1. Fractionalization and its Effect on Social Cohesion:
"It's the Social Exclusion, Stupid!"
Irene van Staveren, Institute of Social Studies
Zahid Pervaiz, Lahore School of Management
ECINEQ conference July 2015, University of Luxembourgh

2. “Missing links” in development literature
Positive effect on growth
Social capital
Institutions
Converging empirically around trust and informal institutions: => social cohesion channel

3. Social constraint on growth
Ethnic fractionalisation
Ethnic polarization
Negative effect of ethnic diversity on growth?

4. Theoretical puzzle
In economics there is a strong case in favour of diversity
Controlling for income inequality (Gini) does not change the negative and statistically significant effect of ethnic diversity on growth
So, how can we explain the empirical result?
We suggest that it’s not diversity but social exclusion, which drives the negative effect

5. Testing for social exclusion:
Income inequality = vertical inequality
Social exclusion = horizontal inequality
Ethnic diversity is an expression of horizontal inequality => controlling for income inequality is inadequate
Why? We therefore argue that what matters for development is not so much the number and size of groups and vertical inequalities, but how groups relate to each other. Do they tolerate and respect each other or do they discriminate each other and fight over scarce resources? Is there generalized trust of others or only in one's own group? Are group members willing to interact with others? Do they feel listened to and accepted as members of society? In other words: do groups contribute to social cohesion or do dominant groups disrupt social cohesion through the social exclusion of other social groups?

6. Literature
Easterly & Levine (1997) on Africa: “… interest group polarization leads to rent-seeking behaviour and reduces the consensus for public goods, creating long-run growth tragedies.”
Sokoloff & Engerman (2000): historical polarization driven by elites explain growth differences in the Americas
=>Problem: both confuse diversity with polarization
Alesina & La Ferrara (2005) recognize that diversity can run along a variety of dimensions => “open questions”
Casey & Owen (2014) recognize that elite capture must be measured rather than income and wealth inequality

7. New ideas emerging
Fedderke, Luiz & de Kadt (2008): ethnic identity is weaker over time
Shcherbak (2012): what matters is a society’s tolerance of diversity
Community-level studies on developed countries: it’s not diversity as such, but socioeconomic deprivation of particular social groups, which results in negative effects on community social cohesion.

8. Our hypothesis
It is not the number or size difference of ethnic groups which matter, but rather the way ethnic groups are socially positioned and its members relate to members of other groups, which affects social cohesion, and thereby economic development processes and growth.

10. Methodology & data
Dependent variable: Intergroup Cohesion from Indices of Social Development, ISS
Independent variable 1: Ethnic fractionalisation (and linguistic and religious fractionalization) as in the literature
Independent variable 2: Inclusion of Minorities from Indices of Social Development, ISS
Control variable: log GDP per capita

11. ISD Database at ISS: www.IndSocDev.org freely online available
Data from 1990 until 2010 in 5-year periods
Almost every country in the world
Continuously updated
Transparent:
Which indicators go into which index
6 indices and 200 indicators are downloadable
Standard errors provided per country and variable

12. 6 indices:
Civic Activism, measuring use of media and protest behaviour
Clubs and Associations, defined as membership in local voluntary associations
Intergroup Cohesion, which measures ethnic and sectarian tensions, and discrimination
Interpersonal Safety and Trust, focusing on perceptions and incidences of crime and personal transgressions
Gender Equality, reflecting gender discrimination in home, work and public life.
Inclusion of Minorities, measures levels of discrimination against vulnerable groups such as indigenous peoples, migrants, refugees, or lower caste groups.

13. Methodology ISD
ISD combines over 200 indicators from 25 independent and reputable sources
Uses ‘matching percentiles’ method used for Corruptions Perceptions Index
Lambsdorff 1999
Rationale for matching percentiles
Combination of sources measuring same phenomenon more reliable than each source separately
Indices broaden the coverage compared to single source
Minimum 3 independent sources to develop index
As mentioned before, ISD combines over 200 indicators from 25 independent and reputable sources – most of which were mentioned before
The methodology we use is that of ‘matching percentiles’, which is a method which was developed for the Corruptions Perceptions Index by
Lambsdorff. Details can be found at the Transparency website – and in future we will produce a similar detailed description for our website
Rationale for this methodology is 2-fold:
Combination of sources measuring same phenomenon more reliable than each source separately
Indices broaden the coverage compared to single source
For the reporting of an index, there need to be a minimum of 3 independent sources
Please note that we do report 2010 data on the website, but that these data are not fully developed and justified yet
I will explain the methodology in the following steps, over 9 slides

14. Estimation methods for social cohesion
Cross-section all developing countries 2010
Panel all developing countries (5 years) with random effects (Hausman test did not favour fixed effects)
Endogeneity is not likely because social cohesion changes only very slowly over time.
Fractionalization data is available for one year only
Robustness checks with two alternative fractionalization variables

15. Table 1a. Summary statistics cross-section 2010
Variable N
Mean
St Dev
Min
Max IC 87
0.6524
0.0879
0.2040
0.7568 EF
0.5385
0.2401
0.0394
0.9302 LF 83
0.4643
0.3126
0.0080
0.9227 RF 86
0.4326
0.2405
0.0023
0.8603 IM 69
0.4377
0.0487
0.3107
0.5394
lnGDPpc
7.1814
1.0586
5.0156
9.3155

16. Table 1b. Summary statistics panel 1990-2010
Variable N
Mean
St Dev
Min
Max IC
346
0.5845
0.0871
0.2040
0.7568 EF
435
0.5405
0.2383
0.0394
0.9302 LF
420
0.4695
0.3128
0.0080
0.9227 RF
0.4352
0.2393
0.0023
0.8603 IM
207
0.4593
0.0769
0.1726
0.8510
lnGDPpc
428
6.9332
1.0338
3.9129
9.3155

17. Table 2. Results of cross-section estimation; dependent variable Intergroup Cohesion
Model 1
Model 2
Model 3
Model 4 EF LF RF ** IM
***
***
***
ln GDPpc ** ** ** C*
***
R-sq N 82 67 63 65
Model 1 shows a small negative effect of ethnic fractionalization on intergroup cohesion: a 0.10 increase in ethnic fractionalization is associated with a decline in intergroup cohesion (both on a scale of 0-1). The parameter, however, is not statistically significant.
Model 2 adds the social exclusion measure and shows that now the sign for ethnic fractionalization is positive, but still not statistically significant. Inclusion of minorities, however, has a statistically significant positive effect: a 0.10 increase in inclusion of minorities (10% of the scale) is associated with a 0.12 increase in intergroup cohesion, which is a substantive effect, larger than one standard deviation in intergroup cohesion.
Model 4 shows that both religious fractionalization and the inclusion of minorities have positive and statistically significant parameters. A 0.10 increase in religious fractionalization is associated with a increase in intergroup cohesion (1% of the standard deviation of intergroup cohesion). Whereas a 0.10 increase in inclusion of minorities is associated with a 0.11 increase in intergroup cohesion, which is more than a standard deviation.
** significant at 5% level; *** significant at 1% level

18. Table 3. Results of panel estimation; dependent variable Intergroup Cohesion
Model 1
Model 2
Model 3
Model 4 EF LF RF
*** IM
***
***
***
ln GDPpc
*** * *
*** C*
***
***
***
***
R-sq N
340
198
190
195
Both in model 1 and model 2, the signs of the parameters for ethnic fractionalization are negative but not statistically significant.
Model 2 shows that the effect of inclusion of minorities is substantive. A 0.10 increase in inclusion of minorities is associated with a sizeable 0.23 increase in intergroup cohesion.
In model 4, we find that religious fractionalization shows a positive and statistically significant correlation with intergroup cohesion, very similar to the results of the cross-section analysis.
* significant at 10% level; ** significant at 5% level; *** significant at 1% level

19. Discussion
The difference between the cross-section results and panel results are small: mainly the size effect of social exclusion is larger in the cross-section
Large difference in R square between cross section and panel
Ethnic and linguistic fractionalization have very small negative effects and not statistically significant
Inclusion of minorities has a large, positive and statistically significant effect on social cohesion (larger than a stdev)
10% higher GDP per capita has a much smaller impact on social cohesion than a 10% point increase (for example from 0.30 to 0.40) along the scale of inclusion of minorities

20. Conclusion
Horizontal inequality generates not merely differences in economic benefits but expresses social exclusion from large parts of the economy
But the analysis needs to be expanded with more controls and perhaps a time-lag
Next steps:
develop a model in which also income inequality is included
develop a growth model with ethnic diversity, social exclusion, and income inequality

21. Indices of Social Development database www. IndSocDev
Indices of Social Development database
ISD team:
Irene van Staveren
Ekaterina Evdokimova
and external advisors