The measurement of left-right and libertarian-authoritarian values: A ... Working class conservatives and middle class socialists, in R. Jowell, ... 'Labour Party Factionalism and Extremism', in A. F. Heath, R. Jowell and J. Curtice eds.Labour's ...
Quality & Quantity 29: 191-206, 1995. 9 1995 KIuwer Academic Publishers. Printed in the Netherlands.
191
The measurement of left-right and libertarian-authoritarian values: a comparison of balanced and unbalanced scales GEOFFREY A. EVANS & ANTHONY F. H E A T H Centre for European Studies, Nuffield College, Oxford, OX I 1NF, Great Britain
Abstract. In this paper we compare the performance of balanced and unbalanced Likert scales
of two core dimensions of political attitudes: left-right and libertarian-authoritarian values. The balanced scales control for the effects of acquiescence response sets, whereas the unbalanced scales do not. Using data from two panel surveys, balanced and unbalanced scales are compared for reliability, stability and validity both with each other, and with other measures of political ideology and values (left-right self-placement and postmaterialism). Both balanced and unbalanced versions of the left-right and libertarian-authoritarian Likert scales are found to be more stable and strongly associated with social characteristics than are the other measures. The unbalanced scales have slightly higher reliability than the balanced scales; they are also orthogonal, whereas balanced left-right and libertarian-authoritarian scales are moderately correlated. Unbalanced scales also display a slightly stronger relationship with social characteristics, but they do not differ in their pattern of association with political preferences. Differences between balanced and unbalanced scales are attributed to the effects of acquiescence bias. This needs to be considered when using the scales for substantive analyses, but its effects are not problematic.
Introduction The core political beliefs and values in British society are of two types: the first involves principles of socialism versus laissez faire and is often referred to as a 'left-right' dimension; the second involves libertarianism versus authoritarianism and is often labelled a libertarian-authoritarian dimension. 1 The socialist/laissez faire domain is concerned with collectivism and government control, as well as economic and political equality. The libertarian/authoritarian domain concerns freedom of thought and conscience, freedom of association, and freedom to pursue one's own course of life. In short, the former can be interpreted as concerned with equality and the latter with personal freedom (cf. Rokeach, 1973). The central objective of our recent methodological research has been to develop and evaluate reliable and valid multiple item Likert scales of these core political values (see, for example, Heath et al., 1994; Evans et al., 1995). 2 As a result, batteries of items designed to measure left-right End libertarian-authoritarian values have for several years formed sections of the British Social Attitude Surveys (i.e. Jowell et al., 1988, 1990, 1991, 1992), the Northern Irish Attitude Surveys (i.e. Stringer and Robinson, 1991, 1992,
192
Geoffrey A. Evans and Anthony F. Heath
1993), the British Election Studies (Heath et al. 1991, 1994), and a recent elite survey (Norris, 1994). In addition, they form the basis of comparative measures of core political values in recent surveys conducted in Eastern Europe (see Whitefield and Evans, 1994). They thus provide a resource for both cross-sectional and time series analyses of core political beliefs and their relation to political behaviour and social structure. Such multiple item indicators of political values are usually to be preferred to single item measures (Nunnally, 1978). 3 Nevertheless, all attitude measurement assumes a shared understanding of the meaning of items, or at least random error but not bias in responses. 4 Consequently the systematic error associated with the presence of response sets is a major threat to validity. In particular, acquiescence response bias is likely to invalidate scales which do not control for direction of question wording. Acquiescence response bias is the well established finding in methodological research on surveys that some respondents have a tendency to agree with whatever statement is presented to them (see Bass, 1956; Schttman and Presser, 1981). Moreover, some types of people (such as the less-well educated) are more prone to acquiescence than others. This has been attributed to the existence of a response set derived from a norm of agreement amongst people that makes the expression of disagreement with a proposition a more demanding task requiring fairly clear views on the issues in question. The normal way in which this problem is addressed in Liken scales is by using balanced sets of questions that present issues to respondents in both positive and negative directions. Ideally, for example, support for a free market economy rather than one which is state regulated should be assessed with a combination of items: some of which ask if respondents agree with the free market; others which ask if respondents agree with opposition to the market; others which enquire whether respondents agree with state regulation of the economy; and yet others which ask if respondents disagree with such regulation. In order for respondents to be understood as having clear preferences one way or another on an issue it is important that they adopt a position consistently, regardless of the direction of the question wording. Usually the advantages of scales which contain a balance between statements worded in one direction on an issue and those worded in the opposite direction is simply asserted (see for example, our own work: Heath et al., 1994). In this paper however we compare empirically the reliability and validity of balanced and unbalanced versions of left-right and libertarianauthoritarian political values scales. We thus can estimate the likely consequences of using unbalanced scales when measuring core political values and by extension, similar types of socio-political attitudes. Such estimates are
Left-right and libertarian-authoritarian values
193
necessary for research which uses publicly available survey data containing unbalanced scales (see Evans et al., 1995).
Method
The reliability of the scales is assessed by examining their internal consistency using Cronbach's alpha. This is an estimate of reliability derived from the average inter-item correlation, which is monotonically related to the number of items in a scale. It is an estimate of the correlation of a test with an alternative form (with equivalent i.e. parallel items) containing the same number of items (or a hypothetical alternative form - Nunnally, 1978). A problem with Cronbach's alpha is that it may fail to identify a multi-dimensional structure (see Cortina, 1993). So we also use factor analysis to estimate the minimum structure to which the covariance matrix can be reduced. The second approach to reliability assessment is to examine the stability of responses through a test-retest study a few months, after the original. The time gap between the two surveys is designed to remove any memory effects, while limiting the possibility of major political events or long-term value changes adversely affecting stability estimates. Primarily, this involves examining correlations between responses to the scales at time tl and t2.5
Validation The effects of bias are also assessed by examining the relations between the scales and other social and political characteristics. This is a form of construct validation - a method that involves assessing whether a measure of a concept is associated with other variables in theoretically prescribed ways. 6 Of course, tests of construct validity are dependent upon the presence of well-established theories about the relations between the construct being tested and the variables used to assess its validity. Fortunately, in the case of political values this is unlikely to be a problem.
Data
The data come from a panel study designed to investigate a number of methodological problems. The panel members were drawn from the 1989 British Social Attitudes Survey, a nationally representative annual survey of the adult population of Great Britain. The 1989 BSAS was a stratified probability sample drawn from the electoral register (which was used to
194
Geoffrey A. Evans and Anthony F. Heath
provide a sample of addresses) and clustered within 152 sampling areas. It had a response rate of 69%, giving an achieved sample size of 3029. 7 1150 of these respondents were selected to form the basis for the panel study. 8 The first round of panel interviews was conducted between November 1989 and January 1990, and interviews were completed with 839 respondents. The second round was conducted between May and July 1990, and interviews were completed with 629 people. The items designed to tap core beliefs and values were administered in two short booklets which the respondents were asked to fill in themselves during the course of the interview. Two versions (Y and Z) were used. In order to check on the stability of responses to the items over time the respondents were given the same version in each round. 9 This analysis just uses the responses to version Y. In the first round 408 respondents answered version Y. In the second round there were 312 respondents.
Constructing the scales The items in the scales were selected to cover the main theoretical components of the two domains. In the case of the left-right scale we designed items to tap collectivism and individualism, government intervention and free enterprise, and economic and political equality. In the case of the libertarian-authoritarian scale we designed items to tap three main concepts - freedom of thought and conscience, freedom of association, and freedom to pursue one's own course of life. The items were asked in an agree/disagree format with five response categories (strongly agree, agree, neither agree nor disagree, disagree, strongly disagree). Respondents were also given a "don't know" option. 1~ Since our aim was to design scales that could be used over a period of years, the items did not address topical policy issues. We preferred items which could be asked in future studies when the political agenda might have changed. Given their more topical content, few existing questions from the British Election Surveys were suitable, but some items were taken from previous British Social Attitude Surveys. Most items were designed specifically for the study (or for a previous one on the same topic). The items for the scales were derived from larger batteries using factor analysis. For the left-right scale six items were selected which were balanced for question direction and covered the three conceptual components of equality, collectivism, and government intervention, la The items were as follows: Ordinary people get their fair share of the nation's wealth.* There is one law for the rich and one for the poor.
Left-right and libertarian-authoritarian values
195
There is no need for strong trade unions to protect employees' working conditions and wages.* It is government's responsibility to provide a job for everyone who wants one. Private enterprise is the best way to solve Britain's economic problems.* Major public services and industries ought to be in state ownership. These six items were summated to form a balanced scale. An unbalanced version of the scale was constructed by replacing the three items worded in a right-wing direction (marked with an asterisk) with ones that were worded in a left-wing direction. The unbalanced scale was thus composed of the following items: Government should redistribute income from the better off to those who are less well off. There is one law for the rich and one for the poor. Big business benefits owners at the expense of workers. It is government's responsibility to provide a job for everyone who wants one.
Management will always try to get the better of employees if it gets the chance. Major public services and industries ought to be in state ownership. Since we are limited to the items included in the original questionnaire, the three additional left-wing items are not exact opposites to the three right-wing items which they replace, but we believe that they tap the same ideological dimension. The correlation between the balanced and unbalanced scales was: 0.82 in round 1; and 0.83 in round 2. We then carried out analogous analyses with the libertarian-authoritarian items. The following six items were selected to form a balanced scale: Young people today don't have enough respect for traditional British values. Censorship of films and magazines is necessary to uphold moral standards. People in Britain should be more tolerant of those who lead unconventional lives.* Homosexual relations are always wrong. People should be allowed to organize public meetings to protest against the government.* Even political parties which wish to overthrow democracy should not be banned.* An unbalanced scale was constructed by replacing the three items worded
196
Geoffrey A. Evans and Anthony F. Heath
Table 1. Alphas and test-retest correlations
Alphas
Balanced left-right scale Unbalanced left-right scale Balanced libertarian scale Unbalanced libertarian scale Abstract self-placement scale Short postrnaterialism index Long postmatedalism index
Test-retest
Study 1
Study 2
correlation
0.64 0.77 0.53 0.67
0.67 0.76 0.52 0.66
0.70 0.76 0.72 0.76 0.58 0.41 0.63
in a libertarian direction (marked with an asterisk) with ones that were worded in an authoritarian direction. The unbalanced scale was thus composed of the following items: Young people today don't have enough respect for traditional British values. Children should be taught to obey authority. The law should always be obeyed even if a particular law is wrong. People who break the law should be given stiffer sentences. Censorship of films and magazines is necessary to uphold moral standards. Homosexual relations are always wrong. The correlation between the balanced and unbalanced scales was: 0.73 in round 1; and 0.74 in round 2. The dimensionality of the scales was checked using factor (principal components) analysis. For both the balanced scales two main factors accounted for most of the explained variance. We have treated these two factors as largely capturing question-direction effects and measuring the same underlying substantive dimension. 12 Thus items whose wording direction differed in the two scales differed in the factors on which they loaded. This interpretation is supported by the finding that for both of the unbalanced scales there was only one prominent factor.
Comparing the reliability of the balanced and unbalanced scales The first step is to compare the internal consistency of the scales. Table 1 presents the alphas in both rounds for both types of left-right and libertarianauthoritarian scales. It can be seen that the alpha of the balanced left-right scale is acceptable for a six-item indicator, whereas the reliability of the balanced libertarian-authoritarian scale is relatively low. This may be because the frequency distributions for the individual items are more skewed than is
Left-right and libertarian-authoritarian values
197
the case with the left-right items, thus reducing the item-total correlationsJ 3 Both of the unbalanced scales have higher levels of internal consistency than the respective balanced scales. Table 1 also presents the stability between rounds 1 and 2 of the panel of the scales. To provide yardsticks for comparison, we have also calculated the corresponding correlations for two frequently-used alternative measures of core political beliefs and values: the left-fight self-placement scale (i.e., Barnes and Kaase, 1979); 14 and Inglehart's (1977) measures of postmatefialist values. Again we find that the unbalanced scales have somewhat higher reliability coefficients than the balanced scales, but these differences are of small magnitude. More noticeably, the findings are in general quite comforting for both versions of the Likert scales, which attain considerably higher levels of stability than do the other measures. 15 The left-fight self-placement scale and the long version of the Inglehart measure have similar - and quite reasonable - over time correlations, but they are lower than any of the Likert scales. The short version of the Inglehart measure, which asks respondents to rank only four items, has very low over time stability. Finally, we examined the correlations between the two types of political values. The balanced libertarian-authoritafian and left-fight scales correlate moderately (R = 0.24; P < 0.01), whereas the unbalanced scales are uncorrelated (R = - 0 . 0 6 ; NS). This is a significant and substantively important difference.
Comparisons among different educational groups As we have mentioned earlier, acquiescence response bias is usually thought to be more prevalent among the less well-educated, who appear to have generally less well-formed political belief systems (Bishop, 1976). Any effects related to acquiescence could therefore be expected to be more problematic among such respondents. To examine this possibility we compared the reliability and stability over time of the different scales among respondents with more and less education respectively. 16 The results are shown in Table 2. As can be seen, even among the less highly educated group the alphas for both balanced and unbalanced scales hold up rather well. With the balanced left-fight scale alpha is the same for both educational groups. The internal consistency of the balanced libertarianauthoritarian scale declines substantially among less well-educated respondents in round one, but not in round two. Both the unbalanced scales had more or less the same level of internal consistency in the two educational groups.
198
Geoffrey A. Evans and Anthony F. Heath
Table 2. Alphas and test-retest correlations for educational groups
Alphas
Balanced LR Unbalanced LR Balanced LA Unbalanced LA Abstract selfplacement scale Short PM index Long PM index
Test-retest correlation
High education
Low education
Study 1/Study 2
Study 1/Study 2
0.70/0.73 0.76/0.77 0.57/0.64 0.64/0.64
0.62/0.58 0.77/0.74 0.44/0.60 0.65/0.60
High education
Low education
0.78 0.80 0.75 0.74
0.59 0.68 0.63 0.72
0.73 0.44 0.61
0.41 0.35 0.60
With respect to stability, we can see that among the less well-educated there are weaker correlations between most of the scales over the two rounds. However, the effect of educational level appears to be more pronounced for the balanced scales than for the unbalanced scales. Among the alternative measures, neither the postmaterialism indices nor the long libertarian-authoritarian scale show much difference in stability between the better and less educated respondents. The greatest difference by far occurs with the abstract left-right self-placement scale, where the correlation between the two rounds is 0.32 lower among the less-educated. These results may well reflect the difficulty of abstract political judgments for less-educated respondents. As with the analyses conducted with the full sample, however, the most interesting finding occurs when we inspect the correlations between the leftright and libertarian-authoritarian scales among the two educational groups. As previous research has suggested, there is a higher correlation between the left-right and libertarian-authoritarian scales among more educated respondents. However, when using the balanced scales the differences between the educational groups are muted: in the higher educational group R = 0.30; in the less well-educated group it is 0.23, a difference of only 0.07 (both associations were significant at P = 0.01). In contrast, when using the unbalanced scales there is substantial difference between educational groups. Indeed, the relationship between the left-right and libertarian-authoritarian scales actually changes direction (from R = 0.14 to R = - 0.15, a difference of 0.29; both were significant at P = 0.05) as we move from the highereducated to the less well-educated group.
Left-right and libertarian-authoritarian values
199
Table 3. The value scales regressed on to social characteristics
Left-right scales
Sex Age Class Religiosity Income Qualifications Constant R2
N
Balanced
Unbalanced
0.63 1.02 0.56 0.19 0.08 0.03 11.49 0.09 354
1.05 1.02 0.61 0.21 0.22 -0.12 9.39 0.16 356
(1.63) (1.44) (4.37) (2.19) (1.34) (0.29) (9.34)
(2.41) (1.29) (4.22) (2.11) (3.47) (-0.94) (6.75)
Libertarian-authoritarian scales
Sex Age Class Religiosity Income Qualifications Constant R2 N
Balanced
Unbalanced
-0.33 0.03 -0.05 0.16 -0.01 0.28 16.17 0.08 357
-0.85 0.05 -0.16 0.24 -0.01 0.37 19.24 0.17 356
(-0.98) (2.77) (-0.31) (2.07) (-0.24) (3.00) (13.61)
(-2.42) (4.34) (-1.04) (2.94) (-0.27) (3.81) (15.39)
Figures in brackets give the ratio of the parameter to its standard error.
Comparing the balanced and unbalanced scales' patterns of association with social and political characteristics We now turn to the relationships between the scales and other variables of interest. First, we regress the two types of scales on respondents' social characteristics.17 The results are shown in Table 3. In general, the patterns of association are the ones we would anticipate from previous research: left-right values are most strongly predicted by class and, for the unbalanced scale, income, while libertarian-authoritarian values are predicted by age, education and religiosity (as indicated by church attendance). The most noticeable difference between the balanced and unbalanced scales is the considerably higher proportion of variance explained in the latter by the social characteristics. In addition, there are differences in the parameter estimates between the balanced and unbalanced scales. Thus when using the unbalanced scales women are found to be more right-wing and
200
Geoffrey A. Evans and Anthony F. Heath
Table 4. Partisanship regressed onto value scales: parameter estimates Balanced LR Balanced LA Constant Unbalanced LR Unbalanced LA Constant
Con vs. Lab
Con vs. Lib/Dem
Lab vs. Lib/Dem
0.25 (7.6) 0.07 (2.5) -0.17 (-0.2)
0.17 (4.2) 0.06 (1.6) 1.86 (1.9)
-0.11 (-2.6) -0.01 (-0.3) 7.43 (8.9)
0.20 (7.8) 0.09 (3.4) 0.14 (0.2)
0.10 (3.4) 0.06 (1.7) 3.00 (3.2)
-0.i5 (-3.7) -0.04 (-1.0) 8.5 (7.9)
Figures in brackets give the ratio of the parameter to its standard error.
more libertarian than men, whereas neither effect is significant when using the balanced scales. People with higher incomes are more right-wing using the unbalanced scale, but not with the balanced scale. The effects of age and education on libertarian-authoritarian values are also somewhat stronger using the unbalanced scales, although they are significant for both types. Nevertheless, none of these differences in the magnitude of the parameters is particularly large, and all significant effects are in the same direction for both types of scales. It is also instructive to compare the effects of the social characteristics with those obtained using the alternative value measures as dependent variables. The explained variance in the left-right self-placement scale (6.8%), and both the short (3.7%) and long (6.6%) versions of the post-materialism index are all substantially lower than those for either the balanced or unbalanced Likert scales. Even if the Likert scales are divided into three broad categories so as to make them comparable with the short post-materialism index, they still have far stronger associations with other variables. Finally, we examine how the values scales relate to politics. As the panel studies were not undertaken following an election we use measures of party identification as a measure of strength of support for each of the three main parties. We then predict each of the three two-party contrasts with the leftright and libertarian-authoritarian scales as independent variables. 18 As the dependent variable is binary, we use logistic rather than OLS regression. Table 4 shows the results. We can see that both balanced and unbalanced left-right scales are associated with all three two-party contrasts. Both types of libertarian-authoritarian scale by comparison have weaker (although still statistically significant) effects on the Labour vs. Conservative contrast, and have no significant effect on the other contrasts. This presumably reflects the rather limited salience of libertarian-authoritarian issues for competition between the main parties in Britain. If we compare the balanced and unbalanced scales the effect
Left-right and libertarian-authoritarian values
201
of the balanced libertarian-authoritarian scale on Conservative vs. Labour preference is a little weaker than for its unbalanced counterpart, but this is to be expected, given the moderately strong correlation between the two balanced scales which would tend to reduce their unique effects. The unbalanced left-right scale also has slightly stronger effects on Labour vs. Liberal Democrat preference than does the balanced left-right scale, whereas the unbalanced left-right scale is more strongly predictive with regard to Liberal vs. Conservative identity. All of these differences in parameters are small, however, which suggests that the scales are not that different at predicting what they are designed to predict.
Conclusions
We have seen that both balanced and unbalanced Likert scales have greater stability and stronger relations with social characteristics than do the abstract left-right self-placement scale and the short postmaterialist values index. At the same time, both versions of the left-right Liken scale were found to have higher levels of internal consistency than the equivalent libertarianauthoritarian scales. Indeed, the balanced libertarian-authoritarian scale had rather low levels of internal consistency, although whether this is because respondents really do have more consistent attitudes about economic equality or because we devised poorer questions about libertarian-authoritarian issues cannot be determined. Similarly, although both dimensions of political values predict party support, the libertarian-authoritarian scale is a considerably weaker predictor of partisanship than is the left-right scale. This may also be a fault of the items which we have used, or more likely it reflects the less central place of such libertarian values in British politics. When the balanced and unbalanced scales are compared we find that by conventional criteria it is the latter which perform somewhat better. Both unbalanced scales have higher internal consistency and slightly higher stability than their balanced counterparts. This superiority is also found when analysing educational strata: among less highly educated respondents the reliability of the balanced scales declined more noticeably than did that of the unbalanced scales. The relations between the scales and social characteristics were on the whole rather similar, but there were some differences. The parameter estimate for income and gender was somewhat larger in the case of the unbalanced left-right scale than it was for the balanced scale. The parameter estimate for age and gender was larger for the unbalanced libertarian-authoritarian scale than it was for the balanced scale. Most noticeably however,
202
Geoffrey A. Evans and Anthony F. Heath
the variance in the unbalanced scales were more strongly predicted by the background characteristics: R z was considerably higher for both unbalanced scales than for their balanced equivalents. In contrast, the relations between the scales and politics were very alike for both balanced and unbalanced versions. This is important given that the scales are measures of political values, and are thus likely to be used to predict partisanship and voting. The higher reliability and stronger social bases of the unbalanced scales remains to be explained however. Why should these differences be found? We suggest that it is the presence of systematic bias in the unbalanced scales that increases their reliability and predictive power. Acquiescence effects reduce the correlations between the items in the balanced scales, but increase them in the unbalanced scales. Acquiescence effects are also likely to be stable and therefore to boost the over time correlations of the unbalanced scales. Poorer people tend to be prone to acquiescence and to also be more left-wing. Older people tend to be prone to acquiescence and to also be more authoritarian. Similarly, the high internal consistency of the unbalanced scales among the less well-educated group probably results from the consistency of acquiescence effects, to which they are considered to be prone, rather than high levels of reliability in the measurement of political values. The relatively high over time stability of the unbalanced scales among the less educated is also probably caused by the high over time stability of acquiescence effects, which may have boosted the correlations. In other words, the apparent superiority of the unbalanced scales with respect to reliability and stability may simply be due to the presence of correlated errors. The substantive differences found between balanced and unbalanced scales in the relations between the left-right and libertarian-authoritarian values can also be understood in this fashion. The items in the unbalanced libertarianauthoritarian scale are worded in an authoritarian direction. Those in the left-right scale are worded in a left-wing direction. If we assume that, at least in liberal democracies, authoritarians tend to be more right-wing on economic matters, and conversely, left-wingers tend to be more libertarian, then the opposing direction of the items in the unbalanced scales would serve to reduce the correlation between the two core values. Among respondents who are prone to acquiescence (i.e., the less well-educated) we would expect to find these effects exacerbated. In contrast, the balanced scales would reflect the actual associations between the core values more accurately, and because they are less prone to acquiescence effects would not be so strongly affected by respondents' level of education. Which is exactly what we found in Table 2.
Left-right and libertarian-authoritarian values
203
To sum up, the unbalanced scales are more internally consistent than the balanced scales and have stronger relations with social characteristics. This could be taken as confirmation of the superiority of the unbalanced scales. However, we should keep in mind that it is preferable to have scales which have better claims to validity, even if their reliability is lower. Over time stability - arguably the best test of reliability - is similar for both types of scales. The high internal consistency of the unbalanced scales is likely to result at least to a degree from correlated errors due to acquiescence response bias, as are their somewhat higher associations with social characteristics. We should also remember that the balanced and unbalanced scales are not exact equivalents that differ only in the reversal of direction of certain items. Some of the difference in their performance may be attributable to substantive differences in the items composing the scales. We propose therefore that the most reasonable conclusion to be drawn from the analysis is that the differences in the reliability - and in the patterns of correlations with criterion variables - of the balanced and unbalanced scales is sufficiently small and restricted in scope to justify the use of both sets of scales for analyses of political values and their correlates.
Notes 1. For example, analyses of the British Election Studies and of the British Social Attitudes Surveys have found that left-right and liberal-authoritarian values are the two main factors that account for the variance in the data sets in question (Heath and Evans, 1988; Heath et al., 1991; Himmelweit et aL, 1985; Robertson, 1984). Fleishman (1988) has found similar patterns in the United States. It is likely therefore that the two values are relatively orthogonal, which may partly account for the low levels of constraint, as indicated by the average correlation between responses to attitude items, often discovered between political attitudes in studies which have assumed that issues can be ordered on a single dimension of political ideology (i.e., Converse, 1964). 2. Much of this work has been done under the auspices of the Joint Unit for the Study of Social Trends, Nuifield College and Social and Community Planning Research. 3. See Evans et al. (1994) for an empirical test of the relative reliability and validity of various single and multiple indicator measures of political attitudes and values. 4. Classical test theory underlies the theory and practice of most approaches to attitude measurement. A n observed score on an attitude scale is composed of: true score + systematic error (bias) + random error. A person's true score is the average score they would obtain if measured an infinite number of times. Systematic error (bias) is a threat to validity: validity is the degree to which an indicator adequately measures the concept it is supposed to be indicating, rather than some other 'contaminating' influence. Random error is random disturbances which if the attitude were measured infinitely would have a mean of zero i.e., the true score. It is a threat to reliability, but not to validity. A n indicator can be completely reliable, and have no validity. To be valid, however, requires that an indicator have some degree of reliability.
204
Geoffrey A . Evans and A n t h o n y F. Heath
5. See discussions of the role of test-retest reliability in, for example, Carmines and Zeller (1979), and Bohrnstedt (1983). 6. A variety of inter-related types of validity testing exist. Concurrent validity, predictive validity, and convergent validity, for example, are all forms of criterion-related validity. For more detail on these terminological and substantive distinctions see, inter alios, Carmines and Zeller (1979), Cronbach and Meehl (1955), Nunnally (1978), and Bailey (1988). 7. 4560 names and addresses were issued to interviewers. 189 addresses were subsequently deemed out-of-scope, and the remaining 4371 addresses constitute the base for calculating the response rate. Full details of the 1989 BSAS are provided in Jowell et al. (1990). 8. Respondents from half the sampling areas used in the 1989 BSAS were intended to constitute the basis of the panel study. It subsequently proved impossible for practical reasons to include the areas in Greater London, and so the panel interviews were conducted in 67 sampling areas. At the end of the original BSAS interview, respondents had been asked whether they were willing to be reinterviewed. Twelve percent were not willing, and these respondents were excluded from the panel study. 9. The items in the two versions were kept unchanged in the second round but the order and distribution of the items in the two booklets was changed in order to check for context effects. 10. In the analyses which follow we have assigned the "don't knows" to the middle category and have excluded other missing data. 11. The introduction to the questions was: "On this sheet are a number of statements people have made about Britain today. Please read each one and tick a box to show how much you agree or disagree with it". 12. Confirmatory factor analysis indicated that a single dimension was an acceptable representation of the underlying structure of both balanced scales. Thus with the left-right items a LISREL analysis to test the goodness of fit of the one-dimensional model yielded a chi2 of 24.0 with 9 degrees of freedom, goodness of fit index = 0.977. A single dimension (with correlated errors to capture the question-direction effects) was also an acceptable representation of the underlying structure of the six items in the liberal-authoritarian scale (chia of 19.08 with 6 degrees of freedom, goodness of fit index = 0.982). 13. In Heath et al. (1994) we also constructed a longer ten-item scale which showed slightly greater stability over time and had greater predictive power than the short scale. However, we were unable to construct a suitable unbalanced equivalent for comparison. 14. Respondents were asked: "In political matters people talk of 'the left' and 'the right'. On this card are boxes running from the left to the right. Please tell me the letter of the box that best describes your own views',. There were eleven boxes in round 1 and 10 in round 2. Such abstract scales have been widely used and discussed, although doubts have been raised in Britain about the ability of respondents to think in the abstract way required. See Butler and Stokes (1974) and Luttbeg and Gant (1985). 15. It could be argued that the higher correlations obtained for the Likert scales are due to their having a greater range of scores than the abstract selfplacement scale and short postmaterialism index scales (25 points possible for the six-item Likert scales but 11 for the abstract scale and only 3 for the short postmaterialism index). However, even when the Likert scales are condensed into scales of equivalent range to the other measures, they still exhibit higher levels of stability. 16. Years of education are not a good predictor in Britain because of the many changes in the minimum school-leaving age. We have therefore used a variable which measures respondent's highest qualification. People with low education are defined as those with qualifications no higher than CSE grade 2. People with better education are defined as those with CSE grade 1 or above. (A CSE grade 1 is recognised as equivalent to an Ordinary level pass at the General Certificate of Education, which in turn is a prerequisite for continuing one's academic education to Advanced level and thence on to university or polytechnic.) 17. These are measured as follows. Social class: 6 ordered categories following the revised
Left-right a n d libertarian-authoritarian values
205
Goldthorpe class schema (Goldthorpe and Heath, 1992). Different orders are used for the two values, reflecting their different class bases. For analyses of left-right values class is coded as follows: petty bourgeois (6), upper service (5), lower service (4), routine nonmanual (3), foreman and technicians (2), manual workers (I). For analyses of liberal-authoritarian values the petty bourgeoisie is recoded to 4, the upper service to 6, and the lower service to 5. Income is coded in 15 bands. Qualifications: 7 ordered categories. Age: years. Church attendance: 7 ordered categories. Gender: male = 0, female = t. 18. We do not include the alternative measures in this analysis as the left-right self-placement scale is most likely a measure of party identification as much as anything else.
References Barnes, S. H. & Kaase, M. (1979). Political Action: Mass Participation in Five Western Democracies, Beverley Hills: Sage. Bass, B. M. (1956). Development and application of a scale for measuring social acquiescence, Journal of Abnormal and Social Psychology 53: 296-299. Bishop, G. (1976). The effects of education on ideological consistency, Public Opinion Quarterly 40: 337-348. Bohrnstedt, G. W. (1983). Measurement, in Peter Rossi, James Wright, and Andy Anderson (eds.), Handbook of Survey Research, New York: Academic Press. Butler, D. E. & Stokes, D. (1974). Political Change in Britain (2nd edition), London: Macmillan. Carmines, E. & Zeller, R. (1979). Reliability and Validity Assessment, Beverley Hills: Sage. Converse, P. (1964). The nature of belief systems in mass publics, in D. E. Apter (ed.), Ideology and Discontent, Glencoe: Free Press. Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology 78: 98-104. Cronbach, L. (1951). Coefficient alpha and the internal structure of tests. Psychometrica 16: 297-334. Cronbach, L. & Meehl, P. (1955). Construct validity in psychological tests. Psychological Bulletin 52: 281-302. Evans, G. A., Heath, A. F., & Lalljee, M. G. (1995). Examining the reliability and validity of measures of left-right and libertarian-authoritarian values, British Journal of Sociology, forthcoming. Feldman S. (1988). Structure and consistency in public opinion: the role of core beliefs and values, American Journal of Political Science 32: 416-440. Fleishman, J. A. (1988). Attitude organization in the general public: evidence for a bidimensional structure, Social Forces 67: 159-184. Goldthorpe, J. H. & Heath, A. F. (1992). Revised Class Schema 1992, JUSST working paper no. 13, Oxford: Nuffield College. Heath, A. F. & Evans, G. A. (1988). Working class conservatives and middle class socialists, in R. Jowell, S. Witherspoon, & L. Brook (eds.), British Social Attitudes: The 1988 Report, Aldershot: Gower. Heath, A. F., Evans, G. A., & Martin, J. (1994). The measurement of core beliefs and values: the development of balanced socialist/laissez-faire and libertarian/authoritarian scales, British Journal of Political Science 24: 115-132. Heath, A. F., Jowell, R., Curtice, J., Evans, G. A., Field, J., & Witherspoon, S. (1991). Understanding Political Change: The British Voter 1964-1987. Oxford: Pergamon Press. Heath, A. F., Jowell, R. and Curtice, J. eds. (1994). Labour's Last Chance? The 1992 Election and Beyond. Aldershot: Dartmouth. Himmelweit, H. T., Humphreys, P., and Jaeger, M. (1985). How Voters Decide (revised edition), Milton Keynes: Open University Press.
206
Geoffrey A. Evans and Anthony F. Heath
Inglehart, R. (1977). The Silent Revolution: Changing Values and Political Styles among Western Publics. Princeton: Princeton University Press. Inglehart, R. (1990). Culture Shift in Advanced Industrial Society, Princeton: Princeton University Press. Jowell, R., Witherspoon, S., & Brook, L. (eds.) (1988). British Social Attitudes: The 5th Report, Aldershot: Gower. Jowell, R., Witherspoon, S., & Brook, L. (eds.) (1990). British SociaIAttitudes. The 7th Report, Aldershot: Gower. Jowell, R., Brook, L.,Taylor, B., & Prior, G. (eds.) (1991). British Social, Attitudes: The 8th Report, Aldershot: Dartmouth. Jowell, R., Brook, L., Prior, G., & Taylor, B. (eds.) (1992). British Social Attitudes: The 9th Report, Aldershot: Dartmouth. Luttbeg, N. R. & Gant M. M. (1985). The failure of liberal/conservative ideology as a cognitive structure, Public Opinion Quarterly 49: 80-93. McClosky, H. & Zaller, J. (1984). The American Ethos: Public Attitudes Toward Capitalism and Democracy, Cambridge, Mass.: Harvard University Press. Norris, P. (1994). 'Labour Party Factionalism and Extremism', in A. F. Heath, R. Jowell and J. Curtice eds. Labour's Last Chance? The 1992 Election and Beyond. Aldershot: Dartmouth. Nunnally, J. (1978). Psychometric Theory, New York, McGraw-Hill. Robertson, D. (1984). Class and the British Electorate. Oxford: Basil Blackwell. Rokeach, M. (1973). The Nature of Human Values, New York: Free Press. Schuman, H. & Presser, S. (1981). The acquiescence quagmire, in Questions and Answers in Attitude Surveys, New York: Academic Press. Stringer, P. & Robinson, G. (eds.) (1991). Social Attitudes in Northern Ireland, Blackstaff Press: Belfast. Stringer, P. & Robinson, G. (eds.) (1992). Social Attitudes in Northern Ireland: The Second Report, Blackstaff Press: Belfast. Stringer, P. & Robinson, G. (eds.) (1993). Social Attitudes in Northern Ireland: The 3rd Report, Blackstaff Press: Belfast. Whitefield, S. & Evans, G. A. (1994). The social background to the December elections in Russia: public attitudes and the transition experience, Post-Soviet Affairs 10.