H7 Electoral system
3.3 Methods employed
This dissertation employs a mix of quantitative (multivariate linear regression and factor analy-‐ sis) and qualitative (case study) methods, in addition to data collection techniques such as question-‐ naires and interviews.
Multivariate linear regression analysis. Bivariate linear regression analysis offers an estimate of the change in the dependent variable when the independent variable increases by one unit (Pollock 2006, 139). For example, when the impact of country-‐level political tolerance43 (tolerance toward the
least-‐liked group) on democracy is analyzed, the b coefficient (i.e., the slope of the regression, which
indicates the magnitude of the aforementioned effect) is 1.7. That is, for every one unit (standard devi-‐ ation) increase in the level of tolerance towards least liked groups, the model predicts, on average, a 1.7 units increase in the level of democracy (operationalized using Voice and Accountability scores). So if the average political tolerance of the citizens of country X is one unit higher than the political tolerance of the citizens of country Y, then country X will be more democratic with 1.7 units on the Voice and Ac-‐ countability scale.
It is seldom the case in social science to be able to offer a satisfactory and comprehensive ac-‐ count for variability in a dependent variable in terms of a single independent variable, and the above example is no exception. What is also known is that tolerance is usually positively correlated with de-‐ velopment and, in turn, development is positively correlated with democracy. That being the case, in order to obtain an unbiased estimate of the impact of tolerance on democracy, development must be included as a control variable in a multivariate regression model, even if the impact of development is not an object of interest.
Indeed, data analysis confirms these contentions. The two independent variables, degree of tolerance towards least liked group and Human Development Index, are highly correlated (0.33), and both are also highly and positively correlated with the dependent variable (0.42 and 0.72, respectively). When both independent variables are included as predictors for democracy, the estimated impact of tolerance (partial regression coefficient), which describes the contribution of this independent variable on the dependent variable, controlling for the impact of development, is only 0.81. This indicates that half of the (apparent) impact of tolerance in the original bivariate model was actually due to develop-‐ ment, rather than tolerance. Table 3.3 shows the relationship among the three variables and depicts the importance of introducing control variables.
Table 3.3. Bivariate Correlations
Democracy Tolerance
Tolerance .421*
Development .725** .330*
**Significant at p=0.01 level *Significant at p=0.05 level N=36
The above example has important substantive implications for this study. For instance, Chapter 4 presents a multivariate analysis of cross-‐national levels of democracy as a function of the average lev-‐ els of political tolerance and social tolerance in those countries, controlling for the level of socioeconom-‐ ic development.
Factor analysis. The goal of factor analysis is the identification of underlying dimensions among a number of variables. These dimensions, called factors, can be seen as “averages” of closely related variables (Lijphart 1999, 245). Often, the factors obtained using the initial factor extraction may be diffi-‐ cult to interpret. Consequently, most researchers use rotation, a technique that helps in obtaining fac-‐ tors that are easier to interpret. In cases such as the study when there are prior expectations about the number of factors, the most appropriate approach is using confirmatory factor analysis.
The comparative method is the “method of testing hypothesized empirical relationships among variables on the basis of the same logic that guides the statistical method, but in which the cases are selected in such a way as to maximize the variance of the independent variables and to minimize the variance of the control variables” (Lijphart, 1975, 164). This study benefits from the comparative meth-‐ od first because “the comparative method does not select its cases in random ways (as do experimental and statistical studies). Rather comparative studies unabashedly select their cases on the dependent variable” (Moses and Knutsen, 2007, 95). The ability to select the cases included in the analysis is con-‐ sidered one of the main strengths of the comparative method. It also allows the researcher to compare
es or most dissimilar cases. In general the shortcomings of this methodological approach are avoided by the researcher employing both deductive and inductive directions of determining the relationship be-‐ tween the variables.
Case study. Two important ways in which a variable like tolerance can be analyzed are multi-‐ case comparisons and single-‐case studies. Scholars such as Inglehart (1997; 2005) and Huntington (1991) looked at dozens of cases to explain cross-‐national differences in political culture or patterns of democratization. One major advantage of this approach is the possibility of testing hypotheses on a large number of cases, thus opening up the prospect of generalizing the results. However, while broad in scope, such studies tend to be short on details, lacking an in-‐depth perspective on processes and indi-‐ vidual cases. This is exactly one of the strengths of case studies, which offer the researcher the oppor-‐ tunity to conduct a fine grained analysis of a single-‐case and subsequently to offer a comprehensive ac-‐ count of the findings. Robert Putnam’s Making Democracy Work is one of the finest examples of using a multi-‐method approach to explain why the same institutions work well in the North of Italy but poorly in the South. Arend Lijphart is one of the most influential authors in comparative politics; however, the starting point for the development of his typology of democratic regimes (1984; 1999) was Politics of Accommodation (1968), a book on his native Netherlands.
Both multi-‐case comparisons and single-‐case studies are valuable tools for research. The two methods are complementary, rather than mutually exclusive. Often, a case study is only the first step in a long term research project, as it was, for instance, in Lijphart’s case. This is also the direction of this research. The main focus of this dissertation is the study of political tolerance and its micro-‐ and macro-‐ level determinants in general, not just in Romania. This study only uses Romania, a country whose citi-‐ zens are comparatively intolerant as previous research demonstrates (e.g. Viman-‐Miller and Fesnic
2010), yet it has not been studied extensively so far, as an exemplary case to develop and test hypothe-‐ ses which can be tested subsequently in other countries.
Quasi-‐experiment. Previous studies on immigration in Western Europe show that the newcom-‐ ers adopt a more democratic profile of citizenship. The final part of this dissertation seeks to add a new dimension to study of political tolerance – i.e., the impact of migration. Does temporary migration influ-‐ ence the level of political tolerance of migrants? In order to answer this question, this study will analyze both qualitative and quantitative data, comparing the tolerance of Romanians who have migrated to that of Romanians who have never left the country, seeking to isolate the independent effects of migra-‐ tion on attitudes. Based on the overview of the literature on socio-‐economic, demographic, and attitu-‐ dinal determinants of political tolerance, Figure 3.2 presents a recursive model of political tolerance de-‐ rived from my theoretical argument. The social and demographic characteristics of respondents are control variables, while exposure to the ideas and institutions characteristics for a more democratic so-‐ ciety is the intervening variable which increases the political tolerance of migrants.
Figure 3.2. Migration and political tolerance: a recursive model
This model is tested in Chapter 7 using data provided by an original survey of Romanian college students. This is a quasi-‐experimental design, and the survey was conducted in Cluj, Romania using stu-‐ dents from “Babes-‐Bolyai” University, a state university44 which is the largest institution of higher educa-‐
tion in the region of Transylvania, with over 50,000 students enrolled in 2008. Two groups were used, a “treatment” group and a “control” group. The first group included students who have traveled to the
44 To see how representative this sample is for the population of state university students in Romania, or for the broader stu-‐ dent population in the country, there are surveys conducted on nationally representative samples of students at our disposal. One such example is a survey conducted on two samples, one that is representative for students enrolled in Romanian state universities, while the other sample is representative for students enrolled in Romanian private universities. (Direcţia pentru
US with the Work and Travel program45 and the second group included students who have never trav-‐
eled to the US.
The survey. The practical means of implementing the survey were self-‐administered question-‐ naires, and the setting was group administration in classes with large enrollment. The cost of this ap-‐ proach was very low and the completion rate was near 100%, and these were major advantages under conditions of limited time and resources (Johnson and Reynolds 2008, 303). Moreover, unlike face-‐to-‐ face interviews, self-‐administered questionnaires facilitated asking sensitive questions (Johnson and Reynolds 2008, 318) and alleviated the pressure to give “socially desirable” answers (Traugott and Price 1992, 246).
To address the problem of an absence of a pre-‐test, multiple items in the questionnaire asked the respondents to self-‐assess their position at the moment of the questionnaire administration com-‐ pared to what it was before their American experience. The survey used a combination of questions to assess students’ political tolerance before and after their involvement in the Work and Travel program. This approach is not novel; social psychology and political psychology research often rely on information obtained from surveys using recall. Ansolabehere and Iyengar (1995) employed the recall method and researched the level of influence that negative advertisement had on voter turn out. They found that negative advertisement tends to depress the voter turnout. Later, Wattenberg and Brians (1999) found that negative campaign advertisement stimulates voter turnout. They based their research on the same methodological approach as their predecessors. Druckman et. al (2011) debate thoroughly the pros and cons of subjects’ memory recall in applied social sciences. They find that in experimental political sci-‐ ence recall cannot be avoided and that researchers must pay more attention to data interpretation.
They make a distinction between recall and recognition (100). A similar approach is also used in political sociology, especially in studies of voting behavior, where survey respondents are often asked how they voted in the last election. Lizotte, Lodge and Taber (2005) argue that results from emotional recall can-‐ not be trusted if they are obtained from a format that asks direct questions. However, their attempt to demonstrate this experimentally was unsuccessful. Table 3.4. presents a summary of the research de-‐ sign, comparing current political tolerance of the treatment group with that of the control group.
Table 3.4. Assessing the impact of Work and Travel experience: a quasi-‐experiment Time:
Group:
t – 1
(before Work and Travel): t: Work and Travel (“Treatment”?) (present): t +1
“Control” Tolerance C, t-‐1 No Tolerance C, t +1
≈ ≠
“Treatment” Tolerance T, t-‐1 Yes (democratic expo-‐
sure & learning) Tolerance T, t +1
A major advantage of surveys is the fact that they provide a large number of cases which enable multivariate analyses – here, testing the impact of migration on tolerance using background and attitu-‐ dinal variables as controls. Survey data are helpful to answer the “if” question, but is less helpful to an-‐ swer the “why” question. If statistical analysis indicates that migration does have an impact on toler-‐ ance, it is still to be determined why that is the case, and the processes through which greater exposure to a democratic culture via migration leads to a change of the migrant’s level of tolerance. Thus, a mul-‐ ti-‐method approach was necessary, one that included both quantitative and qualitative methods.
Interviews. To complement the data provided by the surveys, twenty systematic semi-‐structured interviews were conducted, trying to get a more in-‐depth understanding of the processes that may lead to an increase in tolerance, and also to assess whether such an increase has indeed occurred in the first
place. The richness of interview data collection allowed a better understanding of the impact of toler-‐ ance at the individual-‐level. These were “nonscheduled standardized interviews,” defined by Gray et al. (2007, 161) as a method by which “all questions are asked of each respondent, but they may be asked in different ways and in different sequences.” According to Manheim and Rich (1995, 162), the central goal of this kind of interviewing is not so much “the collection of prespecified data, but the gathering of information to assist in reconstructing some event or discerning a pattern of specific behaviors” – in this case, how migration affects tolerance. Among other advantages of qualitative interviews is the fact that it allows the interviewer to become more personal and gain the trust of the interviewees. This allows and encourages introspection from the respondents (Gray et. al 2007). For this study, respondents who were previously enrolled in Work and Travel programs were selected, discussing with them how this experience has changed their political and social tolerance.
To minimize any bias in the selection of respondents for these interviews, every student previ-‐ ously enrolled in the Work and Travel program who filled the questionnaire was asked if he or she was willing to be interviewed. Given the time and financial constraints, on the one hand, and the relatively low number of people in this sample with previous Work and Travel experience, on the other hand, it was unrealistic to try to interview more than thirty people. The expectation was that, if at least half of those from whom an interview were requested would answer affirmatively, taking into account the goal of having at least one hundred students with Work and Travel experience filling the questionnaire, this should have also ensured about twenty interviews. In actuality, the total number of students from the total sample (N = 1,514) with at least one Work and Travel experience was 129 (8.5%), and the number of interviews completed was 20. The main point of these interviews was to get additional insights on the respondents’ experience in the US (qualitative information), in addition to the quantitative infor-‐ mation provided by survey data. The interviews were not meant to be representative, but they were
designed to provide important additional contextual information that could help illuminate patterns in the survey data.