The Indirect Effect of Democracy on Economic Growth in the MENA Region (1990–2015)

: This paper examines the indirect effect of democracy on economic growth using a dataset of 17 MENA countries from 1990 to 2015. Democracy is assumed to affect growth through a series of channels: education, health, physical capital accumulation per labor, government consumption, and trade openness. A system of six simultaneous equations using 3SLS, is used to estimate the effect of democracy on growth through these channels. For further analysis, the countries are classified into groups according to the democratic status on the one side, and the level of income on the other. The results indicate that democracy enhances growth through its positive effect on health in all classifications of countries within the MENA region. However, the effect of democracy on growth through education and physical capital/labor is non-monotonic. Democracy hinders growth through government size and trade openness. Once all of these indirect effects are accounted for, the overall effect of democracy on growth is negative in less democratic countries and poor countries, but positive in more democratic countries and rich countries.


Introduction
The direct relationship between democracy and economic growth has been widely investigated in the literature in the last 50 years, and several theories have been developed regarding the relationship between the two variables. Democracy is not just one of the factors affecting economic growth; in fact, it creates the appropriate environment for various other factors to work effectively towards enabling growth. Therefore, even though the direct effect of democracy on growth may not be conclusive or significant, the indirect effect of democracy on growth via different linking channels is highly significant.
The main purpose of this study is to estimate the indirect effect of democracy on economic growth through different channels in the Middle East and North Africa (MENA) region for the period 1990-2015. We estimated a full system of equations determining economic growth and the channel variables using panel simultaneous equations model, specifically three-stage least squares (3SLS). First, we identify the most important channels through which democracy can affect economic growth and estimate the effect of democracy on these identified channels. Second, we estimate the effect of these channels on economic growth in the MENA region. Then, we calculate the indirect effect of democracy on growth. We estimate the indirect effect for different democratic groupings within the region separately to investigate whether the relationship relates to the stage of democratic transition of countries or not. Finally, we estimate the indirect effect of democracy on growth in rich and poor countries within the region separately. via different channels. Zghidi (2017) used a panel of 31 African countries over the period from 1986 to 2014 to investigate whether political stability and democracy increase the economic growth rate, using Generalized Method of Moment (GMM) dynamic panel data analysis. She concluded that both democracy and political stability have a positive and significant effect on economic growth. Rachdi and Saidi (2015) used a sample of 17 MENA countries to investigate empirically the effect of democracy on growth during the period from 1983 to 2012, using two different models: a Fixed Effect (FE) and Random Effect (RE) regression model, and a GMM model. They repeated the test five times for each model using several measures of democracy that capture different aspects of democracy and found the effect of democracy to be negative and statistically significant on economic growth for both models for four out of the five measures of democracy.
This work examines the indirect effect of democracy on growth through various channels. The study attempts to classify the estimated models into homogenous groups according to the stage of democratic transition of the groups of countries.

Model Specification
We specified a panel data model to estimate economic growth in 17 MENA countries over the period from 1990 to 2015. Education, health, physical capital per labor, government size, and trade openness were identified as important channels to include in our model. Therefore, the model includes six equations, an economic growth equation, as well as the five channel equations.
The estimated democracy coefficients of each equation are sensitive to the chosen specification, and especially to the exclusion of particular endogenous or exogenous variables. For the growth equation, the specification is derived from an augmented Solow model, with the set of channel variables as independent variables. The equation is expressed as follows: where t = 1, 2, …, 26 years , i = 1, 2, …, 17 MENA countries. lnGDPPit is the natural log of per capita income in period t, at each country i, lnMRit is the natural log of mortality rate, lnEDUSit is the natural log of secondary school enrollment rate, lnGCFLit is the natural log of gross capital formation per labor, lnGZit is the natural log of government size, lnTRit is the natural log of trade openness, and uit residuals. , , , , , are the parameters to be estimated, and according to the theory, it is expected that: > 0, > 0, > 0, > 0, > 0. The democracy index as well as income per capita are included in all the channel equations of the model. In this paper, we measure democracy by the electoral democracy index (EDEMit) obtained from the V-DEM institute, University of Gothenburg (Coppedge et al. 2016). The health equation is represented by infant mortality rate (lnMRit). Female education (lnEDUPFit) is considered an important determinant of health in the MENA region. Primary school education (lnEDUPit), health (lnMRit), among other variables are the determinants of the education equation, which is the second channel equation.
= + + + + + + + (4) = + + + + + + (5) Physical investment per labor (lnGCFLit) is the third channel, expressed by the gross capital formation per labor. Education, exchange rate, urban population and unemployment rate represent the main explanatory variables in equation 4. Equation 5 represents the government size channel, the government final consumption expenditure (% GDP) is an appropriate measure for this variable in the literature (lnGZit). Corruption index (lnCOit), population growth (lnPOPGit), trade openness (lnTRit) are important regressors in this equation. Finally, the openness equation includes the country area (lnLAi), inflation (INFit) and urban population ratio (lnUPOPit) as important determinants.
Data for all the variables, except democracy index is transformed into natural logarithm, therefore the estimated parameters in this form are the elasticities and the difference in logs approximate the growth rates, so the results are interesting and easy to interpret. The data, and its sources and measures, are represented in Appendix A, Table A1.
The previous specification is applied on various models, to test the sensitivity of our results on the one hand and to answer the following two questions on the other. Does this indirect effect of democracy on growth differ according to the state of political transition of each group of countries within the MENA region? Does this indirect effect of democracy on growth depend on the level of income per capita of these countries? To achieve the main aim of this study we classified countries within the MENA region into two groups, relatively democratic countries, which have achieved a minimum degree of democratic transition, and countries that are relatively autocratic within the region. We use, first, the classification of Freedom House of 'free and partly free' countries as one model and 'not free' countries as the other model. Second, we use the V-DEM institute classification of countries as 'electoral and liberal democracy and electoral autocracy' countries group as one model and 'closed autocracy' countries as the other model. Moreover, we classify the MENA countries into democratic countries and autocratic countries based on the average electoral democracy index score over the study period.
We then created another classification according to GDP per capita by dividing the countries into rich countries, which have an income of more than $4000 per capita annually, and poor countries, which have less than $4000 per capita annually. This number was decided as the average of the per capita GDP of all the 17 countries in the sample over the period of the study. This is equivalent to the World Bank classification, illustrated in Section 2.2, since we consider low income and lower-middle income countries as poor, but upper-middle income and high-income economies as rich. To capture the different effects of democracy on growth in poor and rich countries, if any, we included a dummy variable in our different channel equations. This dummy was coded 1 if the country is poor, and 0 otherwise. We then separately interacted this variable with democracy. The resulting interaction terms, democracy in poor countries and democracy in rich countries are included in all of our channel equations to capture the quantitatively distinct effects of democracy on growth via different channels in poor and rich countries. Details of different groups of countries in each model and descriptive statistics for variables in each model are illustrated in Appendix B, Tables A2-A8.

Methodology
An important assumption of ordinary least squares (OLS) estimators is that the regressors are exogenous. If this assumption is violated, regressors are correlated with the error term and OLS are biased and inconsistent. To solve this problem, simultaneous equations model using 2SLS estimator can be utilized instead since it is consistent even if the explanatory variables are endogenous. However, if the errors terms of various equations are correlated, the 2SLS estimators are consistent but inefficient. Seemingly Unrelated Regression (SUR) is efficient because it takes into account the correlation of errors across equations. However, it assumes that there are no endogenous variables on the right-hand side. If endogenous variables appear on the right-hand side of equations, the system estimation of SUR must be combined with the instrumental variables method of 2SLS. The resulting estimator is the 3SLS. While both 2SLS and 3SLS are consistent, the 3SLS is asymptotically more efficient than 2SLS because it uses information on the correlation of the stochastic disturbance terms of the structural equations (Intriligator 1978).
The 3SLS estimation contains three stages. First, all reduced-form coefficients are estimated applying the least square estimator. Second, structural coefficients are estimated, using 2SLS to each of the structural equations. Third, all of the structural coefficients of the system are estimated using Generalized Least Squares (GLS) estimators, using a covariance matrix for the stochastic disturbance terms of the structural equations, which is estimated from the second-stage residuals. Using the information contained in this covariance matrix improves efficiency (Intriligator 1978).
To construct our model, we estimated a full set of six simultaneous equations with six dependent variables, using 3SLS estimators-Following (Tavares and Wacziarg 2001) and (Baum and Lake 2003). The estimated coefficients of the economic growth equation ( )-Equation (1)-yield the effect of the channels on the growth in MENA region, whereas the estimated coefficients of the channel equations -Equations (2)-(6)-represent the effect of democracy on the channels. The product of the coefficient of democracy in the channel equation by the coefficient of the channel variable in growth equation illustrates how democracy affects growth indirectly through this particular channel. The summation of these calculated coefficients yields the total indirect effect of democracy on growth in the MENA region.
As mentioned above, the indirect effects of democracy on growth via each channel variable are calculated by multiplying coefficients across equations. Therefore, the statistical significance of these coefficients is not straightforward. The delta method (Oehlert 1992) is utilized to calculate the standard errors of these calculated coefficients, assuming that the covariance between γ and β is zero, the standard errors of the indirect effect of democracy on growth can be calculated as in Equation (7).
where is the calculated indirect effect of democracy on economic growth via each channel variable, i denotes each channel variable, education, health, physical capital/labor, government size, and trade openness respectively. is the estimated coefficient of each channel variable in Equation (1), is the estimated coefficients of democracy in each channel equation (Equations (2)-(6)). SE is the standard errors of the estimated coefficients.

Empirical Results
First, the effect of the different channels on economic growth is displayed in Section 6.1. Then, the effect of democracy on each channel is represented in Section 6.2. Lastly, the indirect effect of democracy on growth via each channel is calculated in different models based on the stage of democracy in each group of countries within the region, as illustrated in Section 6.3. Section 6.4 displays the estimated indirect effect coefficients of democracy on growth according to the country's standard of living. Finally, Section 6.5 examines the sensitivity of our results to some modifications of the model.
Before displaying the estimated coefficients of our models, diagnostic tests were performed to check the consistency and efficiency of our estimates. First, Hausman test was applied, as illustrated in Appendix C, Table A9, and the results confirmed the endogeneity of our models, hence the use of an endogenous technique such as 3SLS. Moreover, the Variance Inflation Factor (VIF) was applied to our estimated models-after performing 2SLS-and confirmed the absence of multicollinearity in all of our estimated equations, as illustrated in Appendix C, Table A10 2 . Growth (1990Growth ( -2015 Table 1 illustrates the results of estimating Equation (1), which represent the effect of each channel variable on economic growth. The estimated parameters are highly significant and consistent with the economic theory. The effect of education and health on growth is very strong and positive in all the models. This is expected, as human capital improvement enhances productivity, and consequently, increases economic growth. In the more democratic countries 3 , the effect of education 2 Only in two cases out of 36, the VIF coefficients exceed 10. This is in Equation (2) (where education is the dependent variable) in the closed autocracy model and in Equation (6) (where trade is the dependent variable) in the free and partly free model. 3 Authors mean by "more democratic" Free and partly free (from the freedom house classification), Electoral and liberal democracy and electoral autocracy (from V-DEM classification) and Democratic-countries (based on data classification). "Less democratic" means not free (from the freedom house classification), Closed autocracy (from V-DEM classification), and Autocratic-countries (based on data classification).

The Effect of Different Channels on Economic
is stronger than its effect in less democratic ones. A 1% increase in secondary school enrolment leads to a 0.55% increase in economic growth on average in democratic groups, but only 0.28% increase in growth in the less democratic ones. As asserted by many authors (Barro 1996;Freund and Jaud 2014) democratic countries reduce education inequality, which maximizes the effect of education on growth. On the contrary, the effect of health on economic growth is stronger in less democratic countries than in more democratic ones. A 1% decrease in mortality rate leads to an increase in growth by 0.53% and 0.96% on average in more and less democratic groups, respectively. The increase in physical capital formation per labor boosts the economy in all models; a 1% increase in investment per labor induces the growth by 0.87% on average of all models. Physical capital formation increases production and creates more employment opportunities. Furthermore, it leads to technical progress, which helps in achieving economies of large-scale production, increases specialization and provides machines and equipment for the growing labor force. Therefore, it leads to the expansion of the market (Shuaib et al. 2015). Consequently, the increase in physical capital formation is expected to foster the economy strongly. On the other hand, trade openness in the MENA region hinders economic growth in all models, with an elasticity of 0.30 on average. This could be attributed to the fact that the imports of these countries mostly exceed their exports, and that the terms of trade tend to be in disfavor of them. There is no significant difference between the effect of these two variables on growth in democratic and autocratic groups of countries.
Government size enhances economic growth, since more government expenditure on infrastructure, education and health affects economic growth positively. In addition, the effect of government size on growth is stronger in democratic countries than in autocratic countries, with an elasticity ranging from 0.42 on average for the relatively autocratic groups to 0.67 on average for the relatively democratic groups of countries.

The Effect of Democracy on the Link Variables (1990-2015)
Table 2 illustrates the elasticity of the different channels with respect to democracy, represented by the coefficient in Equation (2) to Equation (6), in different models. These coefficients represent the effect of democracy on each channel variable. The effect of democracy on education is significant and positive in the democratic countries, indicating that a 1% increase in democracy index tends to increase education by 0.22% on average. However, the same effect turns negative in the autocratic groups, with a marginal effect that equals −0.65 on average. The effect of democracy on mortality rate is negative and strong in all models. A 1% increase in democracy tends to decrease mortality rate by 1% on average in all models. This is explained by the fact that democracies tend to be usually more responsive to the basic needs of the people than dictatorships. They will choose policies that promote human capital accumulation, even if on account of physical capital. T-statistics are included in Parentheses.
The effect of democracy on physical investment per labor is also positive in democratic countries (0.80 on average), it turns negative in less democratic countries, equal to −1.20 on average. Democracy could affect physical capital formation positively or negatively. Several researchers argue that physical investment grows in a climate of liberty, free-flowing information and property rights. Democracy reduces the extent of political, social and economic uncertainty, and in turn encourages physical capital formation. On the contrary, other researchers claim that democracy may redistribute national income in favor of labor and disfavor of capital, by giving a greater voice to unions and labor wages and interests. Higher wages increase the cost of production, decrease the profits, and thus lower the incentives for private investment (Tavares and Wacziarg 2001).
The effect of democracy on government size is always negative and significant, a 1% increase in democracy decreases government size by 0.47% on average. More likely autocrats intend to increase the size of government to maximize their influence and control over the economy since their power is derived from the resources under their control (Tavares and Wacziarg 2001). On the contrary, democracies tend to decrease government size.
The effect of democracy on trade openness is very strong and positive in all models; a 1% increase in democracy tends to increase trade openness by 1.70% on average for all models. Democracies generally increase economic freedom and benefit a great number of consumers at the expense of a few producers, who receive more advantages from the protectionist policies. Therefore, democracy stimulates economic freedom and trade openness.
To conclude, the effect of democracy on mortality rate and government size is always negative, whereas its effect on trade openness is always positive. The effect of democracy on education and physical capital per labor is non-monotonic; it is negative in less democratic groups of countries but turns positive in the more democratic groups of countries. Growth (1990Growth ( -2015 The product of the coefficient of democracy in the channel equation ( in Equations (2)-(6)) by the coefficient of the channel variable ( ) in the growth equation (Equation (1)) illustrates how democracy affects growth indirectly through this particular channel. The overall indirect effect of democracy on economic growth is significant in the MENA region through the effect of all the selected channels, as illustrated in Table 3. The effect of democracy on economic growth is obtained through the effect of health, education and physical capital accumulation per labor. Although the effect of democracy on economic growth via education is unexpectedly negative in autocratic countries, it turns positive when the country is democratic. The latter result is attributed to the effect of democracy on education, as illustrated in Section 6.2. The effect of democracy on economic growth via health is very strong and positive in all the models. It is greater in the less democratic countries than in the more democratic ones. This result is attributed to the effect of health on economic growth as illustrated in Section 6.1.

The Indirect Effect of Democracy on Economic
Moreover, the effect of democracy on economic growth through physical capital accumulation per labor is significantly positive in the democratic countries, while it is significantly negative in the autocratic groups. An increase in the democracy index by 1% fosters growth in the democratic groups by 0.55% on average, and hinders growth in the autocratic groups by 1% on average. In democratic countries, where there is a climate of liberty, free flowing information, and secured property rights, the increase in democracy tends to enhance growth. On the other hand, where there is no good institutional framework through which democracy could be practiced; higher levels of democracy give a greater voice to unions, labor wages and sectarian interests. Therefore, the cost of production will increase, and profits will decrease, thereby reducing private investment.
The effect of democracy on economic growth through both government size and trade openness is mostly negative in all models with different specifications. An increase in democracy by 1% decreases growth by 0.43% on average via government size, and by 0.49% on average via trade openness. T-statistics of the indirect effect coefficients are calculated using the delta method.
From the above discussion, it is clear that the relationship between democracy and growth is non-monotonic via physical capital per labor and education, but it is monotonic via health, government size and trade openness. Therefore, the overall indirect effect of democracy on growth in the MENA region is non-monotonic as illustrated in Figure 1. It differs according to the stage of democratic transition of the group of countries. Democracy fosters growth in the more democratic groups; free and partly free countries, electoral and liberal democracies, and democratic countries according to the electoral democracy index, and hinders growth in the less democratic groups; not free countries, closed autocracies, and the autocratic countries according to the electoral democracy index, within the region.
Our results are in line with the Kuznets hypothesis (Kuznets 1955), which indicates that at the beginning of the political process, democracy decreases economic growth, but after democracy reaches a more mature stage, it enhances economic growth. In addition, our results are consistent with Barro's findings (Barro 1996). The latter's results indicate that democracy has a more favorable effect on growth at a middle level but negative effect in countries with a very low level of democracy. Our results report a negative effect of democracy in countries that are classified as "not free", "closed autocracy", or "autocratic", which usually score very low on the V-DEM electoral democracy index (0.25 or below on average). In addition, democracy has a positive effect on countries classified as "partly-free and free", "electoral democracies", and "democratic groups", which have a democratic index score ranging between 0.3 and 0.55 on average, which falls within the middle range of democracy defined by Barro.  Region (1990Region ( -2015 Democracy affects economic growth differently according to the level of per capita income in MENA countries. We investigated this hypothesis by estimating the effect of democracy on the different channels, as illustrated in Table 4. The effect of democracy on education is not affected by the level of income per capita; but it is significantly affected by the democratic stage of the group of countries. However, the effect of democracy on health is much stronger in rich countries than in poor countries. The latter effect is highly significant in all rich countries whatever the stage of democracy in the group, whereas it is insignificant in three out of six poor groups of countries within the region. The economic conditions are very important and mostly a pre-requisite for democracy to affect health significantly.

The Indirect Effect of Democracy on Economic Growth in Poor and Rich Countries within MENA
The effect of democracy on physical capital per labor is highly affected by the level of income per capita. Even if the overall effect of democracy on investment is negative in less democratic groups, it turns positive in rich countries. When this effect is positive in the more democratic countries, it is highly significant and strong in rich countries, but turns insignificant in poor countries.
The effect of democracy on government size is mostly negative in rich as well as in poor counties. This effect is stronger-more negative-in rich countries than in poor countries. Moreover, the effect of democracy on trade openness is always positive no matter in poor or rich countries. Turning to the overall indirect effect of democracy on economic growth, as illustrated in Table  5, important points arise. For all the models, democracy has an overall negative effect on economic growth in poor countries but a positive effect in rich countries. A 1% increase in democracy index fosters economic growth by 0.69% on average in rich countries but hinders growth by 1.03% in poor countries. Our findings are in line with the Lipset (1959) hypothesis, which states that development is a prerequisite for democracy, or that democracy cannot survive in poor and uneducated societies. Poor people do not have the luxury of choosing their rulers and having a voice in the political process. They are too preoccupied with earning their livings. T-statistics are included in Parentheses.

Sensitivity Analysis
The sensitivity to sample coverage, obtained from estimating six different models using the same equations' specification and the same methodology but different samples are illustrated in Sections 6.1-6.3. Three different democratic groups of countries based on different specifications have been estimated, and the results are consistent and follow the same trend in the three different estimated models; free and partly free countries, liberal democracies and electoral autocracies, and democratic countries according to the V-DEM electoral democracy index. In addition, the results of the autocratic groups of countries also follow the same pattern in the three different estimated models, not free countries, closed autocracies, and autocratic countries according to the V-DEM electoral democracy index. Moreover, in this subsection we examine the sensitivity of our results to the time period and to alternative specifications of the economic growth variable.
Firstly, we examined the sensitivity of the estimated coefficients to the period of time, therefore, we estimated our six equations over the period from 1995 to 2015. The estimated coefficients are consistent and stable. However, the coefficient of democracy in the trade openness equation takes a lower value in all of the models in the new sample, which has less degrees of freedom, as illustrated in Appendix D, Tables A11 and A12.
Secondly, we estimated the effect of democracy on the channel variables in alternative specifications of the economic growth variable, specifically GDP, GDPG, and GDPPG. The estimated coefficients have not changed in different specifications, indicating that the estimated coefficients are not biased. However, although the effect of democracy on health in the health equation still follows the same pattern, it represents relatively higher or lower values than before, as illustrated in Appendix D, Tables A13-A15. Finally, the effects of the channel variables on economic growth and the effect of democracy on the channel variables are estimated, and then the indirect effect of democracy on economic growth is calculated using three different estimators; 2SLS, SUR, and OLS for all of our six models. The results are illustrated in Appendix D, Tables A16-A33. The estimated coefficients are mostly the same indicating the robustness of the models' results.
Moreover, the FE estimator in simultaneous equations controlling for country fixed effect is also performed. The results are provided in Appendix D. The estimated effect of health and education on growth is comparable to the estimates of 3SLS. The effect of physical capital per labor on growth has lower values than usual, although it is always significant and positive. The effects of both government size and trade openness are less significant in the case of FE than in the case of other estimates. As expected, there is a reduction in the estimated effect of democracy on the channel variables; as controlling for country effects is akin to ignoring some of the between-country variation in the determinants of the channels, which may drive much of their partial covariation with democracy. As a result, the effects of democracy on education, health, and government size are less significant than in other estimators. In addition, democracy has mostly a negative effect on investment. However, democracy has the same effect on trade openness as in the case of other estimators.

Conclusions, Policy Implications and Future Work
Our results conclude that the overall indirect effect of democracy on economic growth is significantly positive in the more democratic countries but turns negative in the less democratic countries within the MENA region. Therefore, the effect of democracy on economic growth is nonmonotonic and changes according to the stage of democratic/political transition of countries. The effect of democracy on education and physical capital is responsible for this non-monotonic relationship. However, the effect of democracy on mortality rate and government size is always negative, and the effect of democracy on trade openness is always positive regardless the stage of democratic transition of different groups of countries.
Moreover, our results indicate that democracy might induce growth in rich countries, but hinders it in poor countries. A minimal level of income-$4000 per capita annually on average-was estimated as the lowest sufficient amount required for people to be educated and affluent enough to have the ability to both seek their political rights and to practice them effectively. Again, the effect of democracy on growth is found to be non-monotonic according to the standards of living in MENA countries. Therefore, improving economic circumstances, namely GDPP in MENA countries, as well as pursuing inclusive sustainable development policies need to be undertaken alongside adopting democratic governance since they are important fundamentals for positive democratic practice in these countries.
Human capital and physical capital accumulation were found to be the most effective links through which democracy affects growth in the MENA region. The health channel is more affected by the economic circumstances of the group of countries, whereas education is more affected by the democratic stage of the countries. Physical capital is highly affected by both the economic circumstances and political circumstances of the group of countries within the region.
It is recommended that policy makers encourage democracy in the relatively democratic groups of countries to stimulate economic growth. However, the autocratic groups of countries should start taking effective measures to transfer their regimes and their institutional framework towards one that can support real democratic practice, so they can reap the effects of democracy on growth.
Some areas in economic growth/democracy models have not been addressed in this study and need to be considered in future work. Electoral democracy is only one aspect of democracy, but institutional development ensures that the elected official is really empowered after election, they also guard against the possibility of elections being rigged. Democracy does not end at the ballot box. Therefore, it would be very beneficial to examine other measures or indices that capture other institutional aspects and components of democracy. In addition, the relationship between democracy and economic growth can be investigated in different groups of countries and for different time periods. Moreover, the indirect effect of democracy on economic growth can be estimated using other seemingly significant channels of influence, especially inflation, corruption and political instability as well as rule of law measures. If the oil production is more than one million barrels per day, the dummy = 1, 0 otherwise -   Appendix D. Robustness Analysis Table A11. The Effect of the Channels on Economic Growth in Different Models . T-statistics are included in Parentheses. Table A12. The Effect of Democracy on the Channels in Different Models .      T-statistics are included in. T-statistics of the indirect effect coefficients are calculated using the delta method.   T-statistics of the indirect effect coefficients are calculated using the delta method.  T-statistics are included in Parentheses. T-statistics of the indirect effect coefficients are calculated using the delta method.