Political Behavior, Vol. 22, No. 2, 2000
LOCAL CONTEXT, RETROSPECTIVE
ECONOMIC EVALUATIONS, AND VOTING:
The 1997 General Election in England
and Wales
Ron Johnston, Charles Pattie, Daniel Dorling,
Iain MacAllister, Helena Tunstall and David Rossiter
Local context is widely believed to influence voting behavior with, for example, the
voters’ evaluation of the state of their local economy affecting whether they choose to
reward or punish the incumbent government. Such reward-punish models apply in the
United Kingdom at the national scale: those who believe that the government has
delivered prosperity vote for its return, whereas those who believe that its policies
have produced a worsening economic situation vote against it. This article shows that
the operation of this calculus varies spatially, according to the level of unemployment
in the voter’s home area: the higher the local level of unemployment the lower the
probability of someone who thought that government polices had delivered national
prosperity voting for the incumbent government. It also shows that this is a consequence
of cross-pressured situations. Those who thought that the government’s policies had
delivered both national and local prosperity were very likely to vote for it; those who
thought that the policies had delivered national but not local prosperity were less likely
to vote for it—especially in areas of high unemployment.
In a recent paper, Books and Prysby (1999) argued that studies of the
relationship between retrospective economic evaluations and voting should
take account of the local context in which those evaluations are formed and
the consequent voting decisions. They presented data on evaluations of President
Bush’s economic performance, taken from the 1992 American Election
Professor Ron Johnston, School of Geographical Sciences, University of Bristol, University Rd.
Bristol BS8 1SS UK (R.Johnston@bristol.ac.uk).
121
0190-9320/00/0600-0121$18.00/0  2000 Plenum Publishing Corporation
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER122
Study, which showed that respondents’ economic evaluations were related to
the rate of unemployment in their home states, clearly suggesting the operation
of a local contextual effect. People respond, they argued, not to the national
situation but rather to their local situation—probably as it is presented to them
in the local media. The present article builds on their foundation, looking at
variations in voting in the 1997 general election in England and Wales according
to respondents’ evaluations of national standards of living.
Studies of retrospective economic voting—both sociotropic and egocentric—
are based on a reward-punish model: Those who believe that the economy has
been performing well, and credit government policies with that, will vote for
the government, whereas those who believe that it has been performing badly,
and blame the government for that situation, will vote against it. But when
voting sociotropically, at what spatial scale do they make their decision regarding
economic performance—at the national level or at some scale smaller than
that? Books and Prysby’s (1999) data, for example, suggested that respondents
took account of their state’s economic performance, but not that of their
community within that state (defined as their home metropolitan area)—largely,
they suggest, because local media focus on state-level economic data (such as
unemployment rates), and it is these that influence voters’ evaluations. On
similar lines, Atkinson and Partin (1995) show that although voting decisions
in senatorial elections are linked to presidential popularity, in gubernatorial
elections perceptions of the condition of the state economy are crucial influences
on the voter’s choice.
Within Great Britain most economic performance data—notably those on
unemployment levels and changes in dwelling prices—are reported for the
country as a whole and for major regions (such as Scotland, Wales, Northern
Ireland, and the nine ‘standard regions’ within England). Relatively little is
reported in the national media at a more detailed scale. The United Kingdom
has recently experienced strong spatial variations in economic performance at
a variety of scales, however. These may be the basis of local contextual variations
in evaluations of economic performance, and the voting decisions that follow,
because they are apparent to voters in their everyday experience. They observe
events in their local environments, talk about them to their neighbors, and, in
some cases, hear about them in local media (local radio and—usually weekly—
newspapers, many of which report local changes in unemployment rates and
major contributors to those changes, such as plant closures and openings). In
this article, we explore the hypothesis that voters respond to their local economic
context, as well as to the national situation. In our study, we combined data
from the 1997 British Election Study for voters in England and Wales with a
newly created ecological data set containing information on each respondent’s
local area.1
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 123
RETROSPECTIVE SOCIOMETRIC EVALUATIONS AND VOTING IN
ENGLAND AND WALES
The 1997 British Election Study (BES) included a cross-sectional survey of
some 2,731 respondents in England and Wales.2
They were asked a number
of questions regarding economic conditions, including the following pair:
General standard of living
( . . . thinking back to the general election of 1992—the one where John Major won
against Neil Kinnock) the general standard of living: has it increased a lot, increased
a little, stayed the same, decreased a little, decreased a lot?
Why
Do you think this is mainly the result of the Conservative government’s policies or
for some other reason?
Cross-tabulation of the responses to the first of these questions against reported
vote in 1997 shows the expected relationship—given that the Conservative
party had been in power for the previous 18 years (Table 1). Those who thought
that the general standard of living had increased a lot were more likely to vote
for the incumbent government than for any other party. Those who thought
it had fallen were more likely to vote for an opposition party, especially Labour,
which had led the Conservatives in the opinion polls since late 1992. The
punish component of the economic voting model is very apparent: those who
thought that the general standard of living had increased a lot were twentyfive
times more likely to vote for the incumbent government than were those
who thought it had fallen a lot.
Turning to the second part of the question, the majority of the respondents
TABLE 1. Vote at the 1997 General Election According to
Evaluations of Changes in the General Standard of Living
(percentages of the row totals)
Evaluation C L LD O DNV N
Increased a lot 60.2 17.7 9.7 3.6 8.8 113
Increased a little 39.3 21.7 14.2 3.5 21.3 756
Stayed the same 24.4 34.4 15.3 4.6 21.3 783
Fallen a little 8.5 50.0 12.6 5.0 23.9 682
Fallen a lot 2.4 61.9 11.6 6.1 18.0 294
Don’t know 13.7 33.3 12.7 9.9 30.4 102
Key to columns: C—Conservative; L—Labour; LD—Liberal Democrat; O—Other; DNV—did
not vote; N—number of respondents.
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER124
associated the changes in general standards of living with government policy.
As the last column of Table 2 shows, however, a larger proportion of those
who thought that it had fallen blamed the government than was the case for
crediting the government among those who thought that it had improved. (Of
those who thought that the general standard of living had ‘increased a lot,’ 73
percent credited the government with this; of those who thought it had ‘decreased
a lot,’ 88 percent blamed the government.) The greatest differences
in voting were between those who credited/blamed the government for changes
in the general standard of living, rather than ‘other reasons.’ Those who credited
government policies with standards of living that had increased a lot, for example,
were thirty-seven times more likely to vote Conservative than those who
blamed government policies for them falling a lot (Table 2). Voting for the
incumbent government was linked very clearly to evaluations of the impact of
its policies on the national economic situation.
LOCAL CONTEXT: DEFINING THE SCALE
In recent decades, there have been substantial spatial variations in economic
performance within the United Kingdom. The economic restructuring that
characterized the 1980s saw major decline in many of the country’s traditional
TABLE 2. Cross-tabulation of Vote at the 1997 General Election
Against Evaluation of Changes in the General Standard
of Living and the Attributed Reason for those Changes
(percentages of the row totals)
Evaluation and Reason C L LD O DNV N
Increased a lot
Government 66.3 12.5 11.3 3.6 6.3 80
Other 46.7 26.7 6.7 3.2 16.7 30
Increased a little
Government 47.5 17.2 10.9 3.5 20.9 478
Other 22.7 31.5 20.7 3.2 21.9 251
Stayed the same
Government 26.0 36.1 14.9 3.7 19.3 462
Other 21.3 31.1 17.1 6.0 24.5 298
Fallen a little
Government 7.1 53.1 12.4 4.3 23.1 590
Other 16.7 29.8 14.3 9.4 29.8 84
Fallen a lot
Government 1.8 63.3 11.3 6.1 17.5 275
Other 11.8 47.1 17.6 0.0 23.5 17
Key to columns: C—Conservative; L—Labour; LD—Liberal Democrat; O—Other; DNV—did
not vote; N—number of respondents.
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 125
manufacturing industries, accompanied by substantial growth in service industries.
There was a clear geography to these consequences of ‘Thatcherism,’
with the northern regions suffering from the manufacturing decline while
regions further south benefited most from the service-sector growth (see Lewis
and Townsend, 1989). This north-south divide was apparent in a number of
indicators, not least the level of unemployment, and it remained in place
through the 1990s, although slightly more muted. These spatial variations in
economic prosperity were related—at an ecological scale—to variations in
support for the Conservative party throughout the 1980s and, again in somewhat
more muted more form, in the 1990s. (On the 1980s, see Johnston, Pattie, and
Allsopp, 1988; on the 1990s see Johnston et al., 1998.) The implication was
that people in the north blamed the Conservative governments for the economic
situation there and voted against them, whereas those in the south credited
them with delivering economic prosperity and voted to return them to power.
These are inferences from ecological data, however, and there have been
few studies of spatial variations in economic voting and their direct relationships
to the local context (see, however, Pattie and Johnston, 1995; Pattie, Dorling,
and Johnston, 1997). One of the problems for such studies in the UK is the
absence of data with which those relationships could be explored: general
election voting returns are available for constituencies only (average electorate
c.70,000), for example. Census data from which indicators of economic wellbeing
can be derived are available at three scales: the constituency, the local
government electoral ward (average population c.5000, though with great variation
between urban and rural areas), and the census enumeration district
(average population c.500). In the past, it has been possible to incorporate
census data at the first two of these scales to enable studies of local contexts.
Few studies have been undertaken, however, with only one published at the
ward scale (Harrop, Heath, and Openshaw, 1991). The extent and nature of
contextual variations in voting behavior in the UK is still largely based only on
informed speculation, as in Miller’s (1978) modeling using survey and ecological
data at the constituency scale.
As part of a research project associated with the 1997 British Election Study
(BES) we have built ‘bespoke neighborhoods’ for each of the respondents
in England and Wales (though not in Scotland, because of data-matching
difficulties). Using the enumeration districts as our building blocks, we identified
for each respondent those districts containing the nearest 500, 1,000, 2,500,
5,000 and 10,000 individuals to her or his home, thus giving us detail on local
contexts never before available. These are used in the analyses here; ward and
constituency data are used for comparative purposes.
Books and Prysby (1999) argued that the unemployment rate is a good
indicator of local economic conditions, as it is a widely cited index in media
discussions. As noted previously, this is particularly apposite in the British
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER126
situation because of the spatial variations in unemployment associated with
Thatcherism and its aftermath. Within the UK, however, there is no equivalent
of the state for which unemployment rates are regularly reported. But many
people are aware of local levels of unemployment without such data, both from
their own social networks and from general observation and discussions of wellbeing
in their area. A majority of the respondents to the 1997 BES reported
that they either read a local newspaper or watch the regional TV news programs:
74 percent said they read a local newspaper at least once a week (most are
weeklies) and only 11 percent said that they never do; 56 percent claimed that
they watched regional TV news broadcasts daily (only 10 percent said that they
never watch them). Discussions of local employment issues—plant closures
and expansions, for example, as well as of national unemployment statistics—are
common in those media, so that if there are local contextual effects in economic
evaluations and voting in the British context, they should be related to the
unemployment rates.
One slight problem with using this index is that the last date for which
unemployment data are available at the enumeration district scale is the 1991
census. Unemployment fell substantially between then and the 1997 general
election, but the spatial variations remained consistent—as demonstrated by
other data available for the constituency scale only: the relative relief of the
map of the geography of unemployment did not vary through the decade, even
though its absolute relief did change substantially. Unemployment data are
available monthly at constituency and ward scales. The correlation between
the number unemployed at the time of the 1991 census and at the time of the
1997 general election was 0.94 at the constituency scale and 0.96 at the smaller
ward scale.
The following analyses are in two parts. For the first, exploratory stage we
have grouped the bespoke neighborhoods (and the wards and constituencies)
into six categories according to their 1991 unemployment rates, as shown in
the top block in Table 3. This enables us to illustrate the very substantial
variations in unemployment at relatively small scales, which analyses at the
constituency level do not disclose (the smaller the scale, the wider the spread
of local unemployment rates). Second, to test whether there are significant
relationships between local context and voting, holding constant economic
evaluations, we use logistic regression analysis.
ECONOMIC EVALUATIONS AND VOTING
Because of the relatively small numbers of respondents who said that the
general standard of living had either increased a lot or fallen a lot, for the
exploratory analyses of variations in voting behavior according to local unemployment
level we have reduced the number of categories to three: those who
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 127
TABLE 3. Voting Conservative by Evaluations of Changes in the General
Standard of Living and Local Context (percentage in each cell
voting Conservative)
Unemployment
Level n500 n1000 n2500 n5000 n10000 Ward Const
Number of respondents in each cell
0.0–2.9 88 37 11 0 0 56 0
3.0–5.9 699 705 637 549 503 605 400
6.0–8.9 665 669 721 820 831 742 931
9.0–11.9 395 443 515 555 545 509 590
12.0–14.9 260 252 281 297 346 258 410
15.0< 557 558 499 443 439 560 400
General standard of living improved
0.0–2.9 41.7 * * * * 50.0 *
3.0–5.9 48.7 50.5 54.9 50.5 48.9 52.3 52.0
6.0–8.9 44.8 41.9 41.7 48.4 49.5 45.7 46.1
9.0–11.9 43.8 43.2 35.4 34.2 34.2 33.1 42.1
12.0–14.9 34.7 41.9 40.6 30.1 34.3 33.3 25.0
15.0< 27.0 24.5 26.5 28.3 27.5 30.4 32.3
General standard of living stayed the same
0.0–2.9 44.4 * * * * * *
3.0–5.9 36.1 37.2 40.8 42.9 42.0 38.9 39.8
6.0–8.9 28.9 29.0 28.6 30.6 31.5 31.1 30.1
9.0–11.9 22.0 20.2 19.4 14.2 15.2 16.6 22.3
12.0–14.9 10.1 11.8 14.1 14.2 12.8 12.3 12.2
15.0< 10.6 12.2 9.3 7.2 11.5 11.8 10.2
General standard of living fell
0.0–2.9 19.0 * * * * * *
3.0–5.9 11.3 12.9 13.6 17.0 15.0 12.8 16.4
6.0–8.9 8.9 8.3 7.3 7.0 8.2 8.2 7.5
9.0–11.9 4.4 6.0 5.5 4.2 4.0 2,5 5.8
12.0–14.9 4.0 1.0 4.0 3.3 3.3 5.3 3.8
15.0< 1.7 2.2 2.5 2.5 3.5 4.0 2.8
*Less than 20 respondents in the cell.
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors;
ward—electoral ward; const—Parliamentary constituency.
thought the general standard of living had (a) increased, (b) stayed the same,
or (c) fallen.
In the context of the 1997 British general election, the important voting
decision for the present analyses is whether or not the respondents chose to
vote Conservative: those wishing to reward the government should have done
so, whereas those wanting to punish it could either have voted for one of the
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER128
opposition parties (there was considerable tactical voting at the 1997 general
election: Evans, Curtice, and Norris, 1998; Johnston et al., 1997) or abstained
(turnout at the election was 71 percent, the lowest level since the Second
World War). Thus, the remaining tables cover Conservative voting only, as the
incumbent party whose economic record was the focus of the voters’ rewardpunish
calculus.
The lower three blocks of Table 3 explore vote by retrospective evaluation
at each of the seven spatial scales—the five bespoke neighborhood scales plus
wards and constituencies. They show that, with very few minor exceptions, at
all spatial scales the higher the local unemployment rate the lower the percentage
voting Conservative—holding constant retrospective evaluations. Thus, for
example, at the constituency scale among those who thought that the general
standard of living had improved, 52.0 percent voted Conservative where the
unemployment rate was less than 6 percent, compared to 46.1 percent where
it was between 6.0 and 8.9, 42.1 percent where it was between 9.0 and 11.9,
25.0 percent where it was between 12.0 and 14.9 percent, and slightly larger
(32.3 percent) where unemployment was more than 15.0 percent.
Table 3 shows that among those who thought that the general standard of
living had improved, at every spatial scale the percentage voting Conservative
in the areas where unemployment was above 15.0 percent was just over half
of the comparable percentage where unemployment was below 6.0 percent.
Among those who thought that the general standard of living had stayed about
the same (the third block in Table 3), the percentage voting Conservative in
the areas of highest unemployment was less than one-third of the percentage
in the areas of lowest unemployment. Finally, among those who thought that
the general standard of living had fallen, the difference between the areas of
highest and lowest unemployment in their Conservative voting percentages
was even larger: At the n5000 scale, for example, the percentage voting Conservative
in the areas of highest unemployment was only one-sixth of that in the
areas of lowest unemployment.
These tabulations suggest that whatever the respondent’s evaluation of the
general standard of living and the scale of the analysis, the higher the local
level of unemployment the lower the Conservative vote. Respondents apparently
took into account both the general (i.e., national) and their local situation
when determining how to vote—and the poorer the situation locally the greater
the likelihood of people voting to punish the government. To test that conclusion
more formally, we conducted logistic regression analyses with Conservative
vote as the dependent variable (coded 0 if the respondent did not vote Conservative
in 1997 and 1 if he/she did). The independent variables were the five
evaluation categories for the general standard of living (Table 1) and the local
unemployment rate. The former were entered as dummy variables, with those
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 129
saying that the general standard had increased a lot as the comparator; the
latter was the percentage unemployed in 1991 at the relevant spatial scale.
The results of these regressions are given in the first block of Table 4. In
the first regression (the left-hand column), only the economic evaluations are
included; the other seven include the unemployment rate at the scale indicated
in the column heading. All of the regression coefficients are statistically significant
at the 0.01 level or better, and have the expected signs. Thus, the worse
the evaluation and the higher the local unemployment rate, the smaller the
TABLE 4. Logistic Regressions of Voting Conservative by Evaluations
of Changes in the General Standard of Living and
Local Unemployment
Scale of Analysis
n500 n1000 n2500 n5000 n10000 Ward Const
Constant −1.45 −0.76 −0.71 −0.56 −0.47 −0.38 −0.62 −0.30
Change in general standard of living (comparator: increased a lot)
Increased a little −0.85 −0.90 −0.90 −0.93 −0.95 −0.95 −0.89 −0.93
Stayed the same −1.54 −1.54 −1.53 −1.56 −1.58 −1.60 −1.54 −1.59
Fallen a little −2.79 −2.77 −2.77 −2.80 −2.82 −2.83 −2.77 −2.82
Fallen a lot −4.12 −4.23 −4.22 −4.23 −4.26 −4.27 −4.10 −4.14
Unemployment rate −0.07 −0.08 −0.09 −0.10 −0.11 −0.09 −0.12
−2 log likelihood 2498 2360 2357 2357 2343 2343 2422 2413
Improvement 138 141 141 155 155 76 85
With 1992 vote added
Constant −1.53 −1.15 −1.12 −1.05 −1.02 −0.95 −1.08 −0.79
1992 vote (comparator: did not vote Conservative)
Voted Conservative
3.22 3.15 3.14 3.13 3.13 3.13 3.15 3.15
Change in general standard of living (comparator: increased a lot)
Increased a little −1.08 −1.11 −1.11 −1.11 −1.11 −1.12 −1.09 −1.12
Stayed the same −1.50 −1.51 −1.50 −1.52 −1.52 −1.53 −1.50 −1.54
Fallen a little −2.62 −2.64 −2.63 −2.63 −2.64 −2.64 −2.62 −2.64
Fallen a lot −3.67 −3.84 −3.83 −3.83 −3.85 −3.86 −3.68 −3.70
Unemployment rate −0.04 −0.04 −0.05 −0.05 −0.06 −0.05 −0.08
−2 log likelihood 1715 1655 1655 1653 1655 1654 1699 1692
Improvement 60 60 62 60 61 16 23
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors;
ward—electoral ward; const—Parliamentary constituency.
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER130
probability of a Conservative vote. Comparison of the results at the various
scales shows very little variation. As the scale of analysis increases, there is a
slight rise in the size of the coefficient for unemployment rate (from −0.07 for
n500 neighborhoods to −0.12 for constituencies), but model improvement
(compared to the initial model, which excludes local unemployment rate) is
less at the ward and constituency scales than the others. The implication is
that people are particularly responding to contextual variations at the small
scales represented by our bespoke neighborhoods.
The nature of the relationship with local context can be more fully appreciated
by using the equations in Table 4 to estimate the probability that a respondent
voted Conservative. For example, among individuals who considered that the
general standard of living had increased a lot, the odds that they would vote
Conservative were 0.45 if local unemployment at the n1000 scale was 1 percent
and 0.10 if it was 20 percent, giving probabilities of voting Conservative of
0.31 and 0.09 respectively. There was thus a substantial difference in the
likelihood of a person who thought that the national situation had improved
voting Conservative according to the local situation, with similar differences at
other spatial scales (at the n10000 scale, for example, the respective probabilities
were 0.38 and 0.07). This was not the case with those who thought that general
living standards had fallen a lot over the previous five years. At the n1000 scale,
the probability of such a respondent voting Conservative was 0.01 where local
unemployment was 1 percent and 0.00 where it was 20 percent. Similar differences
occur at all other spatial scales. Thus, a person who was positive about
the national situation was much less likely to vote for the incumbent Conservative
party if local employment was high than if it was low, whereas a person who
was negative about the national situation was very unlikely to vote Conservative
whatever the local situation. Where local circumstances were inconsistent with
respondents’ views of the general condition of the economy, the local context
created a cross-pressured situation, with a substantial number deciding to vote
against the incumbent government as a consequence. But, for respondents
who considered that general living standards had declined substantially, there
was virtually no support for the incumbents, whatever the local situation.
One possible criticism of the results in the top section of Table 4 is that
they relate to an underspecified model: the relationships shown may be spurious
because they reflect the operation of other, excluded variables. This could be
countered by incorporating a range of other independent variables, such as
social class, selected to reflect known influences on voting in England and
Wales. The process of class dealignment over the last four decades of the
twentieth century has seen the decline in importance of many of these variables,
however, with the implication that British voters are now much more responsive
to the economic context at individual elections (see Sanders, 1997). Economic
voting calculi now apparently dominate. Nevertheless, to incorporate other
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 131
major influences on voter choice in 1997 we have re-run the regressions incorporating
one further variable—whether or not the respondent reported voting
for the Conservative party in 1992.3
A substantial proportion of the British
electorate remains loyal to a party over successive elections—58 percent of
those who reported voting Conservative in 1992 in England and Wales did so
again in 1997 (Johnston et al., 1997). By holding constant the 1992 vote,
therefore, we are not only implicitly incorporating independent variables that
have a long-term influence on party choice, but also making our test of the
influence of the impact of economic evaluations and local context much more
stringent.4
Our expectations are that: (1) those who voted Conservative in 1992
were much more likely to vote Conservative again in 1997; (2) those who
thought that the general economic situation had declined were much less likely
to vote Conservative in 1997, holding the 1992 vote constant; and (3) the higher
the local unemployment rate, the lower the probability of a Conservative vote,
holding constant both of the previous variables.
The second block of regression results in Table 4 confirms these expectations
fully, with all regression coefficients statistically significant at the 0.01 level or
better. Continuity in voting was clearly a major factor in 1997. Those who
voted Conservative in 1992 were some 25 times more likely to vote for that
party again in 1997, according to the first regression, which excludes local
context. The regression coefficient of 3.22—indicating the log-odds of voting
Conservative among those who voted Conservative in 1992—has an exponent
of 25.03, indicating the relative likelihood of a person who voted Conservative
in 1992 voting Conservative in 1997 compared with one who did not vote
Conservative in 1992. Inclusion of local unemployment rate improves the
goodness of fit—though again less so at the ward and constituency scales than
at those of the bespoke neighborhoods. Local context was clearly related to
voters’ decisions whether to support the incumbent government in 1997.
Holding constant the 1992 vote changes the degree of the impact of local
context on vote in 1997, but not the general pattern discussed above. Thus,
for example, among respondents who did not vote Conservative in 1992 and
who thought that the general standard of living had increased a lot, at the
n1000 scale the probability of voting Conservative if local unemployment were
1 percent was 0.24, whereas if local unemployment were 20 percent it was
0.13; at the n10000 scale, the respective probabilities were 0.27 and 0.10. Local
context mattered when it was not in line with the respondent’s evaluation of
the national situation. Among those who thought that the general standard had
fallen a lot and who did not vote Conservative in 1992, however, the probability
of voting Conservative was virtually zero whatever the local situation. Among
those who voted Conservative in 1992, however, and who also thought that
the general standard of living had increased a lot, the probability of voting
Conservative in 1997 if unemployment was 1 percent at the n1000 scale was
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER132
0.88, whereas if unemployment was 20 percent the probability was 0.77; there
were very similar differences at all other scales. The impact of local conditions
on the probability of them voting again was less substantial, though significant,
than was the case with those who did not support the incumbent party at the
previous election.
The Question of Scale
The regressions reported in Table 4 show continuity in voting behavior
between 1992 and 1997—the government party got much more support at the
latter date from those who voted for it at the previous election. Retrospective
sociometric evaluations were also clearly linked to support for the government
party in 1997—the less satisfied voters were about the general economic situation,
the lower the probability that they would vote for that party. In addition,
local economic context was also related to voter choice—the higher the level of
unemployment locally, the lower the probability of a vote for the Conservatives
among those who were positive about the national economic situation.
But which scale was most important to voters—or was their response to
local unemployment scale-invariant, with the same reaction whether the scale
was that of their immediate neighborhood (the nearest 500 persons to their
home) or their Parliamentary constituency (which averaged some 70,000 voters)?
To answer that we have re-run the first block of regressions in Table 4
including local context at two scales instead of one, thus looking at very local
variations within a wider context.
Each of the first block of regressions in Table 5 includes the constituency
scale plus one other, smaller scale—from n500 to n10000. In every case, both
variables are statistically significant (at the 0.01 level, or better), suggesting
that voters were responding not only to the general level of unemployment in
their home area, defined broadly as their Parliamentary constituency, but also
to the level in their immediate local neighborhood. The probability of a Conservative
vote declined the higher the level of unemployment in their constituency
(equivalent in most cases to either their home town or to a major portion of
a large city for those resident in the country’s major urban areas), and declined
even more in pockets of high unemployment within that constituency. People
were responding to the local economic situation at both scales.
The second block of regressions in Table 5 involves a tighter packing of
scales. The first four include both the n10000 scale plus one of the four smaller
ones. There is some evidence that the probability of a Conservative vote was
even lower in pockets of high unemployment at the n500 and n2500 scales,
holding constant the probability at the larger (n10000) scale, but when the
scales are close—the n10000 and n5000—neither coefficient was significant
at the 0.05 level, indicative of collinearity between the two unemployment
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 133
TABLE 5. Logistic Regressions of Voting Conservative by Evaluations
of Changes in the General Standard of Living and
Local Unemployment Rate—Analyses at Two Spatial Scales
a. Constituency and smaller scales
Constant −1.45 −0.30 −0.30 −0.27 −0.24 −0.23
Change in general standard of living (comparator: increased a lot)
Increased a little −0.85 −0.93 −0.93 −0.95 −0.96 −0.95
Stayed the same −1.54 −1.56 −1.56 −1.57 −1.59 −1.59
Fallen a little −2.79 −2.79 −2.79 −2.81 −2.82 −2.82
Fallen a lot −4.12 −4.25 −4.24 −4.25 −4.27 −4.27
Unemployment rate: scale
Constituency −0.07 −0.07 −0.06 −0.05 −0.05
n10000 −0.08
n5000 −0.07
n2500 −0.06
n1000 −0.05
n500 −0.05
−2 log likelihood 2498 2338 2339 2333 2334 2338
Improvement 160 159 165 164 160
b. n10000/n5000 and smaller scales
Constant −1.45 −0.40 −0.40 −0.41 −0.40 −0.48 −0.48 −0.49
Change in general standard of living (comparator: increased a lot)
Increased a little −0.85 −0.94 −0.94 −0.94 −0.95 −0.94 −0.94 −0.94
Stayed the same −1.54 −1.58 −1.58 −1.58 −1.59 −1.58 −1.58 −1.58
Fallen a little −2.79 −2.81 −2.81 −2.81 −2.82 −2.81 −2.81 −2.81
Fallen a lot −4.12 −4.26 −4.25 −4.25 −4.27 −4.25 −4.25 −4.25
Unemployment rate: scale
n10000 −0.08 −0.08 −0.06 −0.06
n5000 −0.05 −0.08 −0.09 −0.06
n2500 −0.04 −0.04
n1000 −0.03 −0.01
n500 −0.03 −0.02
−2 log likelihood 2498 2339 2340 2339 2341 2341 2342 2341
Improvement 159 158 159 157 157 156 157
Underlined coefficients are not significant at the 0.05 level.
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors;
ward—electoral ward; const—Parliamentary constituency.
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER134
patterns. With the n5000 scale held constant (in all of the final three regressions)
there is no evidence of smaller scale variations.
Again, estimating from the coefficients the probability of an individual voting
Conservative in certain circumstances shows that the main impact of local
effects was on those who were positive about the national situation, but lived
in areas of relatively high unemployment. For example, using the first block
of regressions in Table 5, among those who thought that the general standard
of living had increased a lot and who lived in constituencies with very low
unemployment (just 1 percent), the probability of voting Conservative at the
n1000 scale was 0.40 if unemployment in their immediate neighborhood was
1 percent and 0.20 if it was 20 percent. The worse the very local situation, the
smaller the likelihood of the respondent voting Conservative—even though at
the wider, constituency, scale the situation was very good. Similarly, using the
second block of regressions in Table 5, for a respondent who thought that the
general situation had increased a lot and unemployment at the n10000 scale
was just 1 percent, the probability of voting Conservative if unemployment at
the n1000 scale was also 1 percent was 0.38, whereas it was 0.25 if it was 20
percent. There were no differences in the probabilities of voting Conservative
according to the local situation among those who thought that general living
standards had decreased a lot, however; they were virtually zero at all scales.
In responding to their local context, therefore, it seems that British voters
responded to both the general situation in their home area, as represented by
the level of unemployment in their Parliamentary constituency, and also to the
situation in their immediate milieux—represented by the bespoke neighborhoods
used here, relating to the areas containing the nearest 500–10,000 persons
to the respondents’ homes. Those who thought that the national situation
was very good were less likely to vote Conservative if local unemployment were
high, than if they lived in areas where it was low. The less rosy their picture
of the national economy, however, the smaller the differences in their voting
behavior according to the local context, with virtually no differences for those
who thought the situation had deteriorated substantially and did not support
the Conservatives at the previous general election.
REWARD AND PUNISH
Table 2 shows that respondents varied in their support for the Conservative
party not only according to their economic evaluations, but also whether they
attributed credit/blame to government policies for the perceived situation. In
this section we take both of those into account, exploring whether the probability
of voting for the government party was related not only to the general economic
situation, but also to the respondents’ linking of that situation with government
policy. There was a nearly tenfold difference in propensity to vote Conservative
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 135
between those who thought the general standard of living had increased and
attributed that to government policy (66.3 percent of those people: the ‘rewarders’)
and those who thought it had fallen for the same reason (7.1 percent: the
‘punishers’).
Table 6 shows the percentages voting Conservative in these categories at
the various spatial scales. (There were too few observations in a number of
cells for those who thought that the general standard of living had fallen for
‘other reasons,’ and so that block of data has been excluded.) With very few
slight exceptions, the higher the local level of unemployment, at every scale,
the lower the support for the incumbent Conservative party—whatever the
respondents’ evaluations of changes in the general economic situation and
their causes. In addition, the gap between areas in the support given to the
Conservative party is greatest where the general standard of living is perceived
to have fallen because of government policy, and least where it is thought to
have increased: electoral punishment was greatest where the national situation
was pessimistically evaluated and the local context was worst.
These conclusions based on the cross-tabulations were formally tested
through logistic regressions, again holding constant reported vote in 1992.
The results in Table 7 entirely confirm the expectations. Those who voted
Conservative in 1992 were some 24 times more likely to vote Conservative in
1997 than were those who did not support the government party at the first
of those elections.5
With regard to the economic evaluations, compared to
those who said that the general standard of living had improved because of
government policies, all other groups were less likely to vote Conservative in
1997. The biggest difference was between the comparator group and those
who thought that the general standard had fallen because of government policies.
The probability of a Conservative vote among the latter group was only
0.11 of that for the former, comparator group, indicating a tenfold difference
in propensity to vote Conservative according to whether the government was
credited with economic prosperity or blamed for failure. Finally, holding constant
1992 vote and evaluations of changes in the general standard of living,
there was a significant relationship (at the 0.01 level) between probability of
voting Conservative and local unemployment rate: the higher the rate, the
smaller the probability.
Interpolation of the situation in particular contexts for certain voter types
again illustrates the extent of the influence of local context. For example, the
probability of a Conservative vote in 1997 for a respondent who did not vote
Conservative in 1992, and who thought that the general standard of living had
increased a lot because of government policy since then, was 0.24 if he or she
lived in an n1000 area with an unemployment rate of 1 percent and 0.13 if
the unemployment rate there was 20 percent. As before, the probability of
voting Conservative among those who evaluated the national situation positively
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER136
TABLE 6. Voting Conservative by Evaluations of Changes in the General
Standard of Living, Attribution of the Reasons for Changes,
and Local Context (percentage in each cell voting Conservative)
Unemployment
level n500 n1000 n2500 n5000 n10000 Ward Const
General standard of living increased because of government policies
0.0–2.9 * * * * * 55.6 *
3.0–5.9 58.6 57.9 61.9 55.8 55.6 62.2 60.2
6.0–8.9 51.7 48.9 49.4 58.2 57.2 50.3 54.8
9.0–11.9 49.3 52.4 41.1 40.4 42.7 42.4 47.3
12.0–14.9 43.2 50.0 40.2 37.7 38.5 40.0 45.6
15.0< 33.8 32.2 34.2 35.6 37.5 39.2 30.0
General standard of living increased for other reasons
0.0–2.9 * * * * * * *
3.0–5.9 28.2 29.9 36.5 35.4 29.6 33.3 35.4
6.0–8.9 30.3 30.0 25.0 25.0 31.8 27.1 24.8
9.0–11.9 20.8 12.9 22.4 23.2 17.9 19.4 33.3
12.0–14.9 18.8 29.2 22.2 18.5 28.6 23.8 12.8
15.0< 16.7 1.58 16.1 18.8 16.2 1.82 13.9
General standard of living stayed the same because of government policies
0.0–2.9 * * * * * * *
3.0–5.9 38.7 41.2 46.4 47.5 47.7 45.1 41.3
6.0–8.9 33.3 30.8 31.4 33.6 35.3 34.1 34.0
9.0–11.9 20.8 22.5 18.5 15.3 14.6 12.8 24.8
12.0–14.9 9.1 11.2 13.3 17.1 7.7 11.4 8.6
15.0< 10.5 10.8 7.4 11.4 9.3 11.2 9.8
General standard of living stayed the same for other reasons
0.0–2.9 * * * * * * *
3.0–5.9 29.3 29.5 31.3 32.7 30.0 29.2 37.5
6.0–8.9 21.7 24.6 22.8 26.2 25.6 24.0 23.8
9.0–11.9 22.0 17.0 22.0 12.7 15.4 21.9 17.9
12.0–14.9 14.3 14.5 4.2 15.2 16.0 11.8 12.8
15.0< 11.1 11.1 14.0 8.3 10.9 11.7 10.8
General standard of living fallen because of government policies
0.0–2.9 * * * * * * *
3.0–5.9 9.0 12.2 13.0 13.7 12.8 12.2 17.8
6.0–8.9 8.0 6.3 5.6 6.7 7.6 5.6 6.0
9.0–11.9 3.5 4.7 4.3 3.2 2.4 2.4 3.1
12.0–14.9 2.3 0.0 2.8 1.8 3.0 2.8 2.7
15.0< 1.0 1.5 1.7 2.5 2.6 3.5 2.6
*less than 20 respondents in the cell.
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors;
ward—electoral ward; const—Parliamentary constituency.
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 137
TABLE 7. Logistic Regressions of Voting Conservative by Evaluations of
Changes in the General Standard of Living, Attribution of Reasons
for Changes and Local Unemployment Rate
Scale of Analysis
n500 n1000 n2500 n5000 n10000 Ward Const
Constant −1.51 −1.14 −1.12 −1.03 −1.01 −0.97 −1.07 −0.79
1992 vote (comparator: did not vote Conservative)
Voted Conservative
1992 3.18 3.11 3.11 3.10 3.09 3.10 3.11 3.10
Change in general standard of living and reason (comparator: increased because of
government policy)
Increased:other −0.73 −0.75 −0.74 −0.73 −0.72 −0.72 −0.72 −0.75
Same:government −0.64 −0.63 −0.62 −0.63 −0.64 −0.64 −0.64 −0.65
Same:other −1.01 −0.98 −0.97 −0.98 −1.00 −0.98 −0.98 −1.00
Fallen:government −2.20 −2.21 −2.20 −2.20 −2.19 −2.19 −2.19 −2.19
Fallen:other −1.52 −1.53 −1.54 −1.53 −1.54 −1.53 −1.50 −1.51
Unemployment rate −0.04 −0.04 −0.05 −0.05 −0.06 −0.05 −0.08
−2 log likelihood 1651 1597 1597 1594 1596 1596 1638 1630
Improvement 54 54 57 55 55 13 21
Underlined coefficients are not significant at the 0.05 level
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors;
ward—electoral ward; const—Parliamentary constituency.
at one of the extremes according to local unemployment (the highest), was
about half that in the areas of lowest unemployment—a difference that held
at all spatial scales. Similar relative differences also occurred among those who
thought that the general living standard had decreased a lot—though at much
lower absolute levels. At the n1000 scale, the probability of such a respondent
(and who also did not vote Conservative in 1992) voting Conservative in 1997
was 0.06 with a local unemployment rate of 1 percent and 0.03 where it was
20 percent.
ACCOUNTING FOR THE LOCAL CONTEXTUAL EFFECTS:
LOCAL SOCIOMETRIC EVALUATIONS
We have provided strong circumstantial evidence regarding the operation
of local contextual effects in the economic voting calculus employed by the
electorate in England and Wales at the 1997 general election there. Those
who were positive about changes in the general standard of living over the
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER138
five-year lifetime of the previous government, and who credited government
performance with that situation, were most likely to vote for the governing
party’s return to office—but they were more likely to do that in local areas
with low unemployment than they were where it was higher. The implication is
that their propensity to reward the government for delivering general economic
prosperity was tempered by evidence that such prosperity was not being experienced
in their local area.
We can go further than that inference, however, because further questions
used in the 1997 BES enable us to tap voters’ evaluations of their local economic
situation. Two questions similar to those relating to the national situation were
asked.6
Local area change
Compared with other parts of Britain since the last general election in April 1992,
would you say that (this part of Britain/Scotland/Wales) has been getting more
prosperous than average, stayed about average, or been getting less prosperous than
average?
(The show card used to structure responses divided the ‘more prosperous’ and ‘less
prosperous’ categories into ‘a lot more’ and ‘a little more’.)
Why
Do you think this is the result of government policies or for some other reason?
Combining the responses to these two questions gives voting outcomes very
similar to those in Table 2 for the national situation. Those who thought their
area had become relatively more prosperous, and credited that to government
policies, were most likely to vote for the government party (43.2 percent),
whereas those who though their area had become relatively less prosperous,
and blamed that on government policies, were least likely (9.6 percent) to vote
for the incumbent party.7
So, does this second pattern of economic voting (which we term regional
sociometric voting) provide an account for the contextual effects set out in the
tables discussed above? Does it remove the impact of local unemployment rate
on the probability of a Conservative vote? To answer these we have re-run the
logistic regressions reported in Table 7, adding a further set of independent
variables relating to evaluations of changes in area prosperity.
The first regression in the left-hand column of Table 8 excludes the local
unemployment variables. The coefficients for the first two blocks show the
same, highly significant, relationships for 1992 vote and evaluations of changes
in the general standard of living as in Table 7. The new set of variables includes
two that are statistically significant, and adds substantially to the model’s perfor-
mance.8
The significant coefficients are for ‘local area economic situation was
better because of other reasons’ and ‘local area situation was worse because
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 139
TABLE 8. Logistic Regressions of Voting Conservative by Evaluations of
Changes in the General Standard of Living and Local Area
Prosperity, Attribution of Reasons for Changes and
Local Unemployment Rate
Scale of Analysis
n500 n1000 n2500 n5000 n10000 Ward Const
Constant −1.51 −1.10 −1.07 −0.98 −0.96 −0.94 −1.08 −0.69
1992 vote (comparator: did not vote Conservative)
Voted Conservative
1992 3.18 3.10 3.10 3.08 3.08 3.09 3.12 3.11
Change in general standard of living and reason (comparator: increased because of
government policy)
Increased:other −0.57 −0.58 −0.57 −0.56 −0.56 −0.55 −0.55 −0.57
Same:government −0.59 −0.58 −0.57 −0.59 −0.60 −0.60 −0.58 −0.60
Same:other −1.01 −1.01 −1.00 −1.02 −1.01 −1.02 −0.99 −1.01
Fallen:government −2.16 −2.20 −2.19 −2.19 −2.19 −2.18 −2.16 −2.18
Fallen:other −1.44 −1.39 −1.39 −1.39 −1.39 −1.39 −1.41 −1.41
Change in area prosperity and reason (comparator: better because of government policy)
Better:other −0.48 −0.54 −0.54 −0.55 −0.55 −0.54 −0.52 −0.60
Same:government −0.10 −0.13 −0.11 −0.10 −0.08 −0.10 −0.09 −0.14
Same:other −0.14 −0.19 −0.19 −0.18 −0.18 −0.18 −0.18 −0.24
Worse:government −0.65 −0.63 −0.62 −0.60 −0.60 −0.60 −0.62 −0.61
Worse:other −0.36 −0.41 −0.41 −0.40 −0.40 −0.39 −0.38 −0.42
Unemployment rate −0.04 −0.04 −0.05 −0.06 −0.06 −0.04 −0.09
−2 log likelihood 1489 1443 1442 1439 1441 1443 1477 1466
Improvement 46 47 50 48 46 12 23
Underlined coefficients are not significant at the 0.05 level.
Key to local areas: n500—nearest 500 neighbors; n1000—nearest 1000 neighbors; n2500—
nearest 2500 neighbors; n5000—nearest 5000 neighbors; n10000—nearest 10000 neighbors; ward
electoral ward; const—Parliamentary constituency.
of government policy’—in each case relative to the comparator, those who
thought the ‘local situation was better because of government policy.’ The first
of these shows that among those who thought that the local situation was better,
those who attributed this to other reasons were only about 0.6 as likely to vote
Conservative as those who attributed the improvement to government policy.9
The ‘feel-good factor’ had the most impact on support for the government
among those who ascribed their ‘feeling-good’ to the government’s actions.
The second significant relationship shows that those who thought that their
local situation had become worse because of government policies were only
about half as likely to vote Conservative as those who thought government
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER140
policies had produced an improvement in the local situation (holding constant
national sociometric evaluations).10
These results suggest the operation of a regional as well as a national sociometric
evaluation process in the voters’ 1997 calculi. Their joint operation, however,
did not eliminate the significant relationships with local unemployment rates,
whose strength was as great as in the regressions reported in Table 7. Introducing
the local evaluations did not remove the general finding identified throughout
the analyses reported here that the higher the local unemployment rate
the smaller the probability of a vote for the incumbent government. Whatever
their national and local sociometric evaluations, people living in areas of high
unemployment were less likely to vote for the Conservatives in 1997 than were
those living in areas of low unemployment.
Interpolation of voting probabilities using the regression coefficients produces
similar differences to those discussed above. For example, at the n1000
scale, the probability of voting Conservative in 1997 among those who (a) did
not vote Conservative in 1992, (b) thought the government responsible for an
increase in the general standard of living, and (c) thought that their local
economy was more prosperous because of government policy was 0.25 if local
unemployment was 1 percent and 0.13 if it was 20 percent. Among those with
similar responses except that they did vote Conservative in 1992, the respective
probabilities were 0.88 and 0.77. And among those who (a) did not vote Conservative
in 1992, (b) thought that the national standard of living had decreased
for other reasons, and (c) thought that their local was less prosperous for other
reasons, the probabilities were 0.02 and 0.01. Variations in the local level of
unemployment had most impact on non-Conservative voters in 1992 who
thought that government policies had been successful since then, both nationally
and locally—i.e. where the cross-pressures were greatest—and least on those
who did not vote Conservative in 1992 and thought the economic situations
had deteriorated since the election.
CONCLUSIONS
Books and Prysby (1999, p. 11) concluded from their study of one U.S.
election that ‘economic context does affect retrospective economic evaluations.’
Our study of the 1997 general election in England and Wales not only comes
to the same conclusion, but shows very clearly that local economic context
was very influential on how people voted in two ways—as reflected by their
evaluations of regional prosperity and by the unemployment level in their local
area. The government was most likely to be rewarded by people living in areas
of low unemployment who thought that government policies had delivered
both national and local prosperity. Those who thought that the policies had
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 141
delivered the former but not the latter were much less likely to vote for the
government’s return to office, especially in areas of high unemployment.
Books and Prysby began their paper with the assertion that ‘Individuals are
influenced by their context.’ There is considerable dispute among students of
elections in the UK regarding the veracity of that assertion there (as exemplified
by the debate between McAllister and Studlar, 1992, Johnston and Pattie, 1998,
and Dunleavy’s, 1979, general critique). This article, based on data sets never
previously available, comes down firmly on the side of those who believe in
the assertion, and moves it toward the status of a verified hypothesis. It has
provided strong evidence of spatial variations in economic voting in England
and Wales at the 1997 general election: the higher the level of unemployment
in an area, the lower the support for the incumbent government, holding
constant the voters’ evaluations of recent changes in the general standard of
living, whether they attributed those changes to government policies, and also
their evaluation of the state of their local economy. Places matter when voters
are deciding whether to support the incumbent government. They take cues
from their local context and act accordingly.
Acknowledgments. The census data used here were obtained from the Manchester
University Computer Centre (@ Crown Copyright); the British Election Study files
were obtained from the Data Archive at the University of Essex; and linking the two
files was made possible by cooperation with the CREST team who conducted the
survey. We are very grateful to John Curtice and others in that team for their time and
considerable assistance. The work was funded by a grant (R000222649) from the ESRC,
which supported Iain MacAllister and Helena Tunstall for six months during 1998. We
are also grateful to three anonymous referees for very valuable constructive critiques
of earlier versions of this article, and for the editor’s sympathetic consideration.
NOTES
1. These data sets were compiled and the analyses reported here undertaken with a grant from
the UK Economic and Social Research Council (R000222649), whose support is gratefully
acknowledged. We are also grateful to John Curtice and other members of the CREST team
who conducted the 1997 British Election Study for their generous assistance with the work.
2. The respondents in Scotland have been excluded from this study because of difficulties creating
the contextual data sets discussed below.
3. We explored the use of a further explanatory variable—whether the respondent was currently
unemployed. The number was small, however (some 6 percent of the sample), and the variable
was not significant in any of the equations. We also explored whether there were candidate
effects—whether incumbents performed better than non-incumbents, whether ‘Eurosceptics’
performed better than ‘Europhiles,’ whether candidates against whom there had been allegations
of ‘sleaze’ during the previous Parliament performed less well than those against whom
there were no allegations, and so on. Like others who have sought such effects (e.g., Farrell
et al., 1998), we found no evidence of such effects.
4. In the terms of the classic ‘Michigan model’ (e.g., Miller and Shanks, 1996), therefore, we
JOHNSTON, PATTIE, DORLING, MACALLISTER, TUNSTALL, AND ROSSITER142
are incorporating all of the non-proximate causes of the voting decision into the single variable
relating to vote in 1992 and focusing our investigations on the proximate causes.
5. The exponent of the coefficient of 3.18 is 24.05.
6. There are some slight potential difficulties with using these questions since the term ‘this part
of Britain’ is imprecise. It could have been interpreted by respondents as meaning their home
region, town, or even neighborhood.
7. A copy of the full table can be supplied on request to the first-named author.
8. The −2 log likelihood figure is 1489, compared to 1651 for that reported in the left-hand
column of Table 7.
9. The coefficients for that variable range from −0.48 to −0.60; the exponent of −0.54 (the median
value) is 0.58
10. The median value for those coefficients across all spatial scales is –0.62, which has an exponent
of 0.54
REFERENCES
Atkinson, L. R., and Partin, R. W. (1995). Economic and referenda voting: a comparison
of gubernatorial and senatorial elections. American Political Science Review 89: 99–
107.
Books, J., and Prysby, C. (1999). Contextual effects on retrospective economic evaluations:
the impact of the state and local economy. Political Behavior 21: 1–16.
Dunleavy, P. (1979). The urban basis of political alignment: social class, domestic
property ownership and state intervention in consumption processes. British Journal
of Political Science 9: 409–443.
Evans, G., Curtice, J., and Norris, P. (1998). New Labour, new tactical voting? The
causes and consequences of tactical voting in the 1997 general election. In D. Denver,
J. Fisher, P. Cowley, and C. Pattie (eds.) British Elections and Parties Review 8: The
1997 General Election, pp. 65–79. London: Frank Cass.
Farrell, D. M., McAllister, I., and Studlar, D. T. (1998). Sex, money and politics: sleaze
and the Conservative party in the 1997 election. In D. Denver, J. Fisher, P. Cowley,
and C. Pattie (eds.), British Elections and Parties Review 8: The 1997 General Election,
pp. 80–95. London: Frank Cass.
Harrop, M., Heath, A., and Openshaw, S. (1991). Does neighbourhood influence voting—and
why? In I. Crewe, P. Norris, D. Denver, and D. Broughton (eds.), British
Elections and Parties Yearbook 1991, pp. 103–122. Hemel Hempstead: Harvester
Wheatsheaf.
Johnston, R. J., and Pattie, C. J. (1998). Composition and context: region and voting
in Britain revisited during Labour’s 1990s’ revival. Geoforum 29: 309–329.
Johnston, R. J., Pattie, C. J. and Allsopp, J. G. (1988). A Nation Dividing? Britain’s
Changing Electoral Map 1979–87. London: Longman.
Johnston, R. J. et al. (1997) Spatial variations in voter choice: modelling tactical voting
at the 1997 general election in Great Britain. Geographical and Environmental Modeling
1: 153–179.
Johnston, R. J. et al. (1998). New Labour landslide—same old electoral geography? In
D. Denver, J. Fisher, P. Cowley, and C. Pattie (eds.), British Elections and Parties
Review 8: The 1997 General Election, pp. 35–64. London: Frank Cass.
Lewis, J., and Townsend, A. R., eds. (1989). The North-South Divide: Regional Change
in Britain in the 1980s. London: Paul Chapman Publishers.
LOCAL CONTEXT, ECONOMIC EVALUATIONS, AND VOTING 143
McAllister, I., and Studlar, D. T. (1992). Region and voting in Britain: territorial polarization
or artifact? American Journal of Political Science 36: 168–199.
Miller, W. E., and Shanks, J. M. (1996). The New American Voter. Cambridge MA:
Harvard University Press.
Miller, W. L. (1978). Social class and party choice in England: a new analysis. British
Journal of Political Science 8: 257–284.
Pattie, C. J., Dorling, D. F. L. and Johnston, R. J. (1997). The electoral geography of
recession: local economic conditions, public perceptions and the economic vote in
the 1992 British general election. Transactions, Institute of British Geographers NS22:
147–161.
Pattie, C. J., and Johnston, R. J. (1995). ‘It’s not like that round here’: region, economic
evaluations and voting at the 1992 British general election. European Journal of
Political Research 28(1): 1–32.
Sanders, D. (1997). The new electoral battleground. In A. King (ed.), New Labour
Triumphs: Britain at the Polls pp. 209–248. Chatham NJ: Chatham House.