Schools of Thought and Economists' Opinions on Economic Policy

Transcription

1 University of Macerata From the SelectedWorks of Luca De Benedictis Winter September, 2014 Schools of Thought and Economists' Opinions on Economic Policy Luca De Benedictis Michele Di Maio Available at:

2 Schools of Thought and Economists Opinions on Economic Policy Luca De Benedictis University of Macerata Michele Di Maio University of Naples Parthenope Abstract In this paper we bring to data the hypothesis that economists opinions are related to the schools of thought they declare to belong to. Two are the main results of our analysis, based on a novel dataset on Italian economists opinions. The first one is that, on average, differences in the School of Thought predict differences in economists opinions on a large set of normative questions on economic policy, even controlling for individual and other group or community characteristics, spatial and knowledge heterogeneity, and political orientation. Second, the way of grouping together different schools of thought is crucial as far as their usefulness to predict economist opinions: Dichotomous groups (e.g. Mainstream vs Non-Mainstream, Orthodox vs Heterodox) have poorer explicative power than more articulated groups that preserve the differences between schools of thought. Keywords: School of thought, Economists disagreement, Group identity, Economic policy, Italy JEL Classification: A11, A13, C42 DIEF - University of Macerata, Via Crescimbeni 20, Macerata 62100, Italy. Ph: ; Fax: debene@unimc.it DSE, University of Naples Parthenope, Via Medina 40, 80133, Naples, Italy. michele.dimaio@uniparthenope.it. We would like to thank Frederic Lee, Cristina Marcuzzo and Thomas Mayer for helpful comments and suggestions on a previous version of this paper. We would also thank all Italian Economists that participated to our survey and that answered to our follow-up questions. All the supplementary material on the Survey of Italian Economists can be found at the web page entirely our own. All errors and omissions are 1

3 1 Introduction The proclivity of human beings to gather together in communities is a key feature of human social behavior, as well as a fundamental component of human cognitive system. We, as humans, tend to cluster in groups, and in doing so we tend to separate our own community from other communities. 1 The same procedure of grouping and separating that we apply to our social organization we also apply to learning, to the way we evaluate things, and to the explanation of individual behavior. In this paper we analyze the role of communities in a special group of people involved in intellectual production: Economists. In particular, we evaluate whether the difference in the School of Thought (SofT, hereafter) relates to the difference in their opinions about normative issues on economy policy. This empirical evidence comes from a representative sample of academic Italian economists answering to a large set of questions concerning their views on the functioning of the economy and about economic policies to be implemented to improve Italian economic conditions. The main result of the analysis is that differences in the SofT predict differences in economists opinions. The result is robust to the inclusion of several individual and community or group characteristics: Controlling for confounders, such as spatial and knowledge heterogeneity and political orientation, for potential group alternatives, such as geographical common origin, common intellectual environment as undergraduates, and common University department as scholars, SofT remains significantly correlated with the difference in economists individual opinions on economic policy. This should not come as a surprise. For hundreds years economists belonging to different SofT argued about theory and policy (Gide and Rist, 1909; Hutchinson, 1953; Blaug, 1980), giving rise to new ideas and intellectual communities (Dow, 2007). 2 All this is well reported, discussed, and classified by historians of economic thought. To exemplify, let s take the volume by our Italian colleagues Screpanti and Zamagni (2005) as a test-bed. In their Outline 1 See Akerlof and Kranton (2000) on the formation of identity and economics and Bisin and Verdier (2011) on group evolution, socialization and cultural transmission. The distinction between Us and Them has been derived by Eaton, Eswaran and Oxoby (2011) as an efficient strategy in an evolutionary game, where groups of people exploit a common access resource, as in primitive hunter-gatherer teams in the savannas of east Africa tens thousands years ago; group identity may improve market transactions in Prisoners Dilemma like situations (Bowles and Gintis, 2004) as a result of parochial altruism. 2 Economists have disagreed about almost everything. For a long time they also have argued about what economics was about and still today economists are far from unanimous about the definition of their research object (Backhouse and Madema, 2009). 2

4 of the History of Economic Thought the word debate appears seventy-five times, one every seven pages, while the stronger term controversy appears eighteen times, one every thirty pages, being the benchmark-term economics in one out of two pages. Of course, debate is not uniformly distributed along pages and time, but tends to catalyze in specific moments. Famous examples are the Methodenstreit between the German Historical School and the Austrian School at the end of the 19th Century (Anderson et al., 2002; Roncaglia, 2005) or the Capital Controversy between the two Cambridge at the end of the 1960s (Harcourt, 1972), and the most recent debates on rational expectations (Pesaran, 1987) or the behavior of individual economic agents (Kahnneman and Tversky, 1979). The procedure of grouping and separating in the evolution of intellectual production in economics reaches our days, the debates on the use of randomized trials in policy evaluation and of laboratory experiments in guiding utility theory being two examples. Despite the ease of representing debates as the confrontation of polar positions, 3 rarely the competition between SofT ends in a complete prevalence of one school over the other(s). In fact, in the majority of cases reported by historians of economic thought, economic debates reached their climax with the prevalence of a mainstream coexisting with alternative views. Pluralism seems to be the norm rather than the exception. Many authors have argued that this is a blessing and that theoretical pluralism should be considered as a value per sé and that it should be guaranteed and preserved (Maki, 1997). In fact, there are good reasons to think that pluralism foster innovative ideas and that critical thinking acts as a system of checks and balances over economic mainstream, on its internal logic, methodological soundness and ability to cope with real economic problems. 4 3 To give an idea of the popular representation of a debate among economists, the resurgence of the Keynes vs von Hayek confrontation on the role of government intervention to boost effective demand have been recently proposed in a video clip by econstories.tv ( fight-of-the-century-music-video/). 4 In spite of the twenty-three SofT listed in Eatwell et al. (1987) and the report by Dow (2000) that the number of international associations of heterodox economics is steadily growing (as Lee (2008) quantified, it has reached the noticeable number of twenty-four), many worry about the unfair power distribution in the discipline (George, 2007) and the constant risk that an overwhelming influence of the mainstream over alternative approaches could reduce pluralism in economics (Schiffman, 2005). Moreover, as argued by Axarloglou and Theoharakis (2003), not only SofT still guide the research activity of many economists but it also influences journals quality perceptions, the evaluation of individual research and the individual notoriety of economists as public advisors. For an empirical analysis of the differences between mainstream and heterodox economists see Di Maio (2012). 3

5 While historians of economic thought report the evolution of the different intellectual communities along the time dimension, recently a different group of scholars started to explore and to measure the relevance of disagreement among economists on a cross sectional dimension. Kearl et al. (1979) were the first to recollect informations on economists opinions and compare them through distance metrics. Using a survey of U.S. based economists, they find that consensus, measured through an entropy index, was higher about microeconomic issues rather than macroeconomic ones, and that consensus was also higher on positive rather than normative issues. Yet they argue that the apparently common perception of disagreement among US economists on theory and economic policy was wrong. After Kearl et al. (1979) many other surveys followed (see Fuchs et al. (1998) and De Benedictis and Di Maio (2011) for a synthesis of the literature). 5 Most of them confirm that economists express consensual opinions more often that what one might expect. Yet disagreement turns out to be sizable when economic policy issues are considered. As far as the possible sources of disagreement, some of them appear to be relevant time and again across different surveys. The first one is spatial heterogeneity: Frey et al. (1984), and others after them, find a significant difference between the view of economists belonging to different countries. 6 They attribute these difference, among other reasons, to difference in history and culture associated with geography. The second one is knowledge heterogeneity, related to different research field or research topic of the economists included in the sample. Here the evidence is less clear cut. On the one hand, specific competence seems to orient economists opinions also on general economic policy issues, even if the latter are not strictly related to their field of research (Fuchs et al., 1998). 7 On the other hand, Fuchs (1996) shows 5 Alson et al. (1992) conduct a large-scale survey of US economists in all fields asking a subset of the Kearl et al. (1979) questions. They also conclude that the consensus on positive economics is considerable. The Alston et al. (1992) propositions have been then used in a number of other studies, becoming a sort of benchmark. For instance, a subset of Alston et al. (1992) questionnaire is used in Becker et al. (1994) to compare the views of economists, economic educators, teachers, and journalists and by Fuller et al. (1995) to evaluate the opinions of the delegates at the 1992 American political conventions. Fuller and Geide-Stevenson (2003) updates a subset of Kearl et al. (1979) questionnaire. 6 A subset of Kearl et al. (1979) questionnaire is used by Frey et al. (1984) in a survey of economists from four European countries (Austria, France, Germany, and Switzerland) and by Ricketts and Shoesmith (1992) in a survey of UK economists. Their results suggest the existence of significant differences in the opinions between European, UK and US economists. The existence of small but significant difference between American and European economists has been confirmed by Frey and Eichenberger (1993), Mueller (1995) and Aiginger et al. (2001). 7 Some evidence comes from the surveys directed to economists active in specific research fields: Economic 4

6 that, among 50 leading health economists, the disagreement regarding positive questions more related to theory seems to play no role in explaining differences concerning opinions on policies. The third one is the heterogeneity in individual characteristics. Here also the evidence is mixed. Fuchs et al. (1998), surveying labor and public policy economists active in top US departments, find that individual characteristics are correlated to differences in the opinion about policy prescriptions. On the contrary, Caplan (2001) shows that disagreement among a large sample of US economists is not related to their individual characteristics. The fourth possible source of disagreement is heterogeneity in individual political orientation. Most of previous contributions have shown that the political affiliation of economists, expressed in terms of left-wing or right-wing orientation, or in some cases, more precisely matched to the political landscape of their country of origin, is an important source of variation with respect to their judgment about policy prescriptions (Fuchs et al. 1998; Klein and Stern, 2006, 2007; De Benedictis and Di Maio, 2011), even after controlling for other individual characteristics. The role of ideology in economics is well recognized (albeit controversial) at least since Schumpeter s 1948 Presidential address to Annual Meeting of the American Economic Association (Schumpeter, 1948). 8 Ideology matters in economic analysis whenever the inseparability of efficiency and social justice is acknowledged. This is a condition that is frequently met, and here is where the political economy perspective in welfare economics acquires its relevance, and where economists can make their views on government intervention and on economic policy conditional on their political views. Saint-Paul (2011) includes this assertion into a theoretical model where a self-interested economist, knowing historians (Whaples, 1995), labor economists (Whaples, 1996), European and American industrial organization economists (Aiginger et al., 2001), health economists (Fuchs, 1996), resource economists (Randall, 1985), public and labor economists (Fuchs et al., 1998). A direct comparison of the surveys results is of difficult interpretation, being the sample of the economists included in each survey different on several grounds; but, since many of the comparable questions come from the Alston et al. (1992) questionnaire, a simple confrontation of the results confirms that economists working on specific research topics tend to have different orientations on certain economic policy issues. Frey et al. (2009) report that while more than 40% of German speaking economists do not think that there is agreement on the most relevant issues in economics, the percentage significantly decreases when the sample is restricted only to academic economists. This suggests that differences in opinions may also be related to the type of activity (academic, policy advisor, public administration) the respondent is involved in. This possibility is ruled out in our analysis since our sample only includes academics. 8 See also Heilbroner (1993) and Steen (2004) on Schumpeter and on his view of the interplay between economics and ideology. 5

7 the true structural model of the economy, reports an ideologically-biased one so as to influence outcomes according to his own political orientation. 9 In equilibrium, a conservative economist will propose a model with a lower structural Keynesian multiplier, and a greater long-term inflationary impact of output expansions. A progressive economist will do the opposite. Therefore, both empirical evidence and theoretical analysis suggest that ideology is related to individual opinions. In this paper we add to this list another possible source of disagreement, namely the SofT, considering it as a particular kind of community heterogeneity. Apparently, even though it seems a natural issue to explore, previous research on economists disagreement did not pay any attention to the quantification and the significance of the effect that belonging to a different SofT may have on the difference in opinions between economists. Obviously, as we have seen, this does not mean that the characteristics and role of the SofT in economics have not been explored before. In fact, the literature discussing the differences across SofT is quite substantial, as it is the one on how their evolution might have an impact on the profession 10 What is indeed novel in our analysis is the attempt to empirically validate the assertion, coming from historians of economic thought, that SofT are meaningful communities with respect to economists opinions, also as far as broad policy issues are concerned, controlling for other aspects that could confound SofT effect. To best of our knowledge, this is the first paper that makes a bridge between the insights coming from the history of economic thought and the literature on economists disagreement, and it is the first attempt to quantify the role of the SofT in determining differences in economists views on the functioning of the economy and on the adoption of specific economic policies. The paper proceeds as follows. In section 2, we describe the survey used as the primary information source for the data used in the analysis, and we compare the individual characteristics of the Italian economists belonging to different SofT. In section 3, we evaluate - con- 9 In Saint-Paul (2011), the ideologically-biased model influences the expectations of both the people and the government. The latter tries to stabilize an unobserved demand shock and makes different inferences about that shock depending on the model the economist is proposing, the former updates its expectations on both the model and the government action. See also Bénabou (2008) along this line of research. 10 Actually, the importance the profession has given to methodology is one of the aspects that traditionally differentiates European and American economists (Frey et al. 1984; Aiginger et al. 2001). In their classical study, Frey and Eichenberger (1993) argue that one of the specificities of European economists is considering the belonging to a particular SofT as important. Moreover, there is evidence that the Europeans are more open toward heterodox views (Frey and Frey, 1995). 6

8 trolling for spatial heterogeneity, knowledge heterogeneity, individual characteristics, political orientation and several possible communities economists are or were part of - if economists belonging to different SofT do express different views on a set of propositions. Next we consider if grouping economists as polar positions, such as Mainstream vs Non-Mainstream, Orthodox vs Heterodox, still allows to predict the disagreement between economists. Section 4 summarizes our results and concludes. 2 The survey, the sample and the schools of thought in the Italian academia The data we use come from an on purpose on-line survey. 11 In 2007 we asked Italian economists to express their opinions concerning the Italian economic situation. 12 The survey included economists employed in Italian Universities, members of the Italian Economic Association (SIE), members of the International Association of Italian Economists (AIDEI), and participants to major economic conferences held in Italy between 2005 and The questionnaire was composed of 32 questions: 13 6 of them were on the respondent s demographics, 2 were on her/his research field, 6 were about the determinants of the Italian economic decline, 6 (with 40 sub-questions) were on the causes of the current (i.e. 2007) Italian economic conditions and 2 (with 18 sub-questions) were on the policy proposals to put in place to improve Italian economic performance, 5 were on the quality and the effectiveness of the debate on the Italian economy, and 4 were on the respondent s personal view (her/his position in a hypothetical market-vs-state spectrum, attitude toward social mobility, political stance and level of optimism). Finally, there was a question asking the respondent to selfdeclare the SofT she/he belongs to. 11 The details of the survey design are reported in De Benedictis and Di Maio (2011) and the full description of the survey can be found in the webpage surveyofitalianeconomists. 12 The time of the survey turned out to be quite appropriate: 2007 was one year before the starting of the international financial crisis and Italy was in a temporarily stable state both in terms of economic and political conditions, after years of political turmoil and sluggish economic growth. This is crucial in terms of the reliability of the information collected. In general, during periods of high economic distress all respondents tend to express similar (negative) opinion on the (poor) state of the economy. 13 In the advisory notes to the questionnaire it was clearly stated that no specific knowledge was required to answer any question, that no individual statement would have been revealed, and that discretion on the anonymity of respondents would have been guaranteed. 7

9 Four hundred and ninety-six economists responded, a reply rate of 33%. 14 Since for our analysis we want to rely on a representative sample of Italian economists, we exclude from the analysis all individuals coming from a population that could not be enumerated. Therefore, we consider only those respondents that are employed by the Italian Minister of University and Research (MIUR), excluding all Italian economists working abroad, all post-docs and all non-academics. The final sample includes 335 respondents. As in any survey of this kind, respondents self-select themselves, but there is no evidence that the self-selection is somehow related to the object of the analysis. 15 We therefore assume that selection can be taken care of in regression analysis by adjusting for differences in observed covariates. Moreover, to make the sample representative of the entire population we post-weight the data according to the three strata (gender, academic position and region of work) that we could observe in the MIUR dataset To give a comparison, the response rate of American Economic Association (AEA) members in previous surveys are: 34% in Alston et al.(1992), 31% in Fuller and Geide-Stevenson (2003), 36% in Whaples and Heckelman (2005), 27% in Klein and Stern (2006) and 40% Waples (2006). Usually surveys of economists within subfields tend to yield higher response rates: for example 41% of AEA labor economists in Whaples (1996). 15 The ideal comparison between respondents and non-respondents would imply the possibility to compare the opinion of an individual when responding to the questionnaire with the opinion of the same individual when not responding. However the two events cannot be observed contemporaneously. In our case, we cannot follow Rosenbaum and Rubin (1983), assuming that, conditional on a set of covariates, the assignment to the group of respondents is a-priori independent on being a respondent. Since we do not have any information on the characteristics of non-respondents we could not rely on any matching strategy ensuring a reliable conditional strategy. In order to exclude that non-respondents are systematically different for respondents we can only rely on indirect evidence. To this end, we test this hypothesis using the timing in the response to the questionnaire by the respondents. The survey was on-line during one month and we sent a reminder at the beginning of each week. If the fact that respondents and non-respondents might have fundamentally different opinions can be represented as a smooth process, this would imply that individual average opinion would change smoothly depending on the time of response. Therefore, we group respondents in four groups according to the week in which they filled the questionnaire and we test for the difference between means from the four groups. We find no evidence of any difference. We also reject the hypothesis of a differences in the variances of the four groups of respondents. Occordingly, we infer that this absence of difference will extend also to non-respondents. 16 The procedure of post-stratification weighting was based on the reciprocal of the probability of inclusion of each individual (Groves et al., 2004). Given the identity n ijk p ijk = N ijk, where n is the number of individuals in the survey sample, N is the number of individuals in the total population (MIUR), and i = 1, 2, j = 1,..., 4 and k = 1,..., 19 stand for gender, academic position, and region of work. The corrective weight p of the generic stratum ijk is simply the reciprocal of the probability of inclusion. In case of empty cells (N ijk > 0 8

10 2.1 Distribution of Italian economists across schools of thought To minimize errors and misinterpretations, when asking to self-report the SofT they belong to, the respondents had to choose from a given list of SofT. To guarantee at the same time the maximum freedom in expressing their methodological orientation we allowed the respondents to also include an open definition of the methodology that better described their orientation. Only fifteen economists chose a personal definition. Being these very few, we had to reclassify these specific responses according to similarity (e.g. the indication of being a Marxist-Leninist is similar to Marxist, a SofT that was included among the original options). 17 The list of SofT is reported in Table 1. The list is very similar to the one in Axarloglou and Theoharakis (2003): The main difference is that we also included the Eclectic category. Moreover, all the SofT discussed in Colander et al. (2003) are included in our list. 18 Table 1 reports the distribution of Italian economists across SofT. Almost 1/3 of the economists in our sample defines themselves as Eclectic. The other two larger groups are the Neoclassical/Mainstream (18%) and the I do not refer to any specific SofT group (hereafter No-school) (17%). If we sum this latter group and Eclectic we see that almost half of the sample is not categorized in any of the traditional SofT. Among these, after Neoclassical/Mainstream, the larger are Post-Keynesians (11%) and Neo-Keynesians (8%). Even if much less relevant as percentage, our sample also features Marxists/Sraffians, Evolutionary and Behavioral economists, Institutionalists, Regulationists and Austrian/Neo-Austrian economists. These numbers indicate that the Italian academia is characterized by the presence of a substantial variety of SofT, and that the number of economists belonging to each school is not uniform. 2.2 Schools of thought, ideology, pro-market orientation and age On a bivariate dimension, it is possible to rank SofT according to the relative position they hold in relation to some of the characteristics of the respondents. The most basic distinction and n ijk = 0), we imputed them using a nearest neighbor procedure. Since gender was the less distorted dimension of the sample, when necessary (16 cases out of 335) we imputed i = 1 for i = 2 and vice versa. 17 This procedure does introduce a (minimal) level of arbitrariness in the interpretation of individual opinions. In Section 3.3 we show that the re-classification of these observations does not distort the results of the analysis. 18 While we are aware of the distinction discussed in Colander et al. (2003) between Mainstream and Neoclassical approach, we preferred to consider the two together, to minimize the possible confusion by the respondents who were not informed about the content of the debate. 9

11 Table 1: The distribution of Italian economists across schools of thought SCHOOL OF THOUGHT Number of respondents % Eclectic Neoclassical/Mainstream I do not refer to any specific SofT Keynesian/Post-Keynesian Keynesian/Neo-Keynesian Institutionalist/Neo-Institutionalist Evolutionary Marxist/Sraffian/Neo-Marxist Behavioral Regulationist Austrian/Neo-Austrian Total NOTE: The table includes the responses to the question: How will you define your methodological orientation? Multiple answers were possible: respondents that indicated three or more options (7 cases out of 335) were added to Eclectic; double responses (32 cases) were randomly assigned to one of the two options indicated; specific responses (15 cases) have been reclassified according to similarity (e.g. Marxist-Leninist have been classified as Marxist). between SofT is notably the one based on ideology. As pointed out by Schumpeter (1948) and more recently, in different perspectives, by Bénabou (2009) and Saint-Paul (2011), ideology influences the policy outcome that individuals want to see occur, and people might obtain expressive utility from confirming their ideological orientation. To appreciate whether and how SofT differ as for ideology, we project the ideological orientation of the different SofT (as indicated by the political opinion expressed by the economists that belong to each SofT) on the left-right spectrum. The black dots along the horizontal red line in panel (a) in figure 1 depict the average political opinion in each SofT. Marxist are more left-wing oriented and Mainstream and Austrian are more (relatively) right-wing oriented. All SofT are concentrated on the centerleft side of the political spectrum (as shown by the shaded Kernel density of the Italian economists political opinion). The majority of Italian economists is center-left (57%), and only a small minority describes itself as center-right (7%) or right (3%). What is of even more general interest is that SofT (at least the most representative ones) do not cover the entire left-right spectrum in the ideological space. This means that SofT and ideology do 10

12 not overlap and must not be considered as collinear. Furthermore, some SofT are barely distinguishable on an ideological dimension: Marxist and Behavioral economists express the same political orientation, on average, as Evolutionary and Neo-Keynesian economists do. [Figure 1 about here] In panel (b) of figure 1, we project the SofT on the State-vs-Market space. The black dots along the horizontal red line visualize the average opinion in each SofT about the mechanism that should drive the allocation of resources in the economy. Marxist are again more extremely oriented, being more in favor of a stronger presence of the State, while Mainstream, Austrian and Behavioral economists are more pro-market oriented. Remarkably, all schools of thought are concentrated around the Mixed Economy position. The Kernel density is not bell-shaped anymore, showing a tenuous bimodality. The majority of Italian economists is in fact oriented toward a Mixed Economy (45%) or toward a more relevant role of the Market (38% in the less extreme version, and 12% in the most radical version of a Laisez-faire economy). Only a small minority is in favor of a centrally ruled economy (2%). Again, what is kind of a surprise is that SofT, when they are represented by the opinion of their supporters, do not cover the entire State-vs-Market spectrum. As for ideology, the two dimensions, the belonging to a SofT and the pro-market orientation, are not collinear. Finally, since SofT appeared and became prevalent in different times, they could be somehow correlated with the age of economists that received their under-graduate and postgraduate education in different time periods. The evidence, reported in panel (c) in figure 1 is much less clear cut than what was a-priori expected. As in the two previous panels, the black dots along the horizontal red line visualize the average age of economists in each SofT. The majority of Italian economists is concentrated in the two cohorts (32%) and (29%) and the Kernel density is more uniform than in the two previous cases. Mainstream are relatively younger, while, surprisingly, Behavioral economists are the oldest, and Marxist and Austrian economists have the same age. Interestingly, the No-school group is not on the young fringe, bringing some evidence that the lack of interest toward the methodology of research in economics is not a phenomenon strictly related with the age of economist. Young economists seem to be more orthodox than disinterested. Again, the two dimensions, the belonging to a SofT and the age of the economist, are not collinear. This evidence, together with the one concerning ideology and pro-market orientation, calls for a multivariate analysis of the role of SofT. 11

13 2.3 Are economists belonging to different schools of thought different? Are economists belonging to different SofT different? To answer this question on a multivariate dimension, we run a multinomial logistic regression of the respondent s SofT on all the covariates of interest (that we indicate with capital letters). 19 To take into account the possible common effect of having studied in the same University we also include fixed effects for the Universities were the respondents graduated from. 20 Table 2 reports the estimated multinomial logit regression coefficients. 21 Beginning with the variable AGE (an ordered variable with five categories from younger to older), results show that Mainstream economists are, conditionally on other covariates, significantly younger than Neo-Keynesian and Marxist, with the latter including the elder economists on average. As for GENDER (a dummy variable which takes value 1 if the respondent is a male), Neo-Keynesian, Marxist and No-school show a significant larger number of males with respect to Mainstream. Next there are three variables related to if and which type of post-graduate degree the respondent holds. 22 The variable MASTER (a dummy taking value 1 if the respondent has a Master in Economics) indicates that Institutionalists are less likely to have a Master degree with respect to Mainstream while respondents belonging to any other SofT are not. As for DOTTORATO (a dummy taking value 1 if the respondent holds a Dottorato) 23 it turns out that Marxists and No-school are more likely than Mainstream respondents to have this post-graduate degree. On the contrary, the coefficient of the variable PhD (a dummy which 19 We are forced to exclude Austrian, Behavioral and Regulationist economists from our analysis since they are too few to guarantee the necessary within-group variability for estimation. 20 We include a fixed effect for any University with more than 10 respondents in our sample, while the Universities with less than 10 respondents are grouped together in a residual category in order to make the estimate of the model to converge. The Universities with more that 10 respondents are: Bocconi University in Milan, Catholic University in Milan, La Sapienza University in Rome, Bologna University, University of Florence, University of Naples Federico II, University of Pavia, and University of Turin. 21 The interpretation of the multinomial logit coefficient is that for a unit change in the predictor variable, the logit of the relevant SofT relative to the Mainstream is expected to change by its respective parameter estimate given the variables in the model are held constant. For instance, if a subject were to increase her ACADEMIC POSITION from one category to the subsequent, the multinomial log-odds for being an Eclectic with respect to being a Mainstream would be expected to decrease by 0.53 unit, while holding all other variables in the model constant. 22 Including them in our analysis allows us to control for the fact that economists may in fact self-select into different SofT according to their preexisting opinions. 23 Dottorato is the highest post-graduate degree offered by Italian Universities. We include it as a separate control to differentiate between respondents that completed their post-graduate studies in Italy or abroad. 12

14 takes value 1 if the respondent holds a PhD, i.e. she/he completed her post-degree studies abroad) shows that Marxist and Neo-Keynesian are significantly less likely to hold a PhD with respect to Mainstream economists while this is not true for respondents belonging to other SofT. When we look at the ACADEMIC POSITION (an ordered variable, with four subsequent categories from Assistant Professor to Full Professor), it turns out that, ceteris paribus, Mainstream are in a higher hierarchical position with respect to Eclectic, Institutionalist and Neo-Keynesian. As for RESEARCH (an ordered variable, with five categories from pure empirics to pure theory) there are no large differences across SofT, with Eclectic and Noschool resulting to be more empirically oriented than Mainstream. On the contrary, as for MARKET (an ordered variable, with five categories from the more market-oriented to the less market-oriented) 24 there are important difference between Mainstream and all the other economists. Actually, Mainstream are by large more market-oriented than other economists. Not surprisingly, the largest difference is again found with respect to Marxist and Neo and Post-Keynesian. Interestingly, there are no significant differences across SofT as for the role that the Government should play in favoring social mobility (SOCIAL MOBILITY). Instead looking at the political variables RIGHT (an ordered variable, with five categories from the more left-wing to the more right-wing oriented), we find that ceteris paribus only Post- Keynesians are significantly more leftist than Mainstream. 25 There seems not to be large differences as for OPTIMISM, with only Institutionalist and No-school economists being significantly more pessimistic than Mainstream. The result that economists belonging to different SofT are significantly different has some interesting consequences. For instance, it implies that it would be an error to interpret the disagreement across economists belonging to different SofT as a measure of the disagreement between schools. Indeed, our results indicate that individual characteristics have to be controlled for if one wants to correctly evaluate the role of the SofT in explaining economists disagreement. This is precisely what we do in the next section. 24 In this case the respondents were answering to following question: Which allocative mechanism you consider the most appropriate in the organization of the economic activity?. There were five possible alternatives, with the State (classified as 1) and the Market (classified as 5) at the extremes. 25 The point estimate of Marxist, as it is natural to expect, is the highest (with a negative sign), meaning that this SofT is a the left hedge of the political spectrum of the various SofT. At the same time, also the standard errors are very high, meaning that the variability of the political stance of Marxists is also (relatively) high, so to make the point estimate not statistically different from zero, ceteris paribus. 13

15 Table 2: Differences in individual characteristics across schools of thought Baseline SofT category Eclectic Inst. Neo-Keyn Post-Keyn Marxist Evolut. No-school Mainstream AGE ** ** (0.422) (0.564) (0.503) 0.574) (1.710) (0.477) (0.418) GENDER ** ** ** (0.597) (0.922) (0.966) (0.779) (9.332) (0.942) (0.752) MASTER ** (0.542) (0.951) (0.852) (0.746) (1.210) (0.920) (0.620) DOTTORATO ** * (0.740) (1.149) (1.123) (1.066) (3.809) (1.109) (0.832) PhD ** * *** (0.596) (0.995) (0.962) (0.843) (3.476) (0.842) (0.659) ACADEMIC POSITION * * *** (0.302) (0.419) (0.446) (0.457) (0.715) (0.513) (0.387) RESEARCH ** *** (0.236) (0.299) (0.331) (0.316) (0.646) (0.335) (0.282) MARKET *** ** *** *** *** ** * (0.423) (0.556) (0.529) (0.612) (0.992) (0.522) (0.456) SOCIAL MOBILITY (0.340) (0.402) (0.391) (0.369) (1.399) (0.666) (0.389) RIGHT * (0.372) (0.599) (0.463) (0.527) (1.505) (0.557) (0.405) OPTMISM ** *** (0.351) (0.507) (0.470) (0.435) (0.505) (0.525) (0.420) NOTE: The table reports multinomial logistic regression coefficients. The dependent variable is SofT.,, mean significant at the 1%, 5%, 10% respectively. The reference category is Mainstream. A constant and a fixed effects for the University were the respondent did graduate are included but the coefficients are not shown. Robust standard errors are in parenthesis. The regression is weighted, using the procedure in Groves et al. (2004), see also footnote 16 in the text. # of observations is 238. Pseudo-R 2 = 0.31, P rob > χ 2 = Mean values (Standard Deviation) for the control variables for the reference category, Mainstream: AGE 1.96 (0.93); GENDER 0.80 (.41); MASTER 0.52 (0.50); DOTTORATO 0.43 (0.50); PhD 0.44 (0.50); RESEARCH 3.30 (1.16); ACADEMIC POSITION 4.06 (1.05); MARKET 4.06 (0.72); SOCIAL MOBILITY 4.20 (.91); RIGHT 2.40 (.88); OPTIMISM 3.46 (.81). 3 Schools of thought in action: The Italian case To test the hypothesis that differences in economists opinions on normative issues are related to respondents belonging to different SofT, we focus on the two core questions in the questionnaire. The first one is Do you think that THIS is a cause of the unfavorable performance of the Italian economy?, followed by a pre-defined list of forty broad economic, social and political issues considered in academic or general debates as possible causes of the 14

16 Italian economic conditions, such as The adoption of the Euro or Labor Union behavior. The second question is Can IT be effective to revive the Italian economy from the current phase of difficulty?. Also this question was followed by a list of eighteen proposals, all expressed in terms of broad policies to be implemented, such as the opportunity To proceed with more liberalizations or Funding academic research. The complete list of what we will call from now on Causes and Policy Proposals is included in the Appendix. 26 Economists opinions on these forty Causes and eighteen Policy proposals are our main object of analysis. For each of the 58 propositions we separately estimate E(opinion i SofT i, c i ), where opinion i is the opinion of individual i on the proposition, SofT i is the categorical variable indicating the SofT of the individual i and the vector c i denotes the set of control variables that we hold fixed while evaluating the effect of SofT i on the expected value of opinion i. From our analysis in section 2.3, we know that it is indeed crucial to control for the differences in the characteristics of individuals to capture the effect of SofT i. Still, if an individual or community characteristic, not included in the regression, affects both the SofT the economist declares to belong to and the economist s opinions the regression would suffer from an omitted variable bias. While we cannot in principle exclude this possibility, we include in vector c i all information we could collect concerning individual characteristics, spatial heterogeneity, knowledge heterogeneity, political orientation, and other possible community heterogeneity in addition to SofT. The variable opinion i is the detection of a continuous latent variable, opinion i, that can be divided into J ordinal categories, where the thresholds separating the different categories are going to be estimated. In practice, for each of the 58 propositions, economists expressed their agreement on the fact that the content of that proposition was a relevant Cause or an appropriate Policy proposal. The answer was expressed on a Likert scale with four ordered options, coded as follows: Strongly Disagree (SD)=1; Partially Disagree (PD)=2; Partially 26 All the Propositions (Causes and Policy Proposals) that we included in the list are the ones discussed in the academic literature on the performance of the Italian economy during the last 15 years. This procedure was chosen to minimize the possible bias on the side of the survey-designers in choosing which possible alternative issues to be included in the pre-defined list. Therefore, in designing of the questionnaire we decided to refer to the selection revealed by the profession though publications. Note that also the grouping of the Proposition into Causes and Policy proposals has followed the same methodology. De Benedictis and Di Maio (2011) discuss the details on the methodology employed to select the essays from which the Causes and Policy Proposals have been extracted. 15

17 Agree (PA)=3; Strongly Agree (SA)=4. 27 The continuous latent variable can be thought of as the propensity to agree with the proposition, and the observed response categories are tied to the latent variable by the estimated thresholds. We can therefore write the population model as an ordinal regression model: opinion i = β 1 SofT i + c i β c + u i, (1) where u i is a normally distributed error term, with var(u i ) = 1. To make our sample as closed as possible to a generic draw from the population of economists in Italian Universities, we weight the ordinal regression using the same procedure we previously described. 3.1 Economists opinions and the school of thought Given the characteristics of the population model in equation 1, we run 58 separate weighted ordered probit regressions of the respondent s opinion concerning the content of the propositions on her/his SofT, including a large set of controls c i for individual characteristics (AGE, GENDER, MIGRANT, 28 WORRY, 29 OPTIMISM), community heterogeneity (RE- GION OF BIRTH, UNIVERSITY BA, 30 STUDIED ABROAD [1 and 2], 31 ACADEMIC POSITION), spatial heterogeneity (REGION OF WORK), knowledge heterogeneity (MAS- 27 Respondents could also select the option No opinion. Those cases (very limited: One or two for each proposition) were excluded from the sample. 28 This covariate takes value 1 if the respondent is working in a different region with respect to her/his birthplace, and 0 otherwise. 29 This covariate measures the respondent s self-declared level of worry (and its change in the last five years) about the perspectives for the Italian economy. 30 To account for the possibility that respondents who have studied in the same environment may have elaborated similar views on the economy and its functioning we include a set of fixed-effects for the University where the respondent received her/his B.A. degree. As in the analysis in section 2.3, we include a fixed-effect for each of the nine Universities with more than ten respondents in our sample and, for matter of convergence, we group together the remaining Universities. 31 To control for possible differences in economists opinions due to their different exposure to an international environment, we include two dummy variables taking value 1 if the master and the post-graduate degree are respectively earned abroad and zero otherwise. The reason to include this set of variables as controls comes from the acknowledgment that if and where economists earned their post-graduate degree (PhD or else) needs to be controlled for because this initial (to some extent not fully informed) choice might have influenced their future economic view of thinking. 16

18 TER, DOTTORATO/PhD, 32 FIELD OF RESEARCH, 33 ITALY, 34 DEBATE 35 ) and political orientation (MARKET, SOCIAL MOBILITY and RIGHT). Finally, to account for the possibility that the respondent s opinions are influenced by the peers with whom she/he interacts on an everyday basis we cluster standard errors at the level of the University where the respondent is employed. To test whether SofT i is a determinant of economists opinions, we run a likelihood test ratio between the ordinal regression model in equation 1 with and without the variable SofT i. Results indicate that SofT i is significant in 27 out of 40 Causes. In the case of Policy proposals, the role of the SofT is as much as important being significant in 12 propositions out of 18. These results show that, even controlling for a large number of covariates from individual characteristics to political opinions, the SofT plays an important role in explaining differences in economists opinions. While the number of propositions for which SofT i is significant indicates whether the SofT plays any role in explaining economists disagreement, 36 it is also important to know on which issues differences in opinions between economists are related to differences in their SofT. The full list of Causes and Policy proposals for which SofT i is significant are reported in tables 3 and 4. While several aspects of the results could be emphasized, for brevity here we focus only on few of them. First, when the opinion of Mainstream economists is significantly different from the one 32 In addition to MASTER (a dummy variable which takes value 1 if the respondent holds a Master in economics and zero otherwise), we include a dummy variable DOTTORATO/PhD that takes value 1 if the respondent holds an Italian Dottorato or a PhD and 0 otherwise. 33 The fields of research are classified according to the JEL codes. The largest group respondents belong to is Macroeconomic and Monetary Economics (17%). The other large groups are Industrial Organization (12%), Economic Development and Growth (11%), International Economics (10%), Public Economics (9%) and Microeconomics (9%). 34 The variable measures the respondent s knowledge about the Italian economy, proxied by the study of the Italian Economy being her/his primary or secondary field or research. 35 The variable measures whether the respondent has taken part to the public debate on the Italian economic situation. 36 Looking at the full list of Causes and Policy proposals, it should be noted that the propositions are not independent/orthogonal across them. Some indeed differ only for some detail while others are quite specific and unique. It follows that the number of propositions for which SofT i is significant should just be taken as an broad indication of the significance of the SofT in individual opinions. It is also worth mentioning that, on average, the propositions for which SofT i is not significant are the ones that received the highest level of consensus (measured by an entropy index and a different consensus index). See De Benedictis and Di Maio (2011) on this issue. 17