The co-pollutant cost of carbon emissions: an analysis of the US electric power generation sector

Emission data for CO₂ and the co-pollutants SO₂ and NO_x from the electric power generation sector at the utility-level⁷ are obtained from the US Energy Information Administration (EIA 2019a, 2019b). These two co-pollutant species are responsible for >80% of electric power generation's early death impacts (Dedoussi et al in preparation). Using this data, we develop methods to estimate simultaneous and autonomous trends in co-pollutant emissions.

We start by defining each co-pollutant $c$'s co-pollutant intensity ${\rm{C}}{\rm{P}}{{\rm{I}}}_{c}$ as the ratio of co-pollutant emission mass, ${\rm{C}}{{\rm{P}}}_{c},$ to CO₂ emission mass:

$$\begin{eqnarray}&&{\rm{C}}{\rm{P}}{{\rm{I}}}_{c}=\displaystyle \frac{{\rm{C}}{{\rm{P}}}_{c}}{{\rm{C}}{{\rm{O}}}_{2}}.\end{eqnarray} \tag{ 2 }$$

Using this metric, we decompose annual changes in co-pollutant emissions $\dot{\text{(CP)}}$ into (i) a carbon effect $\dot{\text{(CO}{}_{2})},$ which results from changes in use of fossil carbon; and (ii) a co-pollutant-intensity effect $\left(\dot{\text{CPI}}\right),$ which captures changes in co-pollutant intensity over time. This yields equation (3)

$$\begin{eqnarray}&&\mathop{{\rm{\text{CP}}}}\limits =\mathop{{\rm{C}}{{\rm{O}}}_{2}}\limits +\mathop{{\rm{\text{CPI}}}}\limits .\end{eqnarray} \tag{ 3 }$$

The $\dot{\text{CPI}}$ can further be decomposed into (i) induced changes that are contemporaneous with carbon emissions; and (ii) an autonomous component that is orthogonal to CO₂ trends. For example, if regulatory policies deliberately prioritize CO₂ reductions from sources with higher-than-average co-pollutant intensities, declining carbon emissions will cause induced reductions in the average co-pollution intensities. In contrast, regulatory policies exclusively targeting co-pollutant emissions would result in autonomous reductions of co-pollutant intensities. Using γ to denote the induced component in co-pollutant intensity changes, we rewrite equation (3) to

$$\begin{eqnarray}&&\mathop{{\rm{C}}{\rm{P}}}\limits =\mathop{{\rm{C}}{{\rm{O}}}_{{\rm{2}}}}\limits +\gamma \mathop{{\rm{C}}{\rm{P}}{\rm{I}}}\limits +\left({\rm{1}}-\gamma \right)\mathop{{\rm{C}}{\rm{P}}{\rm{I}}}\limits .\end{eqnarray} \tag{ 4 }$$

From equation (4), we infer $\delta ,$ the simultaneous effect of changes in carbon emissions on co-pollutant emissions, which captures the direct carbon effect and the induced effect, as shown in equation (5)

$$\begin{eqnarray}&&\delta =\dot{\text{CO}{}_{2}}+\gamma \dot{\text{CPI}}\end{eqnarray} \tag{ 5 }$$

To empirically decompose observed changes in co-pollutant emissions into (i) the simultaneous change of carbon emissions and co-pollutant emissions $\delta ,$ and (ii) the autonomous change in co-pollutant intensities, we model the autonomous change as a fixed, state-specific trend parameter and $\delta$ as a proportional function of changes in carbon emissions. To consider heterogeneity in the simultaneous trends in co-pollutant emissions and carbon emissions, we estimate this for the different state clusters shown in figure 1: (a) states that decreased electricity production versus states that increased electricity production to allow, for example, for the possibility that states that increase EG prioritize cleaner sources to do so; and (b) states that decreased carbon emissions versus states that increased carbon emissions, thereby capturing potential asymmetric responses of co-pollutant emissions to changes in carbon emissions, e.g. if states that move to decrease CO₂ emissions target dirtiest plants first.

Zoom In Zoom Out Reset image size

This yields equation (6), which is estimated for each state $i$ and each year t

$$\begin{eqnarray}&&{\dot{\text{CP}}}_{i,t}={\alpha }_{i}+{\beta }_{{c}_{i}}\dot{{\text{CO}}_{2it}}+{\varepsilon }_{it},\end{eqnarray} \tag{ 6 }$$

where ${\alpha }_{i}$ is the autonomous change in co-pollutant emissions in state $i,$ ${\varepsilon }_{i,t}$ is an error term with conventional properties, and ${\beta }_{{c}_{i}}$ is the state-cluster-${c}_{i}$-specific simultaneous CO₂-co-pollutant effect. In this model, ${\beta }_{{c}_{i}}\gt 0$ implies that CO₂ reductions yield simultaneous reductions in co-pollutants. In contrast, ${\beta }_{{c}_{i}}\lt 0$ would imply trade-offs between CO₂ emissions and co-pollutant reductions. Using logarithmic first differences of emission data⁸ , equation (6) is estimated using weighted least squares with carbon emission weights and robust standard errors. It is estimated for both aggregate state-level emissions as well as by fuel type to capture heterogeneities in the co-pollutant trends by fuel type. The models do not set out to identify causal impacts, but they decompose trends in co-pollutant emissions into components that are contemporaneous and orthogonal with respect to carbon emissions.

To estimate the economic costs associated with co-pollutant emissions, we follow approaches that quantify air quality impacts through early deaths attributable to long-term effects of population exposure to fine particulate matter, PM_2.5, including secondary PM_2.5 resulting from SO₂ and NO_x emissions (e.g. Caiazzo et al 2013, Dedoussi and Barrett 2014, Fann et al 2013). PM_2.5 and ozone are the most significant known causes of early deaths due to degraded air quality, with PM_2.5 being responsible >85% of these for the electric power generation sector (US EPA 2011, Dedoussi et al in preparation). PM_2.5 exposure has therefore become the predominant metric for quantifying the impacts of degraded air quality from electric power generation.

For estimating PM_2.5-attributable early deaths, we build on Dedoussi et al (in preparation), who apply a state-of-the-art adjoint approach to trace early deaths in every state to emissions from all states in the contiguous US⁹ Specifically, we use source state-level early death data specific to the electric power generation sector for the years 2005 and 2011. The emission species taken into account are SO₂, NO_X, primary PM_2.5, and NH₃. Emissions data for CO₂ and the aforementioned co-pollutants are obtained from US EPA's National Emissions Inventories (NEI) (US EPA 2017b, 2018b).

We quantify the CPCC for year $t$ and state $i$ as shown in equation (7):

$$\begin{eqnarray}&&{\rm{C}}{\rm{P}}{\rm{C}}{{\rm{C}}}_{i,t}={\omega }_{i,t}\,{\rm{v}}{\rm{s}}{{\rm{l}}}_{t}/{{\rm{CO}}}_{{2}_{i,t}},\end{eqnarray} \tag{ 7 }$$

where ${\omega }_{i,t}$ is the number of premature mortalities attributable to electric power generation emissions from state $i$ and year $t$ obtained from Dedoussi et al (in preparation), ${\rm{v}}{\rm{s}}{{\rm{l}}}_{t}$ is the VSL in year $t,$ and ${\rm{C}}{{\rm{O}}}_{{2}_{i,t}}$ is CO₂ emissions from electric power generation reported for state i in year t. The source-wise mortality estimates include all PM_2.5 premature deaths attributable to a state's emissions regardless of the state in which the deaths occur. VSL estimates are obtained following the methodology of US EPA (2014). We adjust the price level of the VSL estimates to year-2017 USD using the GDP price deflator for the US obtained from Federal Reserve Economic Data, and adjust for real income changes using per capita income growth obtained from Federal Reserve Economic Data in combination with an income elasticity of VSL of 0.7 as recommended by US EPA (2016d) and Robinson and Hammitt (2015). Consistent with US EPA practice (US EPA 2004), we use a cessation lag structure that distributes early deaths over a span of 20 years following exposure (30% in year 1, 50% in years 2 through 5, and 20% in years 6 through 20). We assume a 3% discount rate and apply CBO (2017) GDP-per-capita growth assumptions to consider future early deaths. To examine the sensitivity of our results to these assumptions, we also report results using a constant VSL. Finally, we estimate the US national timeline of CPCC between 2002 and 2017, using the NEI emission trends for NO_x, SO₂, primary PM_2.5, and NH₃ for that period and the 2005 and 2011 estimates of early deaths per unit of emission described above.

We start by analyzing emission distributions and trends as outlined in section 2.1. As shown in table 1, the ten states with the highest emissions by species accounted for 46%–65% of total emissions from the electric power generation sector in 2017. This concentration is partly driven by the amount of electric power generation in each state. However, differences in power generation technologies, e.g. the uptake of renewable sources, as well as different fuels, power generation efficiencies, and varying levels of emission control technologies, cause significant variation in the rankings.

This variation is consistent with the observed variations in co-pollutant intensities shown in figure 2. We find that, on average, 0.72 kg SO₂ and 0.70 kg NO_x were emitted per metric ton of CO₂ released from EG for 2017, but these ratios varied among states. In general, we find co-pollutant intensities to be highest in the Midwest states, pointing towards greater use of fuels or power generation associated with higher pollution intensities.

Zoom In Zoom Out Reset image size

Turning to changes in emissions from US electric power generation between 2002 and 2017, we find a nationwide decline in CO₂ emissions by 1.8%/yr. On the state-level, we observe considerable variation in CO₂ trends across states as shown in figure 3(a).

Zoom In Zoom Out Reset image size

Over the same period, nationwide co-pollutant emissions declined at 13.7%/yr for SO₂ and 8.8%/yr for NO_x. Figures 3(b), (c) again show considerable inter-state variations in the trends. These can be attributed to a variety of reasons including decarbonization efforts, fuel switching (e.g. coal to natural gas), more stringent air pollution regulations, and advances in pollution control technologies.

To partition the observed changes in co-pollutant emissions into components that are (i) simultaneous and (ii) autonomous with respect to changes in carbon emissions, we apply the model outlined in section 2.1. The estimation results are presented in table 2.

Table 2.  Summary of regression results.

	ln(SO2)						ln(NOx)
	(1)	(2)	(3)	Coala	Natural Gasa	Petroleuma	(1)	(2)	(3)	Coala	Natural Gasa	Petroleuma
ln(CO2)	1.51*** (0.13)						1.25*** (0.12)
ln(CO2)
States with CO2 decrease		1.57*** (0.14)						1.25*** (0.12)
States with CO2 increase		0.78* (0.35)						1.12*** (0.15)
ln(CO2)
States with electricity generation decrease			1.45*** (0.17)						1.55*** (0.22)
States with electricity generation increase			1.55*** (0.18)						1.05*** (0.11)
ln(CO2)
Coala				1.27*** (0.08)						1.21*** (0.08)
Natural gasa					0.55*** (0.14)						0.59*** (0.09)
Petroleuma						0.95*** (0.08)						0.82*** (0.06)
${\bar{{\boldsymbol{\alpha }}}}_{{\boldsymbol{i}}}$	−0.12***	−0.11***	−0.12***	−0.10***	0.02	−0.11	−0.07***	−0.07***	−0.06***	−0.07***	−0.06	−0.08
N	731	731	731	672	720	730	731	731	731	672	720	730
${{\boldsymbol{R}}}^{2}$	0.46	0.46	0.46	0.57	0.21	0.48	0.51	0.51	0.52	0.59	0.31	0.67

Note. Standard errors reported in parentheses. For ${\bar{\alpha }}_{i},$ the joint significance of all state fixed effects is reported as significance. Table reports weighted average of ${\alpha }_{i}$ using CP level weights. State-level results are shown in figure 4.* Significant at 5% level** Significant at 1% level.*** Significant at 0.1% level. ^aEmissions attributable to specified fuel only.

For the entire set of states (Model 1), a 1% change in CO₂ emissions coincided with changes in the same direction in co-pollutant emission of 1.51% for SO₂ and 1.25% for NO_x. The elasticity >1 implies that CO₂ emissions have an induced as well as direct effect on co-pollutant emissions. When we differentiate between states that reduced or increased CO₂ emissions over the period (Model 2), we find states that reduced CO₂ emissions to have a significantly higher elasticity in SO₂ emissions than states that increased CO₂ emissions. Similarly, states that reduced EG are found to have a significantly higher elasticity in NO_X emissions than states that increased electricity production (Model 3). These findings are consistent with the expectation that states reducing CO₂ emissions or electricity production tend to target dirtier sources first.

The autonomous co-pollutant trends by state, estimated by the parameter ${\alpha }_{i},$ are shown in figure 4. Their average values, reported in table 2, range from −12%/yr in the case of SO₂ to −7%/yr in the case of NO_x. In comparison, the average simultaneous CO₂ and co-pollutant emission changes are smaller (−2.7%/yr in the case of SO₂ and −2.2%/yr in the case of NO_x).

Zoom In Zoom Out Reset image size

When we estimate the model separately by fuel type, we find that autonomous reductions in co-pollutant emissions were significant only for coal-fired power plants (−10%/yr for SO₂ emission and −7% for NO_x emissions). Our estimates also indicate that a 1% reduction in CO₂ emissions coincided with a reduction of co-pollutant emissions by more than 1% for coal power plants only. The elasticities in the fuel-specific models are lower compared to those of the aggregated model (Model 1) since the latter captures fuel switching from coal to natural gas (figure 5) which the fuel-specific models do not.

Zoom In Zoom Out Reset image size

Using the methods described in section 2, we calculate that the national-level CPCC for the electric power generation sector in 2011 was $44.7/mt CO₂ (in 2017 USD). This is ∼20% higher than the IAWG SCC of $37.3/mt CO₂, and ∼8 times larger than the revised SCC of $5.5/mt CO₂ used by the Trump administration (both in 2017 USD), which only considers US climate damages¹⁰ . In other words, the CAQSC of combustion emissions, is found to be about 1.2 times as high as the SCC used by the Obama administration, and nearly an order of magnitude higher than the revised SCC used by the Trump administration.

Inter-state variations in the year-2011 CPCC, reported in table 3, reflect differences in co-pollutant intensity (emissions per metric ton of CO₂), the dispersion of co-pollutants, background atmospheric composition, and population densities in impacted locations. Overall, the results indicate a higher CPCC in mid-Atlantic and Midwestern states. In 2011, New Jersey led the nation with a CPCC >$95/mt CO₂, a result largely attributable to impacts originating from New Jersey power generation emissions on the population in New York City and environs.

Table 3 also reports the CPC-EG, and the SCC Emissions from EG using the SCC values from the IAWG and the current US administration (both at 3% discount rates) per megawatt-hour (MWh) of electricity produced in each state. These reflect the effects of efficiency improvements, fuel switching, and decarbonization. Some states, such as California, Maine and Oregon, had very low CPC-EG and SCC-EG, resulting from their greater reliance on non-fossil based energy sources (61% of power generation in 2017). The average CPC-EG of electricity production was relatively high in the mid-Atlantic and Midwestern states, reaching ∼$70 per MWh in Ohio and Kentucky. These states also rank high in SCC-EG by virtue of more carbon-intensive electricity production.

The ratio of CPC-EG to SCC-EG in 2011 for each state is shown in figure 6. Again we find substantial inter-state variations, with the highest ratios in the Midwest and mid-Atlantic states.

Zoom In Zoom Out Reset image size

While the CPCC was comparable in magnitude to the IAWG SCC in the year 2011, one could expect the autonomous downward trends in co-pollutant emissions to have led to significant declines in the CPCC over the period under investigation. To assess this, we compute yearly CPCC estimates averaged across all states using the method outlined in section 2.2, in combination with the year-2005 and year-2011 estimates of PM_2.5 premature mortalities per unit of emission by species from Dedoussi et al (in preparation), and yearly US aggregate emission data obtained from US EPA (2016b, 2016c)¹¹ .

The results are shown in figure 7. We find that the CPCC declined by 71% and the CPC-EG by 79% between 2002 and 2017. These outcomes can be attributed largely to the reductions in precursor emissions (section 3.1). The slightly larger reduction in CPC-EG is due to lower CO₂/MWh intensities resulting from grid decarbonization, fuel switching (see figure 5), and efficiency improvements.

Zoom In Zoom Out Reset image size

If we compare the excess mortality associated with co-pollutant emissions to that in a counterfactual scenario where the CPCC and CO₂ emissions remained constant at 2002 levels, we calculate using the time-varying VSL that the savings resulting from reductions in co-pollutant emissions at power plants amounted to ∼$1.01 trillion (in 2017 USD) over the sixteen-year period.