Trust and Transition in Eastern Europe and the fsu

5.2 Civic participation, trust in public institutions and reform – preliminary regression results

So far we have presented just simple correlation's. But these leave open the question as to whether the various measures of social capital are jointly significant and what weight to attribute to civic participation and trust in public institutions respectively. We will now use some very simple and preliminary regression to analyse this issue. We also conduct a very basic robustness test by including economic reforms into the model. Many other potential determinants of growth in transition are not included, raising the possibility of spurious results due to model misspecification. To some extent we would argue that such problems are unavoidable when investigating transition outcomes. But the very small number of observations obliges us to limit the number of variables.

We examine a very simple model of the form:

Growth = a + b*Civic engagement + c*Trust in Institutions + d*Reforms + e

We also examine some interaction effects between all three variables. Trust in institutions is measured by the BEEPS investment climate scores. Reforms are measured as the average over the 1989-1998 period using EBRD’s transition indicators. Growth is given as the logarithmic difference in income levels in between 1998 and 1989.

Table 8 presents the results of our various specifications. At the bottom of each column are two test-statistics, which examine whether endogeneity bias is a problem for our results. The Sargan-test for valid instruments essentially tests whether the instruments used also belong in the regression equation. As we established in the previous section, exports to the CMEA and the urbanisation rate would seem to qualify as valid instruments due to the low correlation with growth but high correlation with civic participation. The Sargan test statistic fails to reject the validity of these two instruments in all the regressions. Note that for trust in institutions no instruments could be found and results thus need to be read with caution. The Hausman test checks the assumption of exogeneity directly by comparing the OLS and the IV estimators. Again in all the regressions shown, we fail to reject OLS estimation.¹⁷

The results in columns 1 and 2 confirm the simple correlations presented above. Over the 1989-1998 period, civic engagement is positively associated with economic growth. The value of the coefficient for "Type One" groups is considerably lower than for "Type Two" groups. Yet, the cross-country variance in the latter is also smaller – the value of the coefficients in both cases imply that one standard deviation in civic participation would explain around a 20 percentage point difference in cumulative growth. Moreover, the results suggest that although civic engagement may increase trust in formal institutions, it seems to have an independent positive effect on growth. One way through which civic engagement might directly benefit economic performance is by facilitating self-enforcement of market rules, without the need for recourse to third party enforcement by formal institutions.

Columns 3-4 show results of interacting civic engagement with trust in formal institutions. As mentioned, complementarity between these two variables is implied in Putnam's (1993) view of social capital. If his theory is correct, this also would imply a positive coefficient on the interaction term. Multicollinearity clouds the results in the case of "Type One" groups, although the sign of the coefficient on the interactive term is positive. For "Type Two" groups, we find no evidence of complementarity – in fact the interactive term is significantly negative.

In columns 5-6 of Table 8, we add economic reforms to the set of regressors. Economic reforms are positively correlated with each of these three variables, causing significance levels to drop relative to columns 1-4, although the total fit of the regressions improves considerably. Note, however, that for “Type Two” groups the coefficient remains significant (column 6) and is only moderately reduced in value from the result in column 2.

One way to correct for problems of collinearity is to adopt a two-step procedure, first regressing reforms against civic participation and trust in public institutions and then using the residuals from this regression as a proxy for reforms in the performance equation. The result for “Type One” groups appears in columns 7. Once the correction is used, the positive impact of social capital reappears, while reforms maintain a positive impact on growth. Finally, we attempted to find evidence for a positive interaction between reforms and the level of social capital but did not find convincing evidence that reforms work more effectively where civil participation or trust in the government are high.

Table 8. Social capital, reforms and performance: regression results
	Dependent variable: cumulative growth 1989-1998
			1	2	3	4	5	6	7
	Constant	-1.53*** (0.41)	-1.91*** (0.30)	-0.85 (0.98)	-2.62*** (0.47)	-1.93*** (0.48)	-2.15*** (0.42)	-1.53*** (0.41)
Type I groups (Civic)	0.021** (0.007)	-.026 (0.576)	0.007 (0.007)	0.02** (0.007)
Type II groups (Civic)					0.040*** (0.011)	0.11*** (0.03)	0.029** (0.011)
Quality of Investment Climate (Trust)	0.57* (0.27)	0.75*** (0.18)	0.095 (0.662)	1.25*** (0.34)				0.31 (0.34)	0.50 (0.29)	0.57* (0.27)
Type I*Trust			0.032 (0.038)
Type II*Trust					-.05* (0.02)
EBRD reform index (average 1989-1998)					0.46 (0.27)	0.35 (0.25)
EBRD reform index (residual)							0.46 (0.27)
Adjusted R-squared	0.42	0.56	0.45	0.60	0.52	0.63	0.52
Number of observations	19	19	19	19	19	19	19
Sargan test for valid instruments (X-square)	0.08 (0.00)	0.06 (0.00)	0.09 (0.00)	0.17 (0.00)	0.004 (0.00)	0.23 (0.00)	0.23 (0.00)
Hausmann test OLS vs. IV (F-stat)	1.56 (0.23)	0.00 (0.95)	1.04 (0.32)	0.01 (0.92)	0.76 (0.40)	0.15 (0.71)	0.76 (0.39)
Notes:	19	19	19	19	19	19	19
The regressions use simple OLS. All results were checked using Two Stage Least Squares and Exports to the CMEA and the Urbanisation Rate as instruments.
	The coefficient estimates for civic participation tend to increase in the TSLS estimations but all other results remain unaffected.
		* = 10% significance, ** = 5% significance, *** = 1% significance
			In column 7, the residual of a regression of EBRD reform index against Type II group participation is used as a measure of reform to get rid of multicollinearity.
				The Sargan test rejects the null hypothesis that the instruments do not belong into the regression on their own right for X-square values higher than a critical level.
		The Hausman test accepts the null hypothesis that IV is prefered over OLS for F-values above a critical level.