Likely the most widely-cited and influential analysis of California’s pollution-equity trends in the first several years of its Cap-and-Invest program is a 2018 paper led by Lara Cushing and Rachel Morello-Frosch titled Carbon trading, co-pollutants, and environmental equity: Evidence from California’s cap-and-trade program (2011-2015).
This paper asks and tackles some critical questions, diving deep into the emissions data from two years prior to the program (2011-12) and during the first 3 program years (2013-15). These questions focus on the issue of whether socioeconomically disadvantaged communities (DACs) have seen, or can expect, reductions in environmental inequities from air pollutant exposure as a result of carbon cap and trade programs, using the California program as a case-study. The authors offer some preliminary insights into initial co-pollutant and environmental justice outcomes of the program.
The key findings
The authors report three key findings:
- “Facilities regulated under California’s cap-and-trade program are disproportionately located in disadvantaged neighborhoods.”
- “Co-pollutant emissions from regulated facilities were temporally correlated with GHG emissions, and most regulated facilities (52%) reported higher annual average local (in-state) GHG emissions after the initiation of trading.”
- “Since California’s cap-and-trade program began, neighborhoods that experienced increases in annual average GHG and co-pollutant emissions from regulated facilities nearby had higher proportions of people of color and poor, less educated, and linguistically isolated residents, compared to neighborhoods that experienced decreases in GHGs.”
Assessing the Key Findings
With the exception of key finding 2, the study’s key findings are robust across the years examined. This does not establish cause and effect with the program nor is it predictive of future results.
People Of Color (POC) and other DACs are disproportionately located near polluting facilities. This is clearly shown in the Table 1 of the original study (Figure 1), and is a known and widespread societal problem. Decreasing emissions from these facilities is essential to achieving environmental justice.
Figure 1: Original study Table 1 concerning characteristics of neighborhoods within and beyond 2.5 miles of a facility
Within finding 2, the assertion that “most regulated facilities (52%) reported higher annual average local (in-state) GHG emissions after the initiation of trading,” is commonly applied in discussions about disadvantaged community (DAC) impacts from emissions trading systems. This finding is heavily dependent on the timeframe selected and power sector trends outside of the programs influence. In addition, the correlation between co-pollutants and GHGs is not straightforward.
Combining findings 2 and 3 (higher proportional share of DACs in areas experiencing an increase in emissions), the conclusion has been drawn that DACs have experienced net harm under the early years of the program. There are several reasons why finding 2 is not robust and the claim of demonstrated net harm, either causal or otherwise, is at the very least an overextension of the study results.
At the population-level, the study found substantially more of the population (6.3 million) experienced decreasing emissions from nearby facilities (within 2.5 miles) than experienced increasing emissions (4.0 million), a net of 2.3 million people, as shown in Table 3 from the original study (Figure 2).
Figure 2: Clip of original study Table 3 concerning full population characteristics sorted by GHG trends of nearby facilities
Review of the study results indicates a significant possibility that all groups considered in the study experienced net benefits. The same study results are indicative that the white population in proximity to these facilities could have been more likely to experience emission reductions in the early years (Figure 3), highlighting the potential inequity in the distribution of net benefits. However, due to limitations in the availability of the original study data, these findings are approximate and cannot be presented with statistical confidence.
Figure 3: Approximation of net share of population experiencing a decrease in nearby GHG emissions, shown by characteristic and for two separate mean estimation approaches
Independent of whether emissions increased or decreased in the first three years of the program, these communities were disproportionately DACs, including majority POC populations (Figure 4). Populations less likely to be impacted in either direction, and therefore more likely to experience the status quo with or without the program, were lower proportion POC.
Figure 4: Reported median and median range of people of color population in census blocks experiencing a GHG decrease (blue), GHG increase (yellow), or not in proximity to a regulated facility (red)
The lead study author has noted that the data do not show cause and effect, and we find that much of the reported data lack any statistical significance or predictive power for subsequent years. Notably, the study authors review of other emissions trading programs “found little evidence that they produced socially inequitable outcomes”.
- Key findings are very sensitive to the choice of baseline year, including the most heavily influenced great recession year of 2011 (Figure 5).
Figure 5: Regulated facility emissions before, during, and after the start of the program (black vertical line) and within the study window (red box)
- Unavoidable and unpredictable developments in the power sector greatly reduced typical availability of zero-carbon power plants (nuclear and hydro) after 2011 (Figure 6). This greatly influenced emissions for the sector of the economy with the most facilities in the study (power generation, 83 out of 322 facilities).
Figure 6: In-state production of nuclear and hydropower from 2001-2018
Data source: https://www.energy.ca.gov/data-reports/energy-almanac/california-electricity-data/california-electrical-energy-generation
- No sensitivity of facility-level or net pollution findings were evaluated based on the choice of baseline years or trends beyond the control of the program, such as availability of the prominent sources of carbon-free electricity.
- The original study does not consider impacts of billions of dollars in investments designed with a specific goal of addressing environmental injustice and equity. Through mid-2020, 55% ($3.5 billion) of implemented investments have gone towards benefits for disadvantaged communities.
More recent data indicates either sharp or continued improvements in overall program outcomes, indicating the original study methodology should be applied to more recent program results. In doing so, a focus on net population exposure and statistical significance is critical.
- Emissions from the facilities examined in the original study were 11% lower in 2016-18 than they were in the first 3 years of the program that were included in the study (2013-15), based on CARB Pollution Mapping Tool data accessed in February 2021;
- A recent study, with two more years of data, noted “that the program reversed previously widening EJ gaps in PM2.5, PM10, NOx, and SOx, narrowing gaps to 2008 levels by 2017”. After steadily climbing prior to 2013, the program reduced emissions at annual rates of between 3% and 9% for all pollutants between 2012 and 2017. The newer study findings depend on pollution transport by winds. Therefore, this assessment does not demonstrate significant program cause and effect, much like the Cushing and Morello-Frosch et al. study does not.
While there is a long-established correlation between GHGs and co-pollutants, these correlations are not particularly tight and include wide variability between and even within sectors and pollution types, as shown in Table 2 of the original study. This indicates that directly and quantitatively regulating both GHGs and co-pollutants under the same mechanism presents major challenges. As reported in Table 2 of the original study:
- Across the five co-pollutant sources considered, the correlation for a 1% change in GHG emissions ranged from 0.48% for volatile organic carbons (VOCs) to 0.91% for air toxics. The other three pollutants had correlations of 0.63% to 0.66%.
- Within a sector, there can be wide variability in pollutants. For example, the largest correlation for air toxics was from “other manufacturers”, with a 2.07% change per 1% change in GHG emissions. However, “other manufacturers” correlation with VOCs was the weakest (-0.17%) of any sector and was also one of the lowest for NOx (0.53%).
- Across sectors, there is also wide variability for a given pollutant. For example, correlation with GHG emissions for 2.5 micron particulate matter (PM2.5) range from 0.00% (metal and machinery manufacturing) to 1.08% (public services). And even within sectors, such as the public service sector for PM2.5, the 95% confidence interval is large (0.26% to 1.09%) and the strength of the correlation is weak. An extreme example is a relatively large sector of co-generation (53 facilities) with no average correlation with GHGs, but a 95% confidence interval of -4.19% to +4.17%.
The Cushing and Morello-Frosch et al. study continues to offer very important insights to the conversation about environmental equity and the design or importance of market-based programs. The methodology and lens are worth applying to all programs on a regular basis, which requires at least a similar level of data availability that California has made publicly accessible. It is critically important to extend this lens both to more recent data and to people or population-centric outputs to provide the most comprehensive perspective possible. Such outputs would emphasize net population exposure metrics alongside this study’s focus on both facility-level trends and the difference in median population characteristics.
As with any research, the results should be interpreted within the context of underlying trends and the wider body of literature. It is always important in presenting or summarizing study results to note when and if statistical significance exists, where there is demonstrated cause and effect, and if results hold predictive power for future results or are consistent with previous studies.
If you are interested in some additional context on the topic of equity under the California Cap & Invest program, these articles from other proponents of deep decarbonization provide informative background reading:
- Cap and Trade-Offs: Did California’s landmark legislation help or hurt the state’s most vulnerable? (Grist)
- Dogmatism on Carbon Pricing Mustn’t Derail Climate Progress (Carbon Tax Center)
- Can California’s cap and trade address environmental justice? (Green Biz)
- Can Climate Efforts Be the “Everything Policy Store”? (Haas Energy Institute)
The quote and summary of the “EJ Gap” study referenced in this paper was updated on March 31 based on revisions made to that study in February 2021. The prior quote no longer appears. In addition, one additional background link was added.
Thank you to Jonah Kurman-Faber, Research Director at Climate XChange, for his feedback and insight during drafting of this memo.
For some more details on the topics discussed in this memo, check out Part 2.
 A prominent example is the letter to the Biden-Harris Transition team regarding Mary Nichols as a candidate to lead the EPA which cites the Cushing and Morello-Frosch et al. study: “As warned by environmental justice advocates, cap and trade has increased pollution hotspots for communities of color in California, exacerbating pollution health and safety harms.”
 The data here are based on the Table 3 of the original study, and determined by a method of using the median as equal to the average, or imputing a mean based on the median and “interquartile ranges” using the online calculation tool at http://www.math.hkbu.edu.hk/~tongt/papers/median2mean.html . Without a complete data set, we cannot determine how closely the mean or average results are to the median results. Applying this approach to Table S2 for a 1-mile radius leads to similar results.
 The strength of correlation expressed as the model fit or R2 value, is below 0.4 for 30 of the 42 reported sector-pollutant combos demonstrating this wide variability and often tenuous correlation with GHG emissions.