(1) About Political Learning during Election Campaigns

Campaign learning did occur. In the midst of the June 2001 British general election,
despite the extensive political news available on radio, television, current affairs
programmes, newspaper supplements and the Internet, about half the public appeared
unaware about some of the key issues featured in party political debates, such as the
stance of the Conservative towards the euro and their promise to cut taxes. After
relatively brief exposure to more information about these issues, interspersed by other
election and non-election news, the public’s issue knowledge rose significantly on
eight out of eleven items. Moreover controls monitoring changes in knowledge with
items where no information was provided in the experiments demonstrated that this
was not merely an unintended effect or artefact of experimental conditioning. The
increase in awareness of party policies persisted after the experiments, with minimal
decay, for the period monitored by the call-back study.

Campaign learning was similar among different media. People learnt as much from
exposure to television news and party websites as from newspapers lending no
support to the claims of print superiority.

Campaign learning was greatest among the least informed. The learning gains were
strongest among those who were least knowledgeable prior to the experiments,
thereby closing the knowledge gap found in the pre-test measures. Nevertheless the
less knowledgeable gained as much or more from reading broadsheet as tabloid
newspapers, contradicting our expectations that there would be an interaction effect
between the type of media and type of voter,

(2) About the effects of PEBs

PEBs that contained material that repeated familiar messages from the ‘long
campaign’ had no significant effect on party preferences. This effect assisted the

2
Liberal Democrats, whose media voice had not been prominent during the long
campaign.

There was a strong differentiation between ‘advocacy’ broadcasts (which present a

positive picture of the advertising party) and ‘attack’ broadcasts (which present a
negative view of opposing parties). Advocacy broadcasts were either neutral or
positive in terms of their intended effects on party preferences. Attack broadcasts, in
contrast, were counterproductive – they actually increased support for the target of the
attack. Our data permit us only to speculate as to why this ‘counterproductive’ effect
is observed. However, we suspect that it reflects a general sense of frustration among
UK electors with ‘negative campaigning’ by political parties – and a consequent
desire to ‘punish’ perpetrators of PEBs who employ negative images. ‘Attack’ PEBs
by Labour and the Conservatives in 2001 actually increased support for their
opponents.

(3) About the effects of ‘positive’ and ‘negative’ television news

Positive and negative television news exerts modest, but statistically non-significant,
effects on partisan preferences. This contrasts with the ‘asymmetric’ findings of our
equivalent 1997 study, when we found that positive news enhanced a party’s image
while negative news failed to detract from it. We are currently re-analysing our
combined 1997 and 2001 datasets and will report on our findings in a subsequent
paper.

Implications

The research has relevance for theoretical debates about the character of
contemporary democracy – in particular, the extent to which citizens are, and can be,
well informed about politics. People do learn from what they read, see and hear in the
media. The media have an important role to play in creating and sustaining a
politically informed electorate. The research also has relevance for policy debates
about the role of PEBs in contemporary political discourse. PEBs allow parties that
typically receive relatively little media exposure to get their election messages across
to voters and to benefit accordingly. However, parties that espouse an ‘attack’
approach to PEBs seriously risk alienating the very people – the neutrals and
supporters of other parties – to whom they are trying to appeal. In the UK, an
‘advocacy’ approach to political advertising works far more effectively than ‘attacks’
on opponents.

3
 Full Research Report

General Background

There has been a long-standing debate about the extent to which media messages
affect the political knowledge and political preferences of the general public.
Although it seems ‘obvious’ that such messages ‘must’ have some sort of effect,
establishing their character and extent has proved difficult in previous research, for
two major reasons. First, people tend to seek out those media outlets that reflect their
pre-existing political dispositions and preferences. This in turn makes it difficult to
assess how far people use a particular media outlet because it reflects their pre­
existing political views and how far they adjust their views because of the character of
the messages that the outlet conveys. Second, with survey-based or focus group
research designs, even if the same individuals are interviewed at different points in
time, the researcher is unable to determine the precise set of political messages to
which each respondent has been exposed. Thus it is difficult to show that any
expressed political views, or even changes in views over time, are the result of a
particular set of media exposures – rather than the result of the myriad other
influences and messages to which individual respondents might be exposed over time.
Taken together, these two constraints mean that it is difficult systematically to assess
‘media effects’ using traditional research designs.

The advantage of an experimental design is that people’s political knowledge and

attitudes can be measured both immediately before and after exposure to a known
media stimulus. Experimental studies accordingly provide a very precise, if
necessarily restricted, method of assessing the impact of media messages on public
opinion. Our experiments were conducted with a representative sample (N=919) of
the Greater London population during the 2001 general election campaign.

All empirical social research has to strike a balance between (a) the provision of a
detailed characterisation of how people think and act, and (b) the extent to which any
findings can be generalised beyond the specific group of people studied. Sample
surveys approaches tend to emphasise (b) while focus group and other qualitative
techniques tend to stress (a). The experimental approach employed here attempted to
combine the detailed characterisation of changes in individual perceptions that can be
derived from experiments with the generalisability that is typically associated with the
use of a representative sample. For logistical and financial reasons, our experiments
had to be conducted in a single location. This meant that we were not able to draw on
a sample of respondents (or ‘subjects’) that was directly representative of the UK as a
whole. However, we did assemble a sample (N=919) that was representative of the
Greater London population in terms of both socio-demographics and political
preferences. We were also able to compare the demographic and attitudinal profiles
of our respondents with those of the 2001 British Election Study post-election survey.
The similarities of these profiles (with the exception of ethnicity – there are more
people from ethnic minorities in London than elsewhere in the UK) suggest that our
sample was, in practice, broadly representative of the UK population in general. This
in turn implies that our experimental results are likely to be applicable to the UK
population in general.

4
Objectives

The general objectives of the study were as follows.

1. To extend the experimental analysis of media effects that we conducted during the
1997 general election campaign (Norris and Sanders, 1997; Sanders and Norris, 1999;
Norris et al 1999). This work had shown that media exposure significantly affected
voters’ issue-priorities and their perceptions of the issue agenda. It had also shown
that UK voters in 1997 were far more likely to change their political preferences as a
result of exposure to ‘positive’ as opposed to negative news images of the various
political parties and leaders. We were concerned to establish whether or not our
results from 1997 extended to the different conditions of the 2001 campaign. We also
wanted to determine whether or the observed effects of straightforward television
news broadcasts (which generally strive to be politically ‘neutral’ over time) also
applied to more explicit party political advertising. We accordingly extended our
research design to include Party Election Broadcasts (PEBs).

2. To show, as precisely as possible at the micro-level, the extent to which media

messages could affect the knowledge levels and political perceptions of a
representative group of voters. We know from election surveys, from panel studies
and from aggregate time-series analyses that there are consistent correlations between
media-usage patterns, on the one hand, and political knowledge and political
preferences, on the other. Establishing the direction of causality is extraordinarily
difficult using these other varieties of data. An experimental approach can provide
appropriate micro-level evidence that people’s knowledge and opinions are indeed
affected by the media messages to which they are exposed.

3. The specific questions that the study sought to answer for the 2001 general election
campaign were:

4. What impact, if any, do different types of media (television, press, the

internet) have on political learning for different types of voters?

5. What influence, if any, do Party Election Broadcasts (PEBs) have on voters’

party preferences? In short, how successful are parties’ attempts to persuade
voters to view them in relatively more favourable terms?
6. What impact, if any, does exposure to ‘positive’ or ‘negative’ television news
about political parties have on voters’ party preferences?

Methods

As noted, the research used an experimental design because it was considered the
most feasible method of measuring the effects of political learning and persuasion.
Cross-sectional surveys are the most common way of exploring the influences of the
information sources upon electors but they can say little about the causal effects of
exposure to different varieties of message. Panel surveys are more useful but also
limited in measuring media exposure. Experiments allow the analyst precisely to
control the specific messages that respondents see and hear. For example, exposure to

5
a particular news story or party political broadcast, and 'pre-post' designs allow us to
measure 'before' and 'after' shifts in attitudes and behaviour. This enables the
researcher to make relatively strong statements about the causal effects of exposure to
messages.

Many experiments are limited because they rely upon small groups of student
respondents and it is difficult to generalize from these results to the general
population. In contrast this project used a large cross-section of the public selected by
quota sample to be broadly representative of the Greater London electorate, involving
over 900 participants in total. The experiments were conducted in the midst of the
May-June 2001 British general election campaign. This combination of an
experimental design with a broad cross-section of the public in a natural setting
allows us to draw causal inferences that have application well beyond the particular
population included in our experiments.

Experiments are becoming increasingly common in political science (Iyengar and

Simon 2000), nevertheless the way that they are designed is critical to the results.
Since the methodology is less familiar than other approaches we need to outline the
specific research design in some detail, including the selection of participants, and the
experimental procedures that we employed.

Selection of Participants

In total, 919 participants drawn from the general electorate were included in the total
study, more than most experimental designs. Of these, 480 were used in the tests of
campaign learning, including 91 participants who constituted a control group that was
not shown any campaign information. The remaining participants were involved in
separate persuasion experiments concerned with the impact of positive and negative
news and party election broadcasts.

The experiments were based in a Greater London location, in central Ilford, with
participants drawn primarily from south-east England. The location was selected to
provide a diverse group of Londoners including office-workers and casual shoppers,
in a mixed constituency, drawing participants from around the region who were
attracted to a popular shopping mall. Most came from the borough of Redbridge but
others were drawn from the surrounding boroughs of Barking and Dagenham,
Havering, Newham, Tower Hamlets, and Waltham Forest, as well as elsewhere in
Greater London. Respondents from one area should not be understood to represent a
random sample of the whole UK electorate, but they were selected by a professional
fieldwork agency using quotas for age, gender, class, ethnicity, and past vote to reflect
the social and political background of the Greater London population. The accuracy
with which the demographic profile of respondents matched that of the Greater
London population is shown in Appendix A. Moreover the fact that the media outlets
under comparison are nation-wide, with the same papers read and TV evening news
watched from Cornwall to Cumbria (with the exception of some minor differences in
the media systems in Scotland and Wales) means that in principle we would expect to
find the same results if the experiments had been conducted anywhere else in
England. Nevertheless the generalisability of the results rests not on the selection of a
random sample of participants, as in a survey design, but on the way that subjects

6
were assigned at random to different experimental groups. Any difference in the
response of groups should reflect the stimuli treatment rather than their social
backgrounds or prior political attitudes. The only exception concerned the group of
Internet users, where we only tested those who already had some prior experience of
surfing the Internet. This group can therefore be regarded as reflecting the general
background of the online community, with higher educational and occupational status,
rather than the Greater London electorate as a whole.

Experimental Procedures

One potential problem of experiments is that participants may alter their own
behaviour given the artificiality of the research setting and their perceptions of the
aims of the study. To help overcome this potential problem, and to avoid the danger
of selection bias by discouraging participation by the politically apathetic, the briefing
did not mention that study was about election news. Respondents were informed
(falsely) that we were conducting market research to learn how different sorts of
people evaluate and understand newspapers, television and Internet news. The main
experiments employed a single-shot rather than a repeated design, to avoid
respondents becoming unduly conditioned by the research process itself, although one
call back was used after the election among a sub-sample to monitor any 'decay' effect
of the stimuli. The experiments were conducted during the last part of the election
campaign, from 26th May until 6th June, election eve.

Participants were asked to complete a short (15-minute) pre-test questionnaire about

their media habits, political attitudes, and personal background. They were then
assigned at random to watch a 30-minute video compilation of television news, or to
read selected newspaper stories, or to read selected off-line party web pages that had
been downloaded and edited into a dedicated site. Respondents subsequently
complete a short (15-minute) post-test questionnaire, after which they were paid for
their time and given a letter about the purpose of the experiment. A member of the
research team interviewed respondents unable to read the questionnaire separately on
a face-to-face basis. To reduce the artificiality of the exercise, the atmosphere was
designed to be relaxed, with refreshments provided in a comfortable environment.
The whole process lasted for about an hour for each participant. The experiments
were carried out during the middle weeks of the official general election campaign.
The aim of this timing was to examine the attitudes of participants who had been
subjected to the intensive barrage of political coverage that characterizes an election
period, again to increase the realism of the experimental conditions.

The classification of the ‘type of media’ is relatively straightforward in Britain where

national newspapers and evening television news on the five major terrestrial national
channels continue to be the main sources of campaign television, despite some
fragmentation of channels and outlets from cable, satellite and broadband services in
recent years (Norris et al. 1999). The study compared the effects of exposure to five
sources of campaign information: television news, newspapers (distinguishing
between the ‘quality’ or heavyweight broadsheets and the popular ‘red-top’ tabloids),
the party websites, and election broadcasts by the three major parties. Details of the
methods are given in Appendix B. Approximately 70-140 respondents were tested in
each experimental group. In addition, a TV news control group was shown a typical
30-minute evening news bulletin with content unaltered by any sort of experimental

7
manipulation, and 'sports' coverage instead of election news in the middle. Other
experiments replicated tests for the impact of negative and positive news.

The compilation of television news stories and PEBs was chosen to represent a
‘typical’ evening news programme during the campaign. We drew on stories
recorded from all the main news programmes on the terrestrial channels in the three
months prior to polling day. The videos were edited to follow the same format. This
consists of a ‘sandwich’, with ten minutes of identical, standard footage at the top and
bottom of each programme and one of the different experimental video stimuli (news
or a PEB) in the middle ‘core’. A similar process was followed to select equivalent
information on the same issues in newspapers, campaign website and party political
broadcasts (see Appendix B). Respondents were not told which media was being
shown to which group, or even that different media were being used by different
groups of respondents. The numbers of participants involved in each of our
experimental groups is shown in Table 1.

Table 1: Number of Participants in each Experimental Group

Experimental Group Number of Participants

Campaign Learning Groupsa

Watch Television News 100
Read Broadsheet Newspapers 139
Read Tabloid Newspapers 73
Surf Party Websites 77

Party Election Broadcast Groups

Conservative Advocacy Broadcastb 49
Conservative Attack Broadcast 47
Labour Advocacy Broadcast 46
Labour Attack Broadcast 46
Liberal Democrat Advocacy Broadcast 51

Positive/Negative Television News Groups

Positive Conservative News 50
Negative Conservative News 50
Positive Labour News 50
Negative Labour News 50

Explicit Control Group – shown sport video 91

Total 919
a
Each Learning Group was given a compiled set of stimuli containing the same
factual information about the election campaign. For details, see Appendix B.
b
The Conservative Party broadcast only ‘attack’ PEBs during the 2001 campaign. To
simulate the effects of an ‘advocacy’ PEB, we used a Channel 4 Comment broadcast
by Teresa May, the party’s then Shadow Education Secretary.

8
Hypotheses Tested and Empirical Findings

The detailed theoretical rationales (and appropriate scholarly citations) for the
hypotheses that we tested are outlined in the two papers that we have specified as
‘Nominated Outputs’ from this project. Here we provide only summary statements of
the core hypotheses and the rationales for them. We outline our hypotheses under two
headings: those concerned with political learning and those concerned with political
persuasion.

Political learning hypotheses

1. The overall picture. Our interest in ‘political learning’ focuses on the extent to
which people exposed to identical factual political information acquire greater
political knowledge regardless of whether print, television or the Internet is used to
convey the information. Our first general hypothesis is that

H1: Regardless of the medium that is employed, there will be a significant

knowledge gain among people who are exposed to campaign information.

We assessed ‘knowledge gains’ by asking our respondents a series of factual

questions in our pre- and post-test questionnaires. Our ‘test’ groups were asked 11
factual (true/false) questions. They were then given factual information about these
questions in their test exposures. Finally, they were asked the same questions in the
post-test and the change in response was measured. In order to control for the
possible effects of ‘priming’ that might have derived from asking true/false questions
about particular topics in the pre-test, we also asked 9 (different) factual questions in
the ‘control’ group pre-test. The control group participants were given no further
relevant factual information about these questions. The control group’s responses to
the questions were also measured in the post-test. Table 2 summarises the results of
this exercise. (For full item-by-item details, see Nominated Output #1, Tables 1 and
2). Note that the average percentage answering the set of questions correctly
increased in the test group from 53% to 61% - an average ‘knowledge gain’ of 8
percentage points. This compares with an average knowledge gain of 1% among the
control group. In short, exposing participants to politically relevant information did
(significantly) affect their political knowledge.

Table 2: Average percentage of factual questions answered correctly, pre-post

Pre-test Post-test Pre-Post Change

Test Groups
Average percent answering 53 61 +8
eleven questions correctly
Control Group
Average percent answering nine 45 46 +1
questions correctly

9
2. The impact of different types of media exposure. It is sometimes alleged that
newspapers are more informative news sources than electronic news media. This
putative ‘print superiority’ is held to derive partly from the greater concentration that
is required in reading, as opposed to listening to watching, and partly from the fact
that readers have greater control over the pace at which material is covered. In
opposition to this claim, it is argued that, on the contrary, the combination of voice,
pictures and/or graphics employed in television news and on web-pages is just as
effective as print in conveying political knowledge to the public. In our view, the
message is more important than the medium. We examine these two contending
arguments in our second general (null) hypothesis:

H2: There will be no important differences in the campaign learning effects

that are observed among those exposed to similar factual information using
print, television news, or web-based media.

We tested this hypothesis by constructing an 11-point ‘knowledge’ scale based on the

answers to the 11 factual questions that we asked participants in each of our political
knowledge test groups. We then compared the average individual-level changes in
the knowledge scale scores between the pre- and post-tests. The results, broken down
by type of media, are shown in Table 3. There were significant knowledge gains
across all media types – and the increases were not substantially different across the
different media types. Interestingly, tabloid and broadsheet exposure produced very
similar knowledge gains.

Table 3: Mean Change in the 11-point Political Knowledge Scale. Pre-Post

Pre-test Post-test Knowledge Eta Sig N.

knowledge knowledge Gain .
(Mean) (Mean)
Control (No stimuli) 5.61 5.91 .30 91

Any information source 5.79 6.76 .97 .16 *** 389

Television news 5.50 6.74 1.24 .29 *** 100
Any newspaper 5.95 6.76 .81 .16 *** 212
Broadsheet papers 6.34 7.18 .84 .19 *** 139
Tabloid papers 5.19 5.96 .77 .16 * 73
Party websites 5.75 6.80 1.05 .25 *** 77

Note: The strength (Eta) and the significance of the difference between the knowledge
gains among the control group and each of the groups exposed to information sources
was measured by ANOVA. *** Sig. 0001; ** .01; * .05.

10
3. The impact of prior factual knowledge. We explored several hypotheses relating
to the ways in which prior factual knowledge might affect campaign learning. We
expected to find that people who already knew quite a lot would be less likely than
people who knew very little to increase their knowledge as a results of exposure to
factual information. This expectation was borne out by our data. Controlling for
standard demographics and prior political interest, politically knowledgeable
participants were significantly less likely to make knowledge gains as a result of our
test exposures. (See Nominated Output #1, Table 4). We also expected to find that,
controlling for other relevant factors, exposure to information in tabloid newspapers
would be more likely to produce knowledge gains among the less well informed than
exposure to broadsheets. This expectation was confounded by our data. Among our
‘know nothing’ respondents, exposure to broadsheet and tabloid newspapers produced
virtually identical knowledge gains. (See Nominated Output #1, Table 6).

Political persuasion hypotheses

Our experiments in this context examined the impacts of PEBs and of ‘positive’ and
‘negative’ television news items. Our investigation of the effects of exposure to
television news repeated experiments that we had conducted in 1997. Political parties
go to great lengths to try to persuade journalists to ‘spin’ news stories in their favour.
Journalists typically try to resist these efforts but it is still the case that certain stories
make ‘good news’ for one party or ‘bad news’ for another. There is thus an indirect
persuasion process involved in the genesis of television news. In our 1997 study, we
found an ‘asymmetric’ pattern in which people’s views of political parties were
significantly enhanced by ‘positive’ (‘good’) news about them but were not damaged
by ‘negative’ (‘bad’) news. The results of our 2001 experiments produced a different
pattern in which the effects of positive and negative news on people’s images of the
parties were symmetric but not statistically significant. We are still in the process of
exploring these relationships and will report on the differences between 1997 and
2001 in a later paper.

Our 2001 investigation of PEBs was, in any case, much more directly concerned with
direct attempts at political persuasion. PEBs are one of the few opportunities that
parties get to communicate directly with voters, unfiltered and un-criticised by
journalists or news editors. The question is: do they work? Do they ‘consolidate our
supporters’ or ‘convert the undecided’, or both?

Parties themselves make very simple assumptions about the impact of PEBs.
‘Advocacy’ broadcasts, in which the party presents of a positive image of itself, will
enhance its image in the eyes of voters. ‘Attack’ broadcasts, which present a negative
image of the opposition, will damage the opposition’s image in voters’ eyes. In both
cases, the relative standing of the broadcasting party vis-à-vis it opponents should be
improved. These effects are summarised in what we term the “rational politician’s
core assumption” about the effects of PEBs. This ‘axiom’ can be stated as follows.

H3: If voters are exposed to our party’s unmediated political messages in

PEBs, other things being equal, they will be persuaded to take a more
favorable view of our party. Undecided voters may be persuaded to vote for

11
 us; our own partisans may be strengthened in their commitment; while our
opponents’ supporters may be weakened.

Our analysis of the effects of PEBs is predicated on the assumption that this core
‘party axiom’ looks at only a part of the picture – that it ignores the importance of the
political context in which PEBs are transmitted. We accordingly develop a set of
‘modifying hypotheses’ that try explicitly to take account of the context of the
campaign and the differential impact of PEBs on different types of voter. (For details
of the justifications for these hypotheses, see Nominated Output #2). These
modifying, which also differentiate between the effects of advocacy and attack PEBs,
can be stated as follows:

H4: For the major parties (Labour and Conservative in 2001), where the
public already has already developed fairly stable and familiar images through
constant repetition, further advocacy information obtained by watching a
single election broadcast will have little or no effect on voters’ party images.

H5: Minor parties (like the Liberal Democrats in 2001) are less familiar to the
public, so a single election broadcast will be more likely to shape the public’s
image of these parties.

H6: In a situation where attacks are widely regarded as implausible (as with
the Conservative and Labour attack broadcasts in 2001), attack messages have
the counterproductive effect of boosting support for the target of the attack.

H7: If the partisans of an embattled party (like the Conservatives in 2001) are
exposed to attack messages directed at a rival party, they are likely to take
note of the messages and to take an even more negative view of the rival party.

Our tests of these hypotheses began by measuring pre- to post-test changes in our
participants’ ‘party images’ of the Labour, Conservative and Liberal Democrat
parties. These ‘party image’ scores were constructed by combining participants’
responses to three items (all measured on 0-10 scales): their feelings of affect on
towards the party leader; their liking of the party itself; and the probability that they
ever vote for the party. We then estimated a multiple regression equation that
controlled for a variety of demographic characteristics and prior (pre-test) political
attitudes (including all pre-test party image scores), as well as exposure to the various
PEB stimuli. These stimuli were: Conservative and Labour advocacy PEBs;
Conservative and Labour attack PEBs; and a single Liberal Democrat advocacy PEB:
the Liberal Democrats did not broadcast an attack PEB in 2001. The reference
category for media exposure contained all of our participants who were not shown
PEBs. In order to test H7 we created an interaction term between strong Conservative
partisanship/not and exposure to a Labour attack PEB.

The results of our analysis are reported in Table 4. The results confirm H4-H7 rather
than H3. The Conservative and Labour advocacy broadcasts had no effect on voters’
party images (H4). In contrast, the Liberal Democrats’ advocacy broadcast
significantly improved the LibDems’ party image and simultaneously weakened that
of the Conservatives (H5). The Conservative and Labour attack PEBs had the
counterproductive effect of significantly increasing support for the targets of those

12
Table 4: The impact of party election broadcasts on party image scores

Change in pre- to post-test Party Image Score

Conservative Labour Lib-Dem
b St err b St err b St err
PRIOR CONTROLS
Conservative Partisan .86** .14 .01 .15 .12 .16
Labour Partisan -.02 .12 .48** .13 .04 .14
Liberal Democrat Partisan .27 .16 -.20 .17 .21 .18
Pre-test Cons Image score -.11** .02 -.03 .02 -.01 .02
Pre-test Labour Image score .01 .02 -.11 .02 .00 .02
Pre-test Lib Dem Image score -.06** .02 .02 .02 -.10** .02
Age .00 .00 .00 .01 .00 .00
Male .02 .08 .07 .08 .14 .09
Middle Class (ABC1) .00 .08 .17 .09 -.02 .10
White -.05 .10 .00 .11 -.10 .12
Graduate -.04 .10 -.17 .11 .16 .11
Interest in Politics .04 .05 -.07 .06 -.03 .06
WATCHED PEB
Conservative Attack PEB .33 .18 .49* .23 .12 .22
Conservative Advocacy PEB -.23 .17 -.10 .18 -.12 .19
Cons Partisan who watched -1.3** .44
Conservative Attack PEB
Labour Attack PEB .36* .17 .00 .18 -.23 .20
Labour Advocacy PEB .18 .18 .25 .19 .11 .20
Liberal Democrat Advocacy PEB -.41* .19 .00 .21 .45* .22

Constant .33 .20 .67** .21 .71** .23

2
Corrected R .08 .05 .03
Standard error of estimate 1.03 1.10 1.19
N 775 756 752

Notes: Estimation by OLS. ** Denotes significant at .01 or better; * significant at .05.

N of cases is less than 919 because not all participants completed all questions in
both the pre- and post-test questionnaires.

13
attacks (H6). [See the significant, positive coefficients for the Labour attack PEB in
the Conservative image model (b=.36) and for the Conservative attack PEB in the
Labour image model ( b=.49)]. Our data permit us only to speculate as to why this
‘counterproductive’ effect is observed. However, we suspect that it reflects a general
sense of frustration among UK electors with ‘negative campaigning’ by political
parties – and a consequent desire to ‘punish’ perpetrators of PEBs who employ
negative images. In any event, attack PEBs by Labour and the Conservatives in 2001
appear to have increased support for their opponents. Finally, Conservative partisans
did react differently from other participants to the Conservative attack PEB (H7). The
impact of their exposure to this stimulus was significantly to increase their distaste for
Labour (see the b=-1.3 for the interaction term in the Labour image model). Overall,
the “rational politician’s axiom” about PEBs derives very little support from our data.

Conclusion

The research shows that by conducting media experiments on a representative sample

of the Greater London population, is possible show at the micro-level that people are
indeed influenced by the media messages to which they exposed. Our campaign
learning experiments show that the message is more important than the medium:
people increase their political knowledge at similar rates in response to exposure to
factual information provided using a range of different mediums. These findings have
relevance for theoretical debates about the character of contemporary democracy – in
particular, the extent to which citizens are, and can be, well informed about politics,
and the devices that might be employed in order to increase their political knowledge..
Our experiments into PEBs show that the main beneficiaries of PEBs are parties
which have a relatively low profile prior to the election campaign and which adopt an
advocacy rather than an attack approach to their broadcasts. Party managers in the
UK should note that strong partisans are the only people who react to attack
broadcasts in the way that the “rational politician’s axiom” predicts. For other voters,
attack PEBs can have the counterproductive effects of increasing sympathy for the
target of the attack.

Activities

Pippa Norris and David Sanders, ‘Knows Little, Learns Less? An experimental study
of the impact of the media on learning during the 2001 British general election’.
Papers presented at the Annual Meeting of the American Political Science
Association, San Francisco, September 1st 2001.

Pippa Norris and David Sanders, ‘Knows Little, Learns Less? An experimental study
of the impact of the media on learning during the 2001 British general election’.
Papers presented at the PSA Elections, Parties and Public Opinion Meeting,
University of Sussex, September 15-17th 2001.

Pippa Norris and David Sanders, ‘Message or Medium? Campaign learning during the
2001 British general election’. Paper presented to the ECPR Joint Sessions, Panel on
Participation and Information, University of Turin, April 2002.

14
David Sanders and Pippa Norris, ‘Advocacy versus Attack: The effects of Party
Election Broadcasts in the 2001 UK general election’. Paper presented to the ECPR
Joint Sessions, Panel on Antipolitics and the Media, University of Turin, April 2002.

David Sanders, ‘The state of democracy in contemporary Britain’. Paper presented at

Harvard University, April 2003.

David Sanders, ‘ Television news, economic perceptions and political preferences in

the UK’. Paper presented at Nuffield College, Oxford, May 2003.

Pippa Norris ‘The legitimacy of the European Union: The role of the media.’
Conference on Democracy and Legitimacy of the European Union and International
Organizations.’ European Parliament, Brussels October 11-12 2002.

Outputs

Pippa Norris and David Sanders, ‘Message or Medium? Campaign learning during the
2001 British general election’. Paper accepted for publication in Political
Communications

David Sanders and Pippa Norris, ‘Advocacy versus Attack: The impact of political
advertising in the 2001 UK general election’. Paper under ‘revise and resubmit’ with
Public Opinion Quarterly.

Pippa Norris. 2002. ‘Do Campaigns Matter for Civic Engagement? US Elections
1952-2000.’ Do Political Campaigns Matter? Eds. David Farrell and Rudiger
Schmitt-Beck. London: Routledge.

Pippa Norris. 2002. ‘Political Communications.’ In Comparing Democracies 2: New

Challenges in the Study of Elections and Voting. Eds. Lawrence LeDuc, Richard
Niemi and Pippa Norris. London: Sage.

Pippa Norris. 2003. ‘Zappers in de politiek? De invloed van de media op

verkiezingscampagnes.’ In Ethische Perspectieven 12(2): 86-102. (In Dutch).
Pippa Norris. 2003. ‘Tuned out voters? Media Impact on Campaigns.’ In Ethical
Perspectives 9(3) (In English).

Dataset, Campaign Learning and Persuasion: An Experimental Study, lodged with the
UK Data Archive.

Information about the project at www.pippanorris.com

Impacts

Pippa Norris. 2004. Consultancy report ‘E-government and democracy.’ United

Nations Public Sector Report. New York: United Nations.

15
A submission, summarising our findings, was made to the Electoral Commission’s
consultation on Party Political Broadcasting. The Commission’s report was published
in January 2003.

Future research priorities

This study has consolidated the use of experimental methods to analyse media effects
on political knowledge and preferences in the UK. There remains enormous scope for
the application of experimental methods in UK political science. We have
concentrated our efforts on election campaigns, but it is clearly important to show
how media effects also operate in more ‘normal’ times. A series of smaller scale
experiments, conducted on a continuing basis throughout the electoral cycle, could
add significantly to our understanding of how people come to change their political
and economic perceptions and partisan preferences.

16