Mapping and Explaining the Use of the Left-Right Divide *

This study is about mapping and explaining the use of the Left-Right divide across 14 countries from 5 Continents and relies on the richness of the post electoral mass surveys from the Comparative National Election Project: 14 countries and 18 elections spread over 5 continents. The paper shows not only how extensively the LR divide is used in these 14 countries, but also explains variation across both individuals and countries in terms of the factors determining LR recognition and use. Overall, it is shown that, although seen world-wide, the LR divide (both for self-placement and party placement) is more present in long consolidated and middle-aged democracies and countries with freer media systems than in new democracies and societies with less free media systems. In the case of parties LR placement, party size also counts: larger parties are more easily placed. Additionally, we also show that LR recognition is more socially and politically determined in long consolidated and middle-aged democracies and in countries with more freedom of the press than in new democracies and in systems with less free media system. These findings add to the existing knowledge about these topics because previous studies were either country/Continent specific, or, if global in nature, never invested in explaining individual and system variation across 14 from 5 Continents. Besides, these findings mean that in the long term probably the new democracies will converge with the long consolidated or middle-aged ones, but this is an empirical question to be researched in future studies.

The data necessary to reproduce numerical results can be found in http://bpsr.org.br/files/arquivos/Banco_Dados_Freire_Kivistik.html
At the individual level, the division between left and right functions as an instrument to reduce the complexity of the political universe; at the systemic level, it functions as a code of communication (Fuchs and Klingermann, 1990).
In spite of all the theories about the "end of ideology" (Aron, 2002;Bell, 2000;Lipset, 1987), the "end of history" (Fukuyama, 1989) and about a certain overcoming of the division between left and right (Giddens, 1996), the truth is that these same theories have been shrouded in an ideological character and, soon after being defended, have been followed by the appearance of new ideological forms or a renewed prominence of the "old" ideologies (Heywood, 2003).Furthermore, various studies have documented the remaining (or even increasing) importance of electors' positions on the LR (or liberal-conservative in the US and some other countries) scale as a defining factor for their voting choices in several regions of the world: in long consolidated democracies in Europe and North America (Franklin et al, 1992;Gunther and Kuan, 2007;Gunther and Montero, 2001;Van der Eijk et al, 2005), in new/middle-aged democracies in Southern and Eastern Europe (Gunther and Kuan, 2007;Gunther and Montero, 2001;Kitschelt et al, 1999;Markowski, 1997), in Latin America (Gunther and Kuan, 2007;Luna and Zechmeister, 2010) and in at least some countries in East Asia (Jou, 2010;Lee, 2007).
Having established the continued relevance and usefulness of the LR divide for both long consolidated and new democracies, and therefore the relevance of studying these topics to understand the functioning of democratic political systems, one could reasonably ask: what relevant elements can be added to this area of knowledge?With the present study about mapping and explaining the use of the LR divide across 14 countries from five continents, we believe that we can add the following: first, we can add relevant data and interpretations given the richness of datasets from the Comparative National Election Project (CNEP I-III).The richness relies on the diversity of countries included that allows us to compare and explain variation in LR recognition and use across 14 countries from five continents, i.e., on a scale never done before (usually these type of analyses are continent or country specific: Freire, 2006b and2008;Freire and Belchior, 2011;Knutsen and Scarbrough, 1995;Lee, 2007;Luna Zechmeister, 2010;Markowski, 1997;Zechmeister, 2006 and2010; some cross continent analyses done before map LR recognition but neither explain variation across individuals nor across countries: Dalton, 2006;Gunther and Kuan, 2007;Noël and Thérien, 2008).The reader should bear in mind that CNEP I-III includes post electoral mass surveys from 14 countries and 18 elections (cases) from five continents: Europe, North and South America, Africa and Asia.We have relevant data for the present paper from the following cases included in CNEP I-III: Argentina 2007, Bulgaria 1996, Chile 1994 and 1999, Greece 1996, Hong Kong 1998, Hungary 1998 and   2006, Italy 1996 and 2006, Mexico 2006, Mozambique 2004, Portugal 2005, South Africa   2004, Spain 1996 and 2004, USA 2004, and Uruguay 1994 and 2004. 2   Considering all this, we believe that we can advance scientific knowledge about the recognition and use of the LR divide across 14 countries from five continents by pursuing the following objectives: first, at the individual level, we want to know and explain how extensively the LR divide is used (LR self-placement: LR SP/ LR party placement by citizens: LR PP) in the 14 countries.Second, mostly using the two step hierarchical regression (multi-level modelling) approach (Achen, 2005;Jusko and Shively, 2005;Lachat, 2008), we want to explain variation across countries in terms of the factors determining LR recognition and use.These findings add to the existent knowledge about these topics because previous studies were either country/continent specific, or, if global in nature, never invested in explaining variations across individuals and countries.
But there is more about the relevance of these topics.On the one hand, this fresh CNEP I-III data will allow us to update information about the levels of LR recognition and use worldwide.But on the other, we do not know from previous studies if the structure of individual determination of LR recognition is the same worldwide or not.If it is similar (for example, if education, political interest, media exposure, and party identification work in the same direction as in Europe and North America), then the differences in overall 2 For the US in 1992 and Greece in 2004 we have information about the level of LR recognition but not enough information (relevant variables missing) to explain variation both at the individual and the country level.recognition and use between countries might be due to compositional differences in the aggregate levels of education, political interest, media exposure and party identification in the different countries, and some convergence between countries might be expected in the future when those compositional differences are reduced.But if that structure of determination is rather different, then country specific macro events should weigh more to explain LR recognition and use, and thus convergence between countries will be less probable in future.Moreover, this situation would also raise serious doubts about the possibility of comparing mass use of the LR schema worldwide.We will also try to specify the impact of some macro-level factors (age of the democratic regime, freedom of media, and level of educational development) on LR recognition and use.Again, if these macro-level explanatory factors are indeed relevant to explain variations in both the overall levels of recognition and in the structure of determination, as we expect, then we should see some convergence between countries when democratic regimes (from non-Western parts of the world) become more mature, freedom of the press is more solid, and social development more advanced.This would also give indications of the heuristic value of LR for comparing mass political attitudes and behaviour worldwide.
We will proceed as follows: in the second section, we will review the theory about LR recognition and formulate hypotheses.In the third section, data and methods will be presented.In the fourth and fifth sections, we will analyse the questions related to LR recognition and use.We finish with our conclusions.

Theory and Hypotheses
Following Converse's study (1964), we assume that if people recognize the labels left and right, if they are able to place themselves and parties on the LR continuum, then it is reasonable to claim that this is at least one indicator that these ideological concepts are still meaningful and valid. 3Therefore, in order to assess the importance of LR concepts, we will first analyse the levels of LR SP and how that recognition is influenced by individual and country level factors.Data from advanced Western democracies confirm, for instance, that on average LR is widely understood by almost 90% of people (Knutsen, 1998;McAllister and White, 2007).The other regions of the world are less studied; nevertheless, the existing research supports the idea that, although usually to a variable 3 Of course, there are other indicators of the meaningfulness and validity of the LR divide.See Van der Eijk et al, 2005 for the impact of LR on the vote and Noël and Thérien, 2008: 32-82 for the consideration of "clashes over social equality"/substantive policy content in terms of "social justice".However, we believe that our option is more adequate because is it less demanding to compare new and long consolidated democracies and because it does not make assumptions about a one-dimensional meaning regarding the LR divide.(Barnes, 2002;Colomer and Escatel, 2005;Dalton, 2006;Evans and Whitefield, 1998;Freire, 2006a;Gunther and Kuan, 2007;Jou, 2010;Kitschelt et all, 1999;Lee, 2007;Luna Zechmeister, 2010;Markowski, 1997;Noël and Thérien, 2008;Zechmeister, 2006 and2010).
Although it is known that the LR divide is widely recognized and used by mass publics in several regions of the world, cross continent comparisons are less common (but see Dalton, 2006;Gunther and Kuan, 2007;Noël and Thérien, 2008) Traditionally, the recognition and utilization of the LR divide is associated with the sophistication and awareness of individual citizens -more sophisticated people understand these ideological labels better and in a more coherent way (Freire, 2006a;Freire and Belchior, 2011;Kitschelt and Hellemans, 1990;Zechmeister, 2006).The most popular way to conceptualize this is through education.In addition to education, a person's political interest and age are most often said to increase the level of recognition and use of the LR divide.The more educated and politically interested people are, the more familiar they are with ideological terms, as they have more resources and motivation.Usually, other aspects being held constant, older people are more aware of political terms because they have had more time to learn them through socialization.Only in cases of regime change may this be questionable, because then one could expect young people to adjust to the new situation more quickly (Freire, 2006a).But there are still other factors that have been shown to explain LR recognition like media exposure and party identification -people with more media exposure and stronger levels of party identification are also more likely to be better informed about and/or engaged in politics and therefore are also expected to show higher levels of LR recognition and use (Freire, 2006a;Freire and Belchior, 2011;Kitschelt and Hellemans, 1990;Zechmeister, 2006).From previous findings, we have derived our individual level hypotheses: Hypothesis on education and LR recognition: more educated people recognize and utilize the terms LR better than the less educated; Hypothesis on political interest and LR recognition: more politically interested people recognize and utilize the terms LR better than the less interested; Hypothesis on age and LR recognition: others things being held constant, older people recognize and utilize the terms LR better than younger people.
Hypothesis on media exposure and LR recognition: more informed people recognize and utilize the terms LR better than the less informed; Hypothesis on party identification and LR recognition: people with stronger levels of party identification recognize and utilize the terms LR better than those with weaker levels of party identification.
The problem with the latter two hypotheses is that questions about media exposure and party identification were not asked in all countries taking part in the CNEP I-III, therefore, these two additional variables will be used only in a limited way.Of course this is a common problem when we use secondary survey data but it is not necessarily an obstacle to producing good quality research (Kiecolt and Nathan, 1985) they follow the so-called "political sophistication model" and are pretty common in the literature about LR conceptualization among the mass publics (see Converse, 1964;Freire, 2006b;Freire and Belchior, 2011;Fuchs and Klingemann, 1990;Kitschelt and Hellemans, 1990;Klingemann, 1979;Zechmeister, 2006).As for the macro level explanatory variables (see below), they will be deduced from the theoretical literature cited below.
Let us now pass to the macro level.The formation of ideological (and partisan) identities is a process that continues to evolve during individual socialization (Barnes, Mc-Donough and Pina, 1985;Campbell, 1980;Converse, 1964Converse, , 1969;;Freire, 2006a;Niemi et al, 1985).Moreover, it is strongly dependent upon an existing environment of effective partisan and ideological differentiation, associated with the existence of free political competition.In authoritarian regimes up to the beginning of their democratic transitions, parties were usually a proscribed reality and ideological differences were repressed.Despite the repression of partisan and ideological pluralism, at times it did exist, although it was very limited.Therefore, the conditions for the formation of ideological (partisan) identities were severely reduced, especially when comparing these new regimes to older democracies.Even in competitive political environments, for individuals to identify themselves with certain LR ideological areas, it is necessary that such political organizations and notions, as well as the image and content that define and/or are associated with them, consolidate their presence in the political and media arenas (Barnes, 2002;Barnes, McDonough and Pina, 1985;Converse, 1969;Gunther and Montero, 2001;Niemi et al, 1985).Thus, politicization and political intermediation are preconditions for the formation and relevance of LR attitudes by mass publics (Beck et al, 2002;Gunther and Kuan, 2007).Moreover, the usual instability associated with new party systems is yet another reason for a lower level of LR recognition by citizens in new democracies (Dix, 1992;Freire, 2006a;Gunther and Kuan, 2007;Mainwaring and Torcal, 2005;Rose and Munro, 2003).
Therefore, we expect that in long consolidated democracies and more developed societies individual level variation in terms of recognition and use of the LR divide is more determined by education, income, political interest, media exposure, and strength of party identification than in new democracies and less developed countries.In the latter regimes and societies, lower levels of LR recognition are more widespread, thus more determined by cultural, historical and political factors than by individual level factors.Thus, we derive our macro level hypotheses: Hypothesis on the age of democratic regime and the overall level of LR recognition and use across countries: the longer the democratic tradition in a country (as measured by "the numbers of years since the last democratic transition"), the higher the level of recognition and utilization of the LR divide by mass publics.
Hypothesis on the age of the democratic regime and the individual level determination of LR recognition and use across countries: the longer the democratic tradition in a country, the more the level of recognition and utilization of the LR divide by mass publics is determined by individual level determinants (age, education, political interest, strength of party identification, and media exposure).

Hypothesis on the level of social development of societies and the overall level of LR recognition and use across countries:
The more socially and politically developed a country (as measured by "the Freedom House index of overall press freedom" and "the percentage of people with secondary education or more") the more the level of recognition and utilization of the LR divide by mass publics.

Hypothesis on the level of social development of societies and the individual level determination of LR recognition and use across countries:
The more socially and politically developed a country, the more the level of recognition and utilization of the LR divide by mass publics is determined by individual level determinants.1) Polity IV for "time elapsed since the last democratic regime transition"; 2) "Percentage of persons with secondary education completed or more" -computed from the CNEP mass surveys in each country. Notes: 1) Hong Kong is not a democracy, especially not at the time of the survey, as can be clearly seen in the Freedom House report for that country in 2002.See http://www.freedomhouse.org/template.cfm?page=363&year=2002&country=2471 (retrieved on 29/09/2011).
Moreover, the country was considered "partly free" and scored a 5 in terms of "political rights" (on a scale from 1, "the most free", to 7, "the least free").Thus, this case was either considered "missing" in our macro level analysis or scored with the value "0" for the number of years since the last democratic transition.
In Table 1 we cross "the numbers of years since the last democratic transition", an indicator of the level of democratic consolidation in each country, with an indicator of the level of socio-economic development, "the percentage of persons with secondary education completed or higher".In the breakdowns for each variable in Table 1 there is, of course, some degree of arbitrariness: they are made by taking into account the average point in each distribution, as well as the dispersion around the mean.But there is also some substantive sense in the breakdowns: for example, we consider "new"/"young" democracies those of 24 years or less since the last democratic transition and the middle-aged democracies those between 26 and 41 years old.The long consolidated democracies have been so for more than 41 years.Moreover, when we look at the countries/years in each group, we see that the partition makes sense politically.Similar notes could be made about the partition concerning "the level of development of societies".However, we should bear in mind that our dataset has a larger number of new democracies (13) than middle-aged (3) or long consolidated democracies (4).We can see that in terms of the level of development, there are many more cases with middle (7) and high (9) levels of development ( 16) than with low levels (4).Moreover, we can say that there is a reasonable spread of different regimes across each level of social development.Thus, although new democracies are in greater quantity, we are convinced that both the richness and uniqueness of the CNEP I-III data and the reasonable variation in macro independent variables allow us to proceed with our analysis.
Finally, we have a third set of two hypotheses that apply only to the recognition of the location of parties in the LR scale (LR PP): Hypothesis on party size and the overall level of LR recognition and use across countries: the higher the (electoral) size of the two parties under recognition taken together, the higher the overall level of recognition of parties' location on the LR scale.

Hypothesis on party size and the individual level determination of LR recognition
and use across countries: the higher the (electoral) size of the two parties under recognition taken together, the higher the recognition determination of parties' location on the LR scale by the independent variables in each country.
We expect this effect because larger parties have more chances of being recognized, especially by the more educated, politically interested and older voters, as well as by those with more media exposure and stronger partisan attachments.First, because they are probably older than the smaller parties.Second, because the policy proposals of larger parties usually receive more media attention.Third, because larger parties usually have more resources to publicize their actions and mobilize voters.

Data, Methods and Operationalization
In order to accomplish our objectives and test the hypotheses, we use as many countries and election surveys from the CNEP I-III dataset as possible, a solution that is variable according to the different objectives and information available.We usually use around 12-13 and 17-19 cases.
To estimate the impact of macro-level items on the country level of determination in LR recognition we use the multi-level technique known as "two-step hierarchical regression" (Achen, 2005;Jusko and Shively, 2005).At the individual level (first step), we use logistic regressions to explain variation in our dependent variables in each country.At the First, as Jusko and Shively (2005: 12-13) note, the "two-step strategy draws heavily on the statistical foundations of the hierarchical linear models... but maintains many of the advantages of both portioning and pooling strategies...And it accomplishes this without loss of efficiency as compared to pooling strategies", and with more flexibility as we will explain below.In the first step, we estimate separate logistic regressions to explain the LR recognition in each one of the 12-19 cases.Education, Political Interest, Age, Media Exposure, and Strength of Party Identification are the independent variables at level one.The regression coefficients from the first step are then taken as the dependent variables for the second step, where the independent variables are system level, and the cases ("N") are the 12-19 countries/election surveys.We acknowledge that a larger N would have been better, but it is important to stress three things.First, while ideally N should be greater than or equal to 30, statistical simulations for N=10 reveal "the regression coefficients and lowest-level variance components are again estimated without bias", and only "the group-level variance was over-estimated, with a bias up to 25 per cent" (Maas and Hox, 2005: 90, 91).Since we have 12-19 groups, and are mainly interested in explaining variation in the regression coefficients, we can conclude the sample size at step two is adequate, even if far from ideal.Second, in cross-country comparative political analysis, larger samples for the second step are rather rare, but robust findings have been found with 26 (Jusko and Shively, 2005), 16 (Lachat, 2008), 15 (Weldon, 2006), and 13 groups (Freire and Kivistik, 2013).Moreover, studies about LR among the mass publics usually rely on an even smaller number of cases/countries in level 1: for example, around 13 and 6 cases in Freire and Kivistik (2013) (using CNEP III data); 13 cases in Freire (2006aFreire ( , 2006b)); 13 cases in Knutsen (1997) (the latter three studies using European Value Study 1990-1999 data).
Third, by using the robust standard errors procedure we can be reasonably sure about the robustness of our findings.
Additionally, when we want to explain cross country variation in the overall level of LR recognition (self and parties) and in the overall level of individual determination of LR recognition, we will proceed in yet another way.Using the "percentage of persons that can locate themselves (or the parties) in the LR scale" as our dependent variable, we then regress these values on a set of independent variables.Or using R 2 s for each country, from the individual level models to explain LR recognition, we then regress these values on a set of independent variables.

Left-Right Recognition Across 14 Countries from Five Continents
To begin with, we will analyse LR recognition across countries.First, we will describe the material from each country, then we will proceed to testing the hypotheses at the micro and macro levels.Notes: (1) Countries organized by descending order of overall level of recognition.
Values -1 and system missing are treated as DK, except in the case of CL93, IT06, MX06 PT05, US92, US04 and UR94, where they have been left out of the analysis.
The first measure to evaluate LR recognition in a society is the proportion of people who have placed themselves on the LR scale.As we can see (Table 2), the percentages vary a lot across countries.The general level of recognition accounts for more than three-quarters of the respondents and in Mexico, less than half of the respondents.So we can conclude that the concept is known at large over the world, like many other studies did before us (Dalton, 2006;Gunther and Kuan, 2007;Noël and Thérien, 2008).The expected tendency should be that long consolidated democracies have a greater proportion of recognizers than newer ones and that the level of recognition is increasing with the passage of time in new democracies (Freire, 2006a), but the data confirms this tendency with some deviations.The lowest levels of ideological recognition are in Latin America (Mexico: 46; Argentina 53 percent), Asia (Hong Kong: 69), Africa (South Africa: 70; Mozambique: 73), and Eastern Europe (Hungary: 76).But some consolidated democracies in some years (US 2004, Spain 2004, Italy 1996) are not that much above the level of the new democracies just mentioned.Of course, there are always some short-term factors that explain variation in the same country across relatively short periods of time.For example, we know that the case of Italy, due to the strong turbulence in the Italian party system since 1994, is perhaps an example of that situation (Freire, 2006a).This is probably the case also for the US 2004 and Spain 2004, namely because data from US 1992 and Spain 1993 are much more in line with our expectations.
In any case, taking all countries together, we can still say that in older democracies (in Europe and North America: US 1992) the average level of LR recognition is much higher than in newer democracies in Africa, Asia and Latin America: long consolidated and middle-aged democracies are overwhelmingly at the top of Table 2, first column.Nevertheless, some people might argue that central categories in a LR scale are often used by non-sophisticated citizens as a kind of "no answer" (NA)/"don't know (DK)": "false centrists".To avoid possibly misleading results, in columns three and four of Table 2 we present the same result as before but after collapsing the probably "false centrists" (categories 5 and 6 in 1-10 LR scale).The results seem to be even clearer than before, but exceptions still remain: at the bottom of column 4 we find mainly new democracies, but Portugal is also there; at the top, among long consolidated democracies, we also find Mozambique, Uruguay and Hungary.Nevertheless, to have a clearer picture about this we need to proceed with the regression analysis.But before that, we analyse LR recognition in terms of the LR location of one or two of the largest parties in each country (Table 3).Notes: (1) Countries organized by descending order of the overall level of recognition.
(2) Can place the two largest parties in the LR scale (values 1-10); (1) Can place one of the two largest parties in the LR scale (values 1-10); (0) -DK, NA, Missing values.
Values -1 and system missing are treated as DK, except in the case of CL93, IT06, MX06, PT05, US92, US04, and UR94 where they have been left out from the analysis.
In Table 3 we present the percentage of respondents in each country that have placed the two largest parties on the LR scale.It is remarkable that in almost all countries (except Hong Kong, Mozambique, South Africa, and Argentina) more than two-thirds of the respondents have been able to do that.Of course, in this case we are not dealing with "correct" or "wrong" placements, for example, left-wing parties that are located on the right, or vice-versa.Nevertheless, the difference between older and newer democracies seems to somehow be present again when using this measure.Countries with relatively long democratic traditions (Greece 1996and 2004, Spain 1993, Italy 2006) seem to be well represented among the group, which has the highest levels of recognition of parties' LR location, and those with few democratic years (Bulgaria, Hong Kong, Mexico, Mozambique, South Africa) have lower levels of recognition, but several exceptional cases are still present.The results of the OLS regressions to explain cross country variation in the overall level of recognition reveal the following (Table 4) 4 : first, in terms of self LR recognition, freedom of press and age of democratic regime work in the expected direction -countries with more freedom of press (lower values in the Freedom House's scale of freedoms) and with longer time elapsed since the last democratic transition do show higher levels of recognition.However, the age of the democratic regime is only significant when the dependent variable is self LR recognition and "false centrists" were collapsed.Conversely, freedom of press always works in the expected direction and in a significant way.
For self LR recognition when "false centrists" are collapsed, there is a factor that works contrary to expectations: the higher the level of education in a country, the lower the level of recognition.It is not easy to explain this, nevertheless, we can conclude that both political socialization in the democratic regime (self) and mass media political intermediation (self and parties) are factors that have a significant effect on the overall level of LR recognition.2) The dependent variables are the percentage of respondents in each country/year that are able to place themselves or the two largest parties on the LR scale.

4
From Table 4 to Table 6.3 we present the error margins for the estimations of the coefficients as: *** p<0.01, ** p<0.05, * p<0.1.This means the probability of rejecting H0 when the null hypothesis (i.e., b/regression coefficient = 0) is true.Thus, we only consider rejection of the null hypothesis when the probability of a false rejection (i.e., the error margin) is below 1% (0.01), 5% (0.05) or 10% (0.1).The findings about which individual level variables have an effect on self LR recognition are in Tables 5.1-5.3. 5First, when we compare the explanatory power of the model, 5 When the dependent variable is a dichotomy (1, the respondent is able to locate him/herself in the LR scale, 0 otherwise), like in Tables 5.1-5.4,we use Logistic (L.Reg.) instead of OLS regressions.In L. Reg.we estimate instead the probability of success of a phenomenon (1, the respondent is able to place him/herself in the LR scale).Thus, the relationship between the independent and dependent variables is not linear -it is better described by an S. So, a logarithmic transformation is needed to transform a non-linear relation into a linear one: this is done by the logit (Jovell, 1995).The regression constant (a) represents the value logit of the probability of an event happening when the value of the independent variable (X) is zero.The regression coefficients (b) now represent the change in the odds ratios' logit for an event to happen, for one unit change in the independent variable (X).The odds ratios represent the ratio of the probability of success for a specific event (LR recognition) vis-à-vis the probability of failure for that same event (LR non-recognition).Because it is easier to it seems that the effect of the sophistication level of LR recognition is usually larger in Europe and in the US than in countries from other regions of the world, although with some exceptions.It seems to indicate that in Europe and the US we find our expectations more easily (more educated, politically interested and older people are usually more able to recognize the LR divide), and that in other regions of the globe levels of recognition, or a lack of them, are socially more diffuse.Thus, in these latter countries, political and cultural factors probably weigh more than in Europe and the US.In the case of Europe, Bulgaria is the exception, with one of the lowest levels of determination.But this might mean that the lower explanatory power of the model that we find in non-European and non-North American regions can also be found in some new democracies from Eastern Europe (Freire, 2006a).Anyhow, it is clear that the effect of the sophistication varies a lot: from 0.02 in Uruguay 1994 to 0.24 in Italy 2006.When we consider an alternate treatment of the dependent variable by collapsing "false centrists", the major effect is to usually depress the strength of the R 2 (columns 5 and 6 of Tables 5.1 to 5.4).
If we look at the variables' impact we see that quite often the only variables that have a significant impact on LR recognition are education and/or political interest.second, in some cases (Hungary 2006, Italy, Chile 1993, andPortugal) the effect of age is negative.This means that the oldest group (AgeD3: 60 years old or more) in these countries is less aware of LR than the youngest cohort (18-29 years old), a syndrome that was found to be characteristic of new European democracies elsewhere (Freire, 2006a).
However, in CNEP I-III this seems not to be the case: first, there are long consolidated or middle-aged democracies with a negative effect, and we cannot see the same results in several other new democracies of the CNEP I-III sample.Overall, however, we would say that the results point in the expected direction: more education, age and political interest are usually associated with higher levels of self LR recognition.However, the effect of age is variable across countries.
interpret, we usually consider the inverse of that measure (e b ), which allows us to read the coefficient as the change in the odds ratio of a certain event happening associated with one unit change in the independent variable (X).Since the dependent variable is not an interval variable, we cannot make estimations for regression coefficients using ordinary least squares (OLS), so the maximum-likelihood method is used instead.The Pseudo R 2 is the equivalent measure for the level of variance explained (R 2 ) in OLS regressions, even if less robust.Finally, the Wald statistic allows us to see which variables have more importance to explain variations in the dependent variable by testing the null hypothesis (b = 0).Of course, one could reasonably argue that the previous models are underspecified: media exposure and strength of party identification are missing.This omission is because if we want to include these two extra factors, we pass from the 18 cases under scrutiny to 13.Nevertheless, we performed new regressions for the 13 countries with the relevant variables testing the effect of "media exposure" and "strength of party identification", besides the other three variables (due to spatial limitations, these tables are not shown here).
There are basically three conclusions.First, the impact of the new independent variables is always relevant and the R 2 s usually increase.Second, both factors boost the level of LR recognition.Third, when both are available, both have an independent and significant im- About the variables that have a significant impact on the recognition of parties' location on the LR divide, we conclude that the pattern of determination is rather similar to the one found for LR SP.Let us now pass to explaining cross national variations in a systematic way.First, we take the country by country R 2 s (from the models to explain individual level determination of LR recognition) as our dependent variables and the macro-level items as our independent variables.From these tests we can conclude that LR recognition (self and parties) is always more socially and politically determined in long consolidated (or middle-aged democracies) than in new democracies.Freedom of press also has the expected effect (more freedom of press implies more determined profiles of LR recognition), but these effects are only significant when the fully specified models (i.e., those including education, political interest, age, media exposure, and party identification) are taken into account.Finally, unexpected results are present again for education: lower levels of determination in LR recognition in societies with higher levels of education.
Finally, we come to the second step of the hierarchical regressions, i.e., now we want to explain cross country variation in the patterns of determination of LR recognition for each and every predictor ("education", "political interest" and "age") -Tables 6.2-6.3.
Our dependent variables are now the regression coefficients for the impact of "education", age") has a significantly higher level of LR recognition than the youngest cohort (reference group: 18-29 years old).
In the case of parties' LR recognition, only the variable "age of the democratic regime" has a significant effect, and only on the impact of "political interest": in older democracies any individual "political interest" has more chances of increasing the level of parties' LR recognition than in new democracies.
Summing up, we conclude that although cross country variation on the strength of the impact of the sophistication model (Tables 5.1-5.4;Table 6.1) and in the overall level of LR recognition (Table 4) can be clearly explained by macro level factors (age of democratic regime, freedom of press, and education), the case is less clear when we want to explain cross country variation in the impact of each individual level explanatory factor (Tables 6.2 and 6.3).

Concluding Remarks
We tried to understand and explain, both across individuals and countries from five continents, the level of recognition of the LR divide at the mass level.Previous studies have concentrated either on European cases or, when using a more global approach, on a description of patterns across countries/regions, and were less focused on explaining those differences.Thus, the present study contributes with unique data and approaches to the understanding of the use of the LR divide across the globe.
In terms of LR recognition (both for self and parties), we used CNEP I-III data from around 18-21 cases/countries in five continents.In line with previous studies, we found that the LR divide is widely recognized around the globe, although usually less in non-European and non-North American regions.What was less known was that at the individual level, the usual suspects (the sophistication model: "education", "political interest", "age", "strength of party identification", and "media exposure") were valuable to explain individual level variation in LR recognition, not only in Europe and North America (US) but also in Asia (Hong Kong), Africa (South Africa and Mozambique) and in Latin America (several countries).Thus, this pinpoints the country differences in overall levels of LR recognition and might be due to compositional differences in the aggregate levels of education, political interest, media exposure, and party identification in the different polities, and also that some convergence between countries might be expected in future, when and if these compositional differences are reduced.
Moreover, we found that cross country variation in the total strength of the impact of the sophistication model (R 2 ) and in the overall level of LR recognition can be more easily explained by macro level factors (age of democratic regime, freedom of press, party size, education) than when we want to explain cross country variation in the impact of each individual level explanatory factor.In the first case, the overall level of LR recognition is more dependent on macro level factors: age of the democratic regime (only for self LR recognition when collapsing "false centrists") and freedom of press (for all dependent variables: self and parties) both boost the overall level of LR recognition in a country.
But in the second set of situations, concerning cross country variation in the impact of each individual level explanatory factor, we found that in terms of self LR recognition macro level indicators only have an impact on "education" and "age: oldest cohort vs.
youngest cohort": in more free media systems (for "education") and in more educated societies (age), "education and age" have more of an impact on individual level self LR recognition, respectively.In the case of parties' LR recognition, we concluded that in long consolidated (or middle-aged) democracies, "political interest" has more impact on LR recognition than in new democracies.Thus, in the latter polities the determination of individual level parties' LR recognition is less dependent upon political interest and is therefore more diffuse, i.e., more dependent upon macro-level political and cultural factors.
Overall, however, we found that LR recognition (for self and parties) is more socially and politically determined (at the individual level) in long consolidated democracies and in systems with more freedom of press than in new democracies and in systems with not so free media systems.Thus, both the age of the democratic regime and freedom of press explain cross country variation in the individual level determination of LR recognition across countries.Consequently, there are reasons to expect some convergence between countries when democratic regimes (from non-Western parts of the world) are more consolidated, there is more freedom of the press, and the level of social development is stronger.Moreover, the evidence shown, concerning both the individual level and macro level of determination of LR recognition and use, gives clear indications of the heuristic value of the LR scale for comparing mass political attitudes and behaviour worldwide.
To conclude, some cues for future research are due.We found that not only is LR widely recognized and used across countries from five continents, although usually more so in Europe and North America, but also the individual level determinants LR recognition are heuristically useful world-wide.And we also found that both the sophistication models (for LR recognition) work better in long consolidated (or middle-aged) democracies and in freer societies.But does this really mean that in the long term new democracies will converge with the long consolidated (or middle-aged) ones?This is certainly an empirical question that cannot be answered here.Moreover, we acknowledge that both the micro and macro models are relatively poor in terms of explanatory factors considered, but this is due to the limitations of the CNEP I-III survey data available, as well as the small number of cases at the macro level and the research design adopted (a comparative variable oriented and synchronic approach).There are clear tasks for future research either using more cases or more in-depth historical approaches based on case studies.
Revised by Priscila Moura

Submitted in May 2012
Accepted in August 2013 is widely recognized and used by the mass publics in very different regions of the world macro level (second step), we pick up the regression coefficients from the first step and consider them as our dependent variables.
pact, but in any case these factors (or at least one of them) appear among the most relevant factors to explain individual level variation in LR recognition.Thus, both media political intermediation and political socialization have an impact on LR recognition.Due to spatial limitations, we cannot show the tests of the models to explain individual level variation in the recognition of the two largest parties' LR position across countries.Contrary to what we find in terms of self LR recognition, in the case of parties' LR recognition we now find a mix of long consolidated(US 2004) and new democracies (Bulgaria, Hong Kong, South Africa) with the highest R 2 ; and at the bottom of the ranking, with the lowest R 2 , we find both new(Mexico, Uruguay 2004, Spain 1993) and middle-aged (Greece 2004) democracies.Thus, we have a more mixed situation in terms of the impact of the age of the democratic regime and level of development on LR recognition.

Table 1 .
Typology of countries in terms of age of the democratic regime and level of development Sources:

Table 2 .
Left-right self-placement (LR recognition) in five continents

Table 3 .
Left-right recognition of parties' location in the LR scale in five continents

Table 4 .
Explaining macro level variation in the overall level of LR recognition: self and parties' location on LR scale -OLS regressions Hong Kong collapsed because it is not a democracy; other cases collapsed due to missing data in the macro variables.

Table 6 .
1. Explaining cross-national variation in individual level determination of left-right recognition (self and two major parties)

Table 6 .
2. Explaining cross-national variation in self left-right recognition -2 nd Step of the Two Step Hierarchical Regression (Beta coefficients) -only for countries that have all independent variables at the individual level

Table 6 .
3. Explaining cross-national variation in parties' left-right recognition -2 ndStep of the Two Step Hierarchical Regression (Beta coefficients) -only for countries that have all independent variables at the individual level

bpsr Mapping and Explaining the Use of the Left-Right Divide "
political interest" and "age" on LR recognition in each country, and our independent variables (second step) are the macro-level items.Overall, what we can see in Table6.2 for LR self recognition is that there are only two significant macro-level factors.Freedom of press boosts the impact of "education" on LR self recognition.Countries' overall levels of education boost the impact of "D3 -age" upon LR self recognition: in countries with larger percentages of persons with a secondary education or more, the oldest cohort ("D3 -