• Home
• About
• Latest Issue
• Board
• Search

The ideology of Italian political experts in comparative perspective. An unfolding analysis based on Benoit-Laver expert survey

Since Castle and Mair (1984), expert surveys have gained an increasing popularity as a method to detect the policy preferences of political parties. In this case, as it is well known, a survey is administered directly to country specialists who are asked to locate parties in their “own” countries on a set of predefined policy dimension(s): from the general left-right scale, to a variety of more specific dimensions (see below). The causes of the success of expert surveys are several (Mair 1990).

First of all, such surveys provide information on party policy positions in a common and standardised format, across a wide range of countries. Second, precisely because they reflect the judgements of experts – who are presumably well informed – they acquire weight and legitimacy. Finally, expert judgements are quick and easy compared to other methods (such as content analysis of party electoral programs or legislative behavioural studies). In the present work we will analyze the latest expert survey conducted by Benoit and Laver (2006), probably the most comprehensive and ambition attempt of this sort. However, instead of using experts’ answers to locate the different political parties within a given policy space, we will use them to directly map the policy preferences of the respondents. As a result, we will be able to characterize the “ideology” of the Italian political experts and to compare it with other countries.

The Benoit-Laver Expert Survey and the Italian Case

Benoit and Laver (2006) conducted their survey mostly in 2003, covering 47 countries (all European countries plus Canada, United States, Australia, New Zealand and Japan) and resulting in 1.491 expert responses, the far more comprehensive coverage than any other survey of this type. The typical expert in their survey was an academic specializing in political parties and electoral politics of her country. The lists of specialists was usually provided by the national political science associations (the Sisp in the Italian case) for each of the countries concerned. The parties that experts were asked to locate included all that were politically relevant (i.e., political party that won seats in the national legislature or that at least had won at least 1% of the vote nationally at the country’s most recent election). Regarding the policy preferences of parties, Benoit and Laver deployed a core set of four substantive policy dimension in each surveyed country to allow a directly comparable policy scales among different countries. In addition, they also measured policy positions on more country-specific dimensions, judged applicable on a country-by-country basis. The core four dimensions were: economic policy (interpreted as the trade-off between lower taxes and higher public spending), social policy (interpreted as policies on matters such as abortion, gay rights, and euthanasia), the decentralization of decision making and environmental policy (interpreted as the trade-off between environmental protection and economic growth). For each policy dimension, they used a scale running from 1 to 20, with the lower position indicating the typically “left” position and the higher value the traditionally “right” position. To take two examples, the previous economic policy dimension was defined as (1) “Promotes raising taxes to increase public services” and (20) “Promotes cutting public services to cut taxes”, while the social policy dimension was defined as (1) “Favors liberal policies on matters such as abortion, homosexuality, and euthanasia” and (20) “Opposes liberal policies on matters such as abortion, homosexuality, and euthanasia”. A party that is pro-market and liberal on social issues would then take a value close to 20 in the economic policy and a value close to 1 in the social policy dimension.

In the Italian case, the expert survey was conducted in 2003. The total respondents were 54 (only German, Sweden and Britain had a higher number of respondents in Western Europe) with a response rates of 30% (a value higher than the mean response rate for the entire sample¹). The parties covered were 13 along 9 policy dimensions. In addition to the four dimensions mentioned earlier, the following policy dimensions were included in the survey: deregulation (interpreted as state involvement in economic regulation), immigration (favoring policies designed to help immigrants integrate into the national society vs. favoring policies designed to help immigrants return to their country of origin), EU policy authority (interpreted as whether the domain within which the EU can authoritatively make policy decision should be expanded or restricted), EU accountability (interpreted as whether EU institutions should provide direct links to citizens through representative institutions such as the European Parliament or should be controlled instead by national governments), and a policy dimension on the issue of expanding the role of the EU in collective security, peacekeeping and other military affairs. The experts were also asked to locate each party on a general left-right dimension, taking all aspects of party policy into account.

Besides, and crucially for the analysis deployed in this paper, a “sympathy scale” was added in the survey. This “sympathy scale” asked experts to place all parties on a scale indicating their own closeness to each party’s, taking all aspects of party policy into account. As a result, a party perfectly close to the expert’s own personal policy preferences would take a value of 1, while the farthest party would take a value of 20. The purpose of this question was to test for respondent (possible) bias by checking whether expert placements of parties on substantive dimensions were correlated with their personal sympathy for a party’s policies. All other things being equal, an expert’s placement of a party should be unrelated to her closeness to that party. This same question, however, can be used to locate the respondents’ ideal points with respect to their policy preferences (see below).

Benoit and Laver replied the expert survey for the Italian case in early 2006². In this case 40 experts answered to the survey, while the parties covered grew to 16 along the usual 9 policy dimensions. Of course we do not know if the there is a perfect overlapping between the 2003 and the 2006 set of respondents (their anonymity is preserved). However, the fact that we can rely on two surveys for the same country is particularly interesting because it allows us to check for any temporal trend in the distribution of Italian political experts’ ideal points. This, as we will see, can even be used as an indirect check on the reliability of our results.

Table 1 reports the parties’ position along each dimensions considered for the 2006 survey (the corresponding table for 2003 can be found in Benoit-Laver 2006), while Figure 1 plots the mean left-right as well as the sympathy score for each party considered in the two surveys. As it can be seen, excluding the three more leftist parties (Rc, Pcdi and Verdi) produces an almost perfect linear relationship between the left-right placement of a party and its average level of “sympathy” accordingly to the experts.

Table 1: Summary Data from the Italian expert survey (2006)

p	Economic	Social	Environment	Decentralization	EU Peacekeeping	Immigration	Deregulation	EU Accountability	EU Authority	Left-Right
Rc	3.57	3.34	5.25	13.74	16.89	2.85	3.13	7.26	11.38	2.28
Pdci	3.56	4.09	6.09	13.38	15.94	3.37	3.59	7.53	10.53	3.05
Green	5.44	3.28	2.30	11.30	15.90	3.32	5.53	6.56	7.65	4.15
Ds	6.39	5.78	7.97	9.47	8.00	4.90	7.64	6.36	5.68	6.10
Rose	11.75	1.94	10.41	8.70	5.70	6.43	14.29	5.63	6.45	8.19
Marg	8.97	12.59	9.14	9.71	7.23	7.03	10.17	7.21	6.16	8.51
Iv	8.52	9.04	9.00	10.86	8.06	8.79	8.58	7.33	7.62	9.35
Udeur	9.16	16.36	11.89	11.72	7.33	10.18	9.25	10.14	9.91	10.60
Pri	13.64	8.18	14.70	9.14	5.62	11.53	14.09	8.71	8.88	11.64
Npsi	11.26	6.68	11.80	9.45	6.63	10.11	11.64	9.75	9.58	11.82
Udc	10.67	17.68	12.71	10.77	6.97	11.59	10.43	10.72	9.47	12.33
Fi	16.82	13.84	17.32	8.03	6.39	14.66	15.74	15.30	15.74	14.98
An	10.41	17.68	14.42	13.62	6.00	16.43	8.62	14.65	13.53	16.28
Ln	16.39	18.38	16.26	2.11	12.19	19.33	14.49	16.16	18.76	17.30
Msft	7.78	18.96	12.40	17.91	11.56	19.14	5.40	18.19	17.64	19.03
As	7.81	18.76	12.82	17.67	10.47	19.37	5.52	18.06	17.75	19.08

Figure 1: Relationship between the Left-Right score of a party and its average level of sympathy (Expert survey: Italy 2003 and 2006)

Unfolding the Policy Preferences of Italian Political Experts

As previously said, the “sympathy question” can help us to locate the policy preferences of the respondents. To this aim, employing an unfolding analysis is probably the best solution. In a nutshell, an unfolding model is useful in situations where the researcher has information about respondent’s preferences with respect to the spatial location of a given set of stimuli (in our case, political parties) along a given set of dimensions. More in details, each respondent is assumed to have a position of maximum preference with respect to the dimensions considered. This identifies the respondent’s “ideal point” because a stimulus located at that position would be preferred over all other stimuli. The ideal point is located such that the distance to the stimulus points correspond to preferences. The greater the respondent’s preference for a stimulus, the smaller the distance from the ideal point to that stimulus point, and vice versa. Objective of the unfolding analysis is to estimate the locations of each ideal point, using the respondent’s expressed preferences for the stimuli (Ingwer – Groenen 2005; Jacoby 2006).

We estimated our unfolding analysis under two different scenarios³. In the first one, we assumed that the respondents, when judging their own closeness (i.e., the closeness of their own ideal points) to each party, had in mind mainly the left-right dimension, considered as the single dimension that most constraints parties’ positions across a broad range of policies (Gabel-Huber 2000). In this sense, the left-right dimension becomes the dimension that better allows to take all aspects of party policy into account (as expressly asked to experts in the sympathy question).

However, the left-right division can be too crude to accommodate many important political division in a way that makes any sense. For example, a classical liberal (as well as a libertarian), which combines support for socially liberal policies with laissez-faire economics – has no unambiguous place on a left-right scale. As a result, in our second scenario we assumed that experts judged their own closeness to each party in terms of two distinct (at least on a theoretical grounds) dimensions: the economic (i.e. taxes vs. spending) and the social one. We chose these two dimensions (instead of others) for two main reasons: on one side, they remain a classical source of structure in political competition and in the definition of a personal ideological position that goes back to the birth of modern democracies⁴; on the other side, and more pragmatically, they allow to directly compare Italy with other countries, given that the economic and the social policy dimensions are two of the four dimensions deployed, as already stressed, in each country surveyed. While we must remain ever alert to the possibility that the “same” policy dimensions mean different things in different countries, it nonetheless remains the case that these two dimensions are at least broadly equivalent in their substance.

In both our scenarios, we constrained the stimulus points locations (i.e., the parties locations) according to the mean score each party obtained along the dimension(s) considered. To better clarify the logic we followed, let’s take the first scenario as an example. In this case, we have two sets of information: 1) the answers from each respondents to the sympathy question. 2) the mean position of each party along the left-right dimension as it results from the expert survey. The unfolding analysis just tries to take advantage of both set of information treating parties’ scores as the “anchors” through which recovering the ideal points of the respondents. The same applies to the two-dimensional case⁵.
The fit measures of the unfolding analysis in both scenarios are very similar and acceptable⁶. In other words, both the left-right dimension as well as the two-dimensional space comprised by the economic and social dimension are able to effectively recover the ideal points of the respondents with respect to the stimulus locations. In this sense, figure 2 plots the locations of the Italian parties in the unidimensional scenario as well as the Kernel density distribution of the ideal points of the respondents in the two surveys. As we can see, the shape of the distribution in the two periods remains quite stable. This is a good news for our analysis. Indeed, given such a short time span elapsed between the two Italian surveys, a sharp difference between the two distributions would have made us quite suspicious about the reliability of our results.
Accordingly to our estimation, roughly the 90% of respondents’ ideal points in 2003 (and the 85% in 2006) places itself to the left of the midpoint of the Left-Right scale, producing a mean position that remains in both surveys fairly close to Ds (mean score in 2003: 6.02 – standard error(s.e.): .40; mean score in 2006: 6.42 -s.e.: .51)

Figure 2: The Kernel density distribution of the ideal points of the respondents along the left-right dimension (Italy 2003 and 2006): 1=Left, 20=Right. Mean expert position reported with its .95% c.i.

Figure 3 plots the locations of the experts’ ideal points under the bi-dimensional scenario in the two survey, together with the expert mean location and the position of each party. In an interesting way, the four quadrants that appears in the figure can be used to describe four broad belief systems (see Lielie-Maddox 1984). The South-West quadrant identifies a social-democrat “character”, i.e., a person who supports government economic intervention and has a liberalish position in the social dimension. The North-East quadrant identifies its exact opposite: a Conservative type. The South-East quadrant identifies a Classical liberal (or a libertarian) type, i.e., a person who supports expanded individual liberties but opposes economic intervention. The final quadrant (the North-West one) is the more difficult to describe. In this quadrant we have who supports government economic intervention while being conservative from a social point of view. Lielie and Maddox use the term “populist”, but we prefer to use “paternalist” to avoid the ambiguities often involved in the previous term⁷.

Accordingly to this typology, in both Italian surveys almost all of the respondents falls between the social democrat and the libertarian quadrants, while the mean position along the two dimensions clearly identifies a “social democrat type” in Italy 2003 (economic mean: 8.69 – s.e. .56; social mean: 6.53 s.e. .26) as well as in Italy 2006 (economic mean: 8.89 – s.e. .62; social mean: 7.09 – s.e. .30)⁸. More in details, a stable 67% of the respondents falls in the social democrat quadrant in both surveys, while 29,4% and 25% of experts (respectively in 2003 and in 2006) falls in the libertarian quadrant. As in the previous scenario, also in the bi-dimensional space the mean ideal point remains fairly stable in the two surveys. However, while in 2003 the closest party (in Euclidean terms) to the mean export position is Sdi (followed by Ds and Iv), in 2006 it is replaced by Iv (followed by Npsi and Ds).

Figure 3: The mean expert position in the two-dimensional scenario

A cross-national comparison

We replicated the unfolding analysis employed in the Italian case for each country surveyed in Benoit and Laver (2006) that satisfied the two following conditions:

1. at least 20 respondents to the sympathy question (to guarantee a reasonable pool of respondents for each country under investigation⁹).

1. at least 5 parties considered in the survey (a fewer number of stimuli would undermine the reliability of the unfolding analysis). So, for example, we did not analyze United States, despite its 167 respondents, precisely because in this case the American experts had to evaluate their own closeness to just two parties. These two conditions left us with 21 countries out of 47 surveyed (plus the Italian survey of 2006), and resulting in 958 expert responses (61.6% of the total expert responses reported in Benoit-Laver 2006).

Quite interestingly, both the average goodness-of-fit as well as the average badness-of-fit of the unfolding analysis for the countries analyzed show a better fitting under the bi-dimensional scenario compared to the unidimensional one¹⁰. In this sense, and contrary to what happened in the Italian case, the bi-dimensional space seems better able to recover the ideal points of the respondents with respect to the stimulus locations than the unidimensional solution. The only countries, besides Italy, where the left-right dimension presents an ability to recover the ideal-points of the experts comparable to the bi-dimensional scenario are the following: Britain, Germany, Belgium, Netherlands, Hungary, Israel, Portugal (and, to a lesser degree, Spain).

Figure 4 plots the mean location of the experts’ ideal points in each country analyzed under the first scenario (i.e., considering the left-right dimension) as well as their 95 percent confidence intervals. We even reported the Overall mean for the entire sample¹¹. As we can see, the mean position of the Italian political experts (both in 2003 and in 2006) is the most leftist one, followed by Netherlands, Israel and New Zealand, while the ideal points of Japan, Hungary and Germany are the most moderate. Even when we take into consideration the uncertainty associated with each mean (i.e., its confidence intervals), we can say that the Italian mean position (in both years) is statistically different (and more to the left) than the Overall mean for the entire sample (that it is equal to 8.56 – s.e. .12).

Figure 4: Cross-national comparison of the ideal points of the respondents along the left-right dimension: 1=Left, 20=Right. Mean expert position reported with its 95% c.i.

The results under the second scenario are plotted in Figure 5¹². As we can see, all the expert means (with the partial exception of Japan) fall into the two lower quadrants. The Italian mean in both years is slightly to the left of the Overall mean along the Economic dimension (albeit just in 2003 in a statistically significant way), while along the Social Liberalism dimension the two Italian means and the Overall mean are remarkably close to each other. The coordinates of the Overall mean for the entire sample of coutries are respectively: 10.29 along the Economic dimension (s.e. .16) and 6.49 along the Social Liberalism dimension (s.e. .13). In this sense, the Overall mean position is placed right between the Social Democrat and the Libertarian quadrants.

Figure 5: The mean expert position in the two-dimensional scenario (all countries considered)

Table 2 reports at this regard the percentage of the respondents’ ideal points in each country investigated that according to our analysis falls into each of the four quadrants. The highest percentage of Social Democrats can be found among the experts in Netherlands and in New Zealand (around 90%), the highest percentage of Libertarians can be found in Germany, Sweden and Hungary (above 70%), while both the highest percentage of Paternalists and Conservatives can be found in Japan.

Table 2: Percentage of experts in each belief-system quadrant

Country	Social Democrat	Libertarian	Paternalist	Conservative
Italy 2003	66.7%	29.4%	2.0%	2.0%
Italy 2006	67.5%	25.0%	2.5%	5.0%
Germany	22.0%	71.4%	2.2%	4.4%
Sweden	52.5%	39.3%	4.9%	3.3%
Switzerland	66.0%	31.9%	0.0%	2.1%
Spain	72.0%	20.0%	6.7%	1.3%
Britain	62.1%	29.3%	3.4%	5.2%
Finland	68.8%	25.0%	3.1%	3.1%
Slovenia	39.7%	53.4%	0.0%	6.9%
Japan	27.3%	22.7%	15.9%	34.1%
Irleland	43.1%	52.9%	2.0%	2.0%
Canada	43.4%	42.4%	5.1%	9.1%
Hungary	2.7%	70.3%	5.4%	21.6%
Czech Republic	25.7%	37.1%	5.7%	31.4%
Poland	20.7%	51.7%	3.4%	24.1%
Denmark	43.5%	47.8%	4.3%	4.3%
Israel	50.0%	39.3%	0.0%	10.7%
Norway	72.2%	22.2%	5.6%	0.0%
Netherlands	90.0%	5.0%	5.0%	0.0%
Portugal	10.0%	50.0%	5.0%	35.0%
Belgium	76.2%	14.3%	9.5%	0.0%
New Zealand	89.5%	5.3%	0.0%	5.3%
Overall Mean	48.2%	39.3%	4.3%	8.3%

Conclusion

In this paper we used the latest expert survey of Benoit and Laver (2006) to unfold the policy preferences of the Italian political experts and to compare them with the preferences of experts of other countries. Our interest has been mainly descriptive, but our results can be helpful for investigating more analytical questions. As Steenbergen and Marks recently wrote (2007: 361) “reliance on expert judgements is an attractive option for measuring complex phenomena such as party positions about policies. But how valid are expert judgements really?”. For example, it can be shown that in some countries (like precisely Italy¹³) there is a systematic evidence (i.e., a statistically significant one) of bias in expert placements, especially against right-wind parties. Less sympathetic experts tend to place these parties as more extreme than do more sympathetic experts. When this happens, the bias can jeopardize the usefulness of expert scores for any empirical analysis, given than now we do not have anymore any reliable scores (along the different policy dimensions) for the biased party. A natural question that arises is then why only some specific countries do present this kind of (ideological) bias. This, and other research puzzles about expert surveys, can be investigated using precisely the methodological tools we employed here.

References

• Barisione M. (2004). Tradizionalismo etico, liberalismo economico: oltre la destra e la sinistra? Polis, XVIII: 237-66
• Benoit, Kenneth and Michael Laver (2006) Party Policy in Modern Democracies. London: Routledge.
• Busing F. – P. Groenen – W. Heiser (2005). Avoiding degeneracy in multidimensional unfolding by penalizing on the coefficient of variation. Psychometrika, Vol.70(1): 71-98
• Castles F. – P. Mair (1984) Left-Right Political Scales: Some Expert Judgements, European Journal of Political Research 12: 73-88
• Enelow J. – M.J. Hinich (1984) The Spatial Theory of Voting: An Introduction. New York: Cambridge University Press.
• Ingwer B. – P.Groenen (2005). Modern Multidimensional Scaling. Berling: Springer
• Jacoby W.G. (2006) Multidimensional scaling: theory and applications for public opinion research. Papor Conference, December 7, 2006
• Maddox W. – S. Lilie (1984). Beyond liberal and conservative. Reassessing the political spectrum. Washington: Cato Institute
• Mair, Peter (2001). Searching for the Positions of Political Actors: A review of approaches and a critical evaluation of expert surveys, in Michael Laver (ed) Estimating the Policy Position of Political Actors, pp.10-30. New York: Routledge

• Riker W.H. (1982) Liberalism against Populism. Illinois: Waveland Press

Footnotes

¹ In our analysis, we implicitly assume that the set of respondents of the Italian survey is a good sample of the Italian population of political experts. In this sense, for example, we discard the presence of any possible self-selection bias. Unfortunately, we have no way to control for it directly. However, the high response rate actually registered in the Italian survey should reassure us (at least partially) on this point.

² We thank Kenneth Benoit and Michael Laver for having kindly provided us the data.

³ The analysis has been conducted using the preference scaling (Prefscal) procedure included in Spss.

⁴ See Barisione (2004) for an application to the Italian case.

⁵ This method is called “external unfolding model” because the stimulus point locations along a given set of dimensions are already known (in our case, from the expert survey) and we only seek to locate the ideal points with respect to the stimuli. An “internal unfolding analysis” seeks to locate both ideal points and stimulus points simultaneously. See Enelow – Hinich (1984) for a classical application of this latter method.

⁶ The Pearson correlation between the disparities and the scaled distance is .83 (Italy 2003) and .81 (Italy 2006) in the unidimensional scenario and .83 and .77 in the bi-dimensional scenario, while the Spearman rank is, respectively, .82 and .79 (unidimensional scenario) and .82 and .77 (bi-dimensional scenario). Either of these correlations assesses the degree to which the scaled distances (from the unfolding analysis) are monotonic to the dissimilarities data: a higher value implies a better goodness-of-fit measure. Regarding the badness-of-fit measure, the penalized-stress value (Busing et al. 2005) is slightly less in the bi-dimensional scenario (.66 and .71 respectively) than in the unidimensional scenario (.73 and .74). Smaller values indicate better scaling solutions.

⁷ See Riker (1982)

⁸ The fact that the standard error of the economic mean is roughly two-times the standard error of the social mean, illustrates how the respondents present a higher diversity of preferences along the first dimension compared to the second one.

⁹ The only two exceptions are Norway (18 respondents) and New Zealand (19 respondents). In both cases, however, the response rate to the survey is higher than the mean response rate for the entire sample. Viceversa, we did not consider Turkey, despite its 29 respondents, precisely because its rate of response was very low (9%). France was not considered because the sympathy question was not measured in its survey.

¹⁰ Under the bi-dimensional scenario, the average Pearson correlation is .81, the average Spearman rank is .79 and the average penalized stress value is .65. On the contrary, under the unidimensional scenario these values are respectively .74, .72 and .74

¹¹ As the fit of the unfold analysis improves, we get more confident about the results (including the location of the mean), as well as about the comparability of the results. In this sense, in Fig.4 we did not report three countries (Ireland, Poland and Czech Republic) that recorded a really low fit (i.e., a Pearson correlation as well as a Spearman rank less than .65)

¹² The economic dimension in Figure 5 comes from one of two sources, depending on the type of country. For most countries, economic policy is represented by the “tax cuts versus spending increases” dimension (like in the Italian case). For post-communist countries, however, following Benoit and Laver (2006) we employed the dimension of “state ownership of business and industry versus privatization” dimension. To represent liberal versus conservative social policy, we used the aforementioned “social” dimension with the exception of New Zealand where social policy was not measured. In this case, following once again Benoit and Laver (2006), we employed the immigration dimension.

¹³ By replicating the method proposed in Benoit – Laver (2006), we found in 2003 a systematic bias against Forza Italia, Northern League and Alleanza Nazionale (but, quite interestingly, not against the extreme-right party Msft). On the contrary, in 2006 the bias is only against the two extreme-right parties (As and Msft).