Making Connections? A Study of Interethnic Dialogue in a Divided Community in Bosnia and Herzegovina

ABSTRACT

Dialogue can have positive effects for participants, but it is not clear how it affects members of the wider community. This study examines a dialogue project in Prozor-Rama in Bosnia and Herzegovina using survey data from 2012 and 2015. Results show that there is a general positive trend in interethnic relations, however, dialogue in Prozor-Rama had the effect of lowering in-group trust and dampening the increase in civic engagement relative to a control town. This suggests that in the short-term dialogue can have negative effects. They also suggest the need for studies of the long-term effects of dialogue. This study is significant because it moves beyond looking at the effect of interethnic contact on attitudes and prejudice and also examines its effect on social ties, social interaction, and civic participation.

Acknowledgements

The author would like to thank James, McCann, Samuel McCormick, Sarah Mustillo, Aaron Hoffman, Anne Holthoefer, Laura Heideman, Inger Skjelsbæk, Daniel Aldrich, Dino Djipa, Elvir Djuliman, Edita Zovko, Steinar Bryn, Chih-Chien Huang, Kevin Doran, Elizabeth Rickenbach, Kristine Adams, Scott MacNeil, and several anonymous reviewers.

Notes

1 This study design allows for innovative longitudinal comparisons. As explained in greater detail below, however, it was not feasible to field a panel survey where the same individuals were tracked over time.

2 This leads to a total N = 1201.

3 Wooldridge (2013, p. 457) writes,

A natural experiment occurs when some exogenous event— often a change in government policy— changes the environment in which individuals, families, firms, or cities operate. A natural experiment always has a control group, which is not affected by the policy change, and a treatment group, which is thought to be affected by the policy change. Unlike a true experiment, in which treatment and control groups are randomly and explicitly chosen, the control and treatment groups in natural experiments arise from the particular policy change.

4 The director of NDC Mostar stated that in choosing a site for dialogue NDC wanted a town in Herzegovina-Neretva canton, that had divided schools. Possible towns included Mostar, Stolac, Capljina, and Prozor-Rama. Since they had already worked in Stolac and Mostar this left Capljina and Prozor-Rama. He said they choose Prozor-Rama because it was further away than Capljina and they wanted to spread out their activity. In other words, the reason the NGO chose Prozor-Rama over Capljina was not related to the observed outcome of interest of this study (e-mail 26 April 2017).

5 Federalna zavod za statistiku, Federacija Bosne i Hercegovine (Federal Bureau of Statistics of Bosnia and Herzegovina). See http://www.px-web.fzs.gov.ba/pxweb/bs-Latn-BA/?rxid=bee0833d-88c4-4ca3-8444-d8fa2227fb24; Retrieved January 2019.

6 Of the 300 respondents sampled in Prozor-Rama in 2015 only six people say they participated in a dialogue seminar organized by NDC. Controlling for this small number of participants and running all models without them does not alter any of the findings in this paper.

7 The variables used in this paper were produced using questions from the Social Capital Community Benchmark Survey 2000, modified for Croat and Bosniak identities.

8 Croats were asked about Bosniaks, and Bosniaks were asked about Croats. Only four respondents did not identify as one of these two groups; four Serbs and one ‘other’. This analysis only includes Croats and Bosniaks. Bosniaks were surveyed by Bosniaks and Croats by Croats to avoid social desirability bias.

9 Wooldridge introduces the difference in differences approach for OLS regression. I also use this model with logistic and ordered logistic regression. In the case of logistic regression, the model is linear and the coefficient on the interaction term is the treatment effect, and the dependent variable is the log odds of the outcome occurring. If the coefficient on the interaction term is positive and statistically significant, then we can conclude that the treatment has the effect of increasing the log odds of the outcome occurring. However, we cannot draw conclusions about the size of the effect. Graphs of predicted probabilities are included below as recommended by Long and Mustillo (2018).

10 Similarly, Ditlmann and Samii (2016, pp. 384–385), citing Angrist and Pischke (2009), note that ‘The difference-in-differences design accounts for baseline differences between participant and control groups that would otherwise bias a simple cross-sectional comparison.’

11 2015 corresponds to d2 and Prozor to dT in Equation (1) above.

12 In the linear case the average treatment effect is equal to the difference-in-differences or the outcome for the treated group minus the outcome for the control group at time two, minus the outcome for the treated group minus the outcome for the control group at time one. This means that the average treatment effect is taking into account changes that might be occurring in the control group from time one to time two (Wooldridge, 2013). In an MLE model the interaction term still represents the treatment effect, but the left side of the equation 1 is the log of the odds ratio of the outcome occurring. We can therefore make inferences about the direction of the effect and whether it is statistically significant, but not directly the size of the effect. For the latter I report predicted probabilities using graphs as recommended by Long and Mustillo (2018).

13 Based on Model 2, Table 2. The results are nearly identical when run with the cem model.

Additional information

Funding

This work was supported by Norwegian Embassy in Sarajevo, Bosnia and Herzegovina [grant number BHZ15/008].