4.1 Environmental degradation, income inequality and freedom of the press

Since environmental degradation has significant negative externalities, to avoid high levels of pollution, the market failure problem needs to be addressed. It is widely accepted by the literature that as GDP per capita increases, the society becomes better able to solve this market failure. To better understand the relationship between economic growth, environmental abatement, and preferences, we follow the theoretical background presented in Magnani (Reference Magnani2000). She takes from Baldwin (Reference Baldwin, Goldin and Winters1995) the division of pollution into two components: incipient pollution and abatement. Incipient pollution, represented by I _it, is the amount of pollution a country would produce at its current level and composition of output if there were no environmental costs. Hence, if a country has high taxation on pollution and large environmental costs, then the level of incipient pollution will be much higher than the country's actual level of pollution. This difference will be considerably reduced for a country that has relatively low taxes on pollution and low costs for environmental damage. The difference between incipient pollution and actual pollution is referred to as abatement and is represented by E _it. Countries with the highest levels of taxation on pollutants will thus also have higher abatement levels. Hence, the level of abatement is given by

(1)\begin{equation}{E_{it}} = {I_{it}} - {m_{it}},\end{equation}

where E _it is the abatement level, I _it is the incipient pollution of a country, and m _it is the pollution level in any given country i at time t.

Stern et al. (Reference Stern, Common and Barbier1996), among others, argue that growth has a positive impact on incipient pollution, ${I_{it}} = f({Y_{it}}),I^{\prime} (.)\gt 0$, where Y _it represents the economic wellbeing of a country at time t. For instance, when developing countries start to grow, they begin to industrialize and thus the incipient pollution level, I _it, also rises. Economic growth often translates into increased consumption levels, which also increases incipient pollution. This scale effect could be greater than the effect that countries experience when moving into the service industries and out of manufacturing, the so-called composition effect (Grossman and Krueger, Reference Grossman and Krueger1995). Countries start to better protect their environment when citizens' demand for higher environmental quality is large enough. We would then expect that environmental abatement, E _it, increases alongside economic growth. This is especially true if there is a high income-elasticity for environmental standards (Antle and Heidebrink, Reference Antle and Heidebrink1995). If we assume that countries' environmental policies reflect individual preferences, then it is plausible to assume that environmental policy totally adjusts to these preferences, and hence

(2)\begin{equation}(\partial {E_{it}}/\partial {D_{it}}) = 1,\end{equation}

which results in:

(3)\begin{equation}(\partial {m_{it}}/\partial {Y_{it}})\lt 0\textrm{ if } \epsilon \gt (\partial {I_{it}}/\partial {Y_{it}})({Y_{it}}/{I_{it}}),\textrm{ where } \epsilon \equiv (\partial {D_{it}}/\partial {Y_{it}})({Y_{it}}/{D_{it}}),\end{equation}

where ε represents the income elasticity of the demand for environmental quality; D _it represents the demand for environmental quality in country i at time t; Y _it equals the economic wellbeing; and I represents incipient pollution (Magnani, Reference Magnani2000). This shows that if the demand for environmental quality is high, then in the latter stages of growth there will be an increase in environmental abatement and hence more stringent environmental policies. For instance, when individuals have a high-income elasticity of demand for environmental quality, the level of environmental protection will be highly dependent on how much income they have. It is only when the income elasticity of demand is high that the income levels play a strong role in improving environmental outcomes. The income elasticity curve shows why the EKC predicts an inverted U shape with respect to economic growth. However, as Magnani (Reference Magnani2000) pointed out, it is worth questioning whether economic growth alone is enough to predict environmental abatement.

We hypothesize that inequality and freedom of the press also play a role in determining environmental abatement. If the power to elect change is in the hands of too few individuals, then it is possible that the outcome will disproportionately benefit those in power. When economic means and political power are concentrated in the hands of a few, those at the top of the income distribution will have a greater chance of having their preferences met as opposed to those at the bottom of the spectrum.

Magnani (Reference Magnani2000) applies the median voter theorem (Congleton, Reference Congleton, Rowley and Schneider2004) when investigating taxes on pollution, as the preferences are single-peaked and there is a monotonic relationship between the voter's relative income and his or her ideal policy. A monotonic relationship refers to the fact that voters' preferences most likely align with their income levels, and that as their income levels change, their preferred environmental policy also adjusts accordingly. If τ represents taxes, l represents an individual, m is the index for the median voter, γ_l is the preference parameter, and R = (y _m/Y) – meaning R represents the ratio between median income, y_m, and average income, Y – then a politician is able to maximize the indirect utility function of the median voter and the equilibrium tax rate is given by

(4)\begin{equation}\tau^{\ast} = 1 - R(1/{\gamma _m}),\end{equation}

to get $\varepsilon \equiv (\delta {\gamma _m}/\delta R)(R/{\gamma _m})\gt 1$. This shows that if the income elasticity $\varepsilon$ of the median voter's preference γ_m for lower levels of environmental degradation is greater than 1, then pro-environment public expenditure is an increasing function of income equality; thus

(5)\begin{equation}E^\ast = f(Y,{y_m}/Y).\end{equation}

Moreover, according to Magnani (Reference Magnani2000), whether or not abatement would respond to increasing per capita income positively depends on the interdependence between economic inequality and average income. For this reason, the effect of income inequality on environmental stringency depends on the level of per capita income.

In addition, according to Barrett and Graddy (Reference Barrett and Graddy2000), the level of abatement and the induced environmental-policy response do not only depend on the prosperity of the countries, but also on the possibilities that individuals have to acquire information about the level of environmental quality. For instance, those policy responses will depend on the level of political and civil freedoms.

Therefore, we test the reduced form in (5) in the next section of this paper, extended with a component that reflects the level of information of the citizens and a second that accounts for the structure of the economy.

4.2 Model specification

The model specification follows the same logic as Magnani's (Reference Magnani2000), and builds off of the previous research on environmental standards that often focused on per capita emissions; we instead focus on environmental policy as the outcome variable. The baseline model includes GDP per capita (x _it), the net Gini coefficient (r _it), the index of freedom of the press (p _it), and the share of manufacturing/services value added in total va (va_it),

(6)\begin{equation}{e_{it}} = \alpha + {\beta _0}\,{x_{it}} + {\beta _1}\,{r_{it}} + {\beta _2}\,{p_{it}} + {\beta _3}\,v{a_{it}}\ + {\varphi _t} + {u_{it}},\end{equation}

where e _it is our environmental policy variable of interest. The error term can be decomposed (Magnani, Reference Magnani2000) into unobservable country-specific effects, represented by μ_i, and a white noise component denoted $\varepsilon$_it, with u _it = μ_i + $\varepsilon$_it. The unobserved effects are controlled for by estimating (6) using panel data techniques. Technology progress is assumed to be common to all countries and time variant and is captured by the year-specific dummy variables, φ_t, which are added to all models.

In order to test if there is a non-linear relationship between GDP per capita and environmentally-friendly legislation – such as posited by the EKC - we include the square of GDP per capita in our second specification, and in a third one we investigate whether the effect of inequality on the outcome variables may be non-linear by adding also the squared term of the Gini.

Furthermore, we investigate if the inequality effect depends on the level of development in the country through the interaction term between the Gini variable and the GDP per capita variable, as hypothesized by Magnani (Reference Magnani2000) and Grunewald et al. (Reference Grunewald, Klasen, Martinez-Zarzoso and Muris2017), and also consider an interaction term between our two target variables, namely, lack of freedom of the press and the Gini coefficient.

Panel data methods allow us to control for unobserved heterogeneity. Fixed effects methods treat the unobserved differences between countries, μ_i, as a set of fixed parameters that can be directly estimated using country dummies or taking within differences of the variables in the estimating equations (Allison, Reference Allison2009). As an alternative, the random effects model uses the unobserved heterogeneity to construct the variance of the residuals. In a generalized least squares procedure, the estimated variance is then used to re-weight the observations. However, this is only valid if the unobserved effects are uncorrelated with all of the explanatory variables. We use a regression-based Hausman-type test to determine if we should use random effects (RE) or fixed effects (FE) (Hausman, Reference Hausman1978; Mundlak, Reference Mundlak1978). If the test rejects the null hypothesis that the estimates from both RE and FE are statistically similar, it is likely that the RE model is biased (Wooldridge, Reference Wooldridge2016). The reason is that FE is still unbiased whether or not the unobserved variables are correlated with the explanatory variables. We tested the hypothesis of no systematic difference between FE and RE and found that it was rejected by the test for the smaller sample of OECD + BRIICS countries, but only in some cases for the second sample and thus we present the corresponding results in our main analysis. The results from the RE regressions augmented with the averages of the time variant variables (Mundlak approach) are located in the appendix,Footnote ⁵ in tables A2 and A3. Those results are used to run a regression-based Hausman test, which tests for the joint significance of the averages of the time-variant coefficients. The test results are shown at the bottom of tables A2 and A3. Regarding year FE, φ_t, those are included in all the regressions as a proxy for technology progress. It is important to notice that whereas the time specific effects explain 39 per cent of the variation of the first dependent variable (environmental policy stringency, EPS), they only explain less than 1 per cent of the second, environmental tax revenue per capita (ETR). The country specific effects instead explain 34 per cent of the variability of EPS and 94 per cent of the variability of ETR. In addition, given the nature of the dependent variables, we also estimated a panel-Poisson model, whose results can also be found in the appendix, in tables A4 and A5.

4.3 Main results

The models specified in the previous subsection have been estimated for two different dependent variables and the corresponding sample of countries using data over the period 1994–2015. Panel data methodology has been used to account for unobserved country-specific heterogeneity that is country specific and time invariant. For the sample of OECD + BRIICS, an RE model was always rejected according to the Hausman test,Footnote ⁶ whereas for the sample of 82 countries this was not the case.

Table 2 shows our results for the panel regression using two-ways fixed – country and time FE – considering as explanatory variables of environmental policy stringency: GDP per capita, the Gini coefficient, lack of freedom of the press, and the share of manufacturing value added in the economy.

Table 2. Main results for stringency of environmental laws

Notes: Robust standard errors clustered by country in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1. The dependent variable is stringency of environmental laws. GDP per capita is in US$1,000. Country and year dummies added in all columns, not shown to save space. Sample of 33 countries over the period 1994–2015.

Dependent variable: environmental stringency.

The results in specification (1) of table 2 indicate that the coefficient of lack of freedom of the press had the expected negative and significant correlation with environmental stringency at the 5 per cent level. GDP per capita shows a positive coefficient, significant at the 1 per cent level, which means that as people's incomes increase, their concern for the environment also increases. In particular, an increase of GDP per capita of US$13,500 is associated with an increase in the index of environmental policy stringency by 1 standard deviation. However, the Gini coefficient is not statistically significant. It is worth noticing that the Gini coefficient had a negative and significant coefficient when the model was estimated without country FE.

In specification (2) in table 2, we test whether the relationship between per capita income and environmental stringency is non-linear, as suggested by the EKC hypothesis. We add the square of GDP per capita as a regressor and find that it is not statistically significant at conventional levels, thus implying that the relationship between GDP per capita and environmental stringency is linear and environmental stringency monotonically increases with GDP per capita. The statistical significance of lack of freedom of the press is now weaker in this model, as expected. Similarly, we test whether the Gini coefficient has a non-linear relationship with environmental stringency and find that the squared Gini coefficient is not statistically significant (column (3)). In specification (4) of table 2, the interaction between GDP per capita and the Gini coefficient is added to the model to test whether the effect of inequality varies with income. We find that the corresponding coefficient is also not statistically significant. In all specifications we allow for the model to be flexible by allowing for non-linear time effects through the use of year dummies. The coefficients of the year dummies are (jointly) significantly different from zero and explain almost 40 per cent of the variability in the dependent variables, as pointed out before. Freedom of the press is found to be significant in determining environmental stringency in all models and has the expected negative sign.Footnote ⁷ If there is less freedom of the press (the index increases from 20 to 40, for example), environmental policy stringency will decrease by 0.3 (since the EP index varies from 0.33 to 4.13, this is an economically non-negligible effect). In columns (4) and (5), we add two measures that are closely related to lack of freedom of the press, namely, lack of civil liberties and lack of political rights, respectively. The former enters the model with the expected sign, but its coefficient is not statistically significant, whereas the latter is statistically significant only at the ten per cent level and does not change the sign or the significance of the coefficient obtained for lack of freedom of the press. Since the correlation between the three proxies is high (around 0.91 per cent), we also estimated the model with only the latter two variables and the results indicate that neither civil liberties nor political rights are statistically significant, not even at the 10 per cent level.

In table 3, we present the main results for the second dependent variable considered, namely environmental tax revenue per capita. As indicated by the results of the regression-based Hausman test (table A3 in the appendix), an FE specification is preferred when the squared term of GDP per capita is included in the model, otherwise the RE specification is preferred. Therefore, we estimate the model with RE in all columns of table 3, apart from (3) and (4).

Table 3. Main results for total environmental tax revenue per capita

Notes: Robust standard errors clustered by country in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1. The dependent variable is total environmental tax revenue per capita (in US$1,000). GDP per capita is in US$1,000. Country and year dummies added in all columns, not shown to save space. Sample of 82 countries over the period 1994–2015. FE (RE) denotes that a model with country fixed effects (random effects) has been estimated.

Dependent variable: environmental tax revenue per capita

For the first specificationFootnote ⁸ in table 3 we find that, as expected, less press freedom is negatively correlated with environmental tax revenue per capita, with a coefficient significant at the 5 per cent level in model (1). This means that when the press in a country starts to become less free (their freedom of the press score moves from a 10 to a 30, for example), then the amount of environmental tax revenue per capita decreases by about €52. This could be the case as people have less reliable information available to them, which is needed if they are going to make an informed decision regarding environmental policy. Also, similar to our regressions with environmental stringency, GDP per capita shows a positive and significant coefficient at the 1 per cent level and the coefficient of the Gini has the expected negative sign, which is statistically significant at the 5 per cent level. An increase in the Gini of 10 percentage points is associated with a decrease in environmental tax revenue per capita of about US$83.

In specification (2) the squared of GDP per capita is added, whose coefficient is statistically significant at the 5 per cent level, indicating that there is a non-linear relationship between income and the outcome variable. Freedom of the press loses its significance under this specification. We also test whether there is a non-linear relationship between the Gini coefficient and environmental tax revenue per capita and find that the coefficient of the Gini squared is not statistically significant at conventional levels in column (3) of table 3. Given that the turning point for the EKC with respect to GDP per capita is out of sample, with the only exception being Luxembourg, we proceed in the next column including an interaction between GDP per capita and the Gini coefficient instead of the squared term.

In the 4th specification of table 3, the results show that the interaction between the Gini and GDP per capita shows a negative and significant coefficient, and the corresponding marginal effect is statistically significant at the 5 per cent level, indicating that the partial negative effect of inequality on environmental taxes increases with income per capita. In columns (5) and (6) we add lack of civil liberties and lack of political rights respectively as control variables. Whereas the former is statistically significant at the 1 per cent level and shows the expected negative sign, the second is not. An improvement in political rights moving from the top to the middle of the ranking (from 7 to 3.5) will be associated with an increase in environmental tax revenues of about €126.50 per capita. In this model, lack of press freedom is not shown to be statistically significant, but since the correlation between this variable and civil liberties is around 90 per cent, this is not surprising. Finally, column (7) includes as a control variable the value added in the manufacturing sector as a percentage as total value added and shows that it is not relevant in this model and even reduces the sample size. Similar to the models in table 2, time FE have been added in all models and are jointly significant, however in this case they explain less than 1 per cent of the variability of the dependent variable.

In our next estimations, we divide up the sample of countries by income per capita levels into two groups. We create a ‘high income’ group and a ‘middle-low income’ group using the World Bank definition in 2016 (countries with GNI per capita higher than US$12,000; see table A1 in the appendix). The reason for doing that is that the correlations of the Gini coefficient and freedom of the press with environmental stringency and environmental tax revenue per capita could vary depending on how rich countries are. For instance, when countries have a lower GDP per capita, they are more concerned with catching up to the rest of the world than with protecting the environment. Thus, they might vote for policies that would increase their economic wealth as opposed to protecting the environment. On the other side of the spectrum, when countries reach a certain level of GDP per capita, citizens' preferences may change and cause them to care more about the environmental quality. This is the idea behind the EKC. It could be the case that inequality and freedom of the press only play a significant role for certain income levels, which is the reason why we divided up the sample into two groups.

Table 4 shows the results when environmental stringency is used as a dependent variable and for above-average income per capita (seen in the left side of the table, specifications (1)–(4)), and below-average income per capita (seen in specifications (5)–(8)). Similar results are presented for the second outcome variable, environmental tax revenue per capita, in table 5. When investigating the effect of the Gini coefficient in table 4, we see that the Gini only has a significant effect for specifications (5) to (8) and the effect is non-linear, suggesting that the inequality has a negative effect on the outcome variable only before a certain inequality level is reached (the turning point is for Gini = 46.25). Therefore, rising inequality is in general negatively correlated with the income dedicated to protecting the environment and could indicate that more egalitarian societies tend to better protect the environment.

Table 4. Results by income group for stringency of environmental law

Notes: Robust standard errors clustered by country in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1. Sample split by GDP per capita. GDP per capita is in US$1,000. Country and year dummies added in all columns, not shown to save space. Sample from 1994–2015 with 25 countries in left panel with GDP per capita>12 in 2015, and 8 in the right panel (listed in table A1, sample 1). Countries with lower income per capita are: Brazil, China, India, Indonesia, Mexico, Russia, South Africa and Turkey.

Table 5. Results by income group for environmental tax revenue per capita

Notes: Robust standard errors clustered by country in brackets. *** p < 0.01, ** p < 0.05, * p < 0.1. Sample split by GDP per capita. GDP per capita is in US$1,000. Country and year dummies added in all columns, not shown to save space. The sample includes 36 high-income countries with GPD per capita>12 in 2015 in the left-panel and 46 in the right panel (listed in table A1, sample 2).

The group of countries considered here are Brazil, China, India, Indonesia, Mexico, Russia and South Africa; therefore it is important to remark that, for these new industrialized countries, both freedom of the press and inequality play an important role in shaping environmental policy stringency. Concerning inequality, reductions in inequality will favor increases in environmental stringency for not very high levels of Gini. However, the Gini coefficient is not statistically significant for rich countries in any of the specifications. Concerning lack of freedom of the press, the results indicate that this variable is strongly relevant for specifications (5) to (8), indicating that for new industrialized countries more freedom of the press (decrease in the variable) will be positively associated with more environmental policy stringency.

The results shown in table 5 when the outcome variable is environmental tax revenue per capita are along a similar line as those found for environmental stringency. In particular, less freedom of the press contributes to more environmental revenues per capita for rich and poor countries. But for rich countries, the effect vanishes once the EKC specification is considered in model (2), whereas for less-rich and poor countries, the coefficient is statistically significant after including the squared GDP term and the squared of the Gini in specification (7) and also when including the interaction between Gini and GDP per capita in (8). Hence, a non-linear relationship is also found for the Gini index, which is more prevalent for specifications (5)–(8) including less-rich countries.

4.4 Robustness checks

As robustness, we first estimated several variations of the models presented in the previous section. When including the interaction between the Gini and lack of press freedom, it was never statistically significant. The same was the case for the interaction between Gini and political rights.Footnote ⁹

As a second robustness check, we estimated a model with interactions between the dummy variable for high-income countries and all the regressors and tested for the equality of the coefficients. The test indicates that only for the coefficients on the income variables (GDP per capita and GDP per capita squared) and the time dummy variables can we reject the null hypothesis that the coefficients are equal. Indeed, it can be seen in tables 4 and 5 that the turning points for the income variable are very different in the left- and right-hand side of the tables.

Finally, we estimated a Poisson pseudo maximum likelihood model, due to the fact that the dependent variables have limited support. The results are presented in tables table A4 and table A5 in the appendix and in general confirm our main outcomes.