Ground Truth Trekking

Methods and Analysis

 Energy Prices and Generation Mix:

Data on yearly energy prices and generation mix in each state from 1990 to 2009 were acquired from the Energy Information Administration, which provides numbers for the "Average Price by State by Provider" and "Net Generation by State by Type."  For the analyses in this study, all values were adjusted for inflation to 2009 dollars using the Consumer Price Index (CPI).

In our analysis, we lumped natural gas and other gases together into a single "Gas" category.  We also lumped different types of biomass into a single "Biomass" category.  We disregarded "Pumped Storage" as a generation type.

Externalized Costs Data Sources:

Externalized costs are difficult to calculate, and contain a fair amount of uncertainty. A 2011 study by Paul R. Epstein et al. estimated the externalized costs of coal-fired power to be an additional 17.8 cents per kilowatt hour above the price paid by the consumer. Epstein provided a low, best, and high estimate for each category, and we used the "best" or middle estimate for all our analyses. Some of these costs are coal-specific (like the costs of mountaintop removal mining), while others apply more broadly to other fossil fuels (like costs due to climate change impacts). However, no energy source is entirely free of externalized costs. To accurately compare coal with the other generation sources in our study, we attempted to create the best estimate of externalities for all generation types.

For our base estimate, we used the ExternE study from 2001. This Europe-wide study has the most comprehensive externalities assessment we have found, with estimates across 15 countries and 9 different energy generation types. To convert these numbers to a base estimate in this study we assumed a normal distribution for the ranges in the original data and then took an average across all counties for each generation type. The ExternE study contained no analysis for geothermal energy, which we set equal to wind at 0.19 cents per kilowatt hour. From these averages, we converted from Euros to 2001 US dollars and scaled for inflation to 2009 US dollars.

Although they cover an array of energy sources, the analyses in ExternE were conducted in the 1990s and lack the most up-to-date estimates of climate change and other externalities found in the Epstein study, as well as the costs of mountaintop removal mining (which are US specific).

Externalized Costs Best Estimate Calculation:

To combine these analyses we used the number from the Epstein study for coal, and scaled the ExternE estimates for other generation types according to the relevant elements of the Epstein study. Without scaling we would underestimate externalities in other generation types, since the more recent Epstein coal analysis takes more factors into account than ExternE's older analysis.

To perform this scaling we divided Epstein's analysis into three pieces: coal-specific, fossil-fuel general, and subsidies. Into coal-specific we placed "public health burden in Appalachia" and "abandoned mine lands", which together accounted for 4.8 cents/kwh. Subsidies accounted for 0.16 cents/kwh. Into fossil-fuel general we placed everything else, which included a variety of costs, with the highest being "emission of air pollutants from combustion" at 9.31 cents/kwh and "climate change total" at 3.51 cents/kwh.

Epstein's coal number of 17.8 cents/kwh was used directly as our best estimate for coal externalities. The fossil-fuel general category was used to scale the ExternE results for oil and natural gas. Both these fuels incur externalized costs from drilling and transport, air pollution from combustion, and climate change from the CO2 released. Per kilowatt hour these factors might be smaller for oil and natural gas than for coal, because it takes a larger mass of coal to get a single kilowatt hour of electricity. However, determining how much smaller is difficult. To account for the difference, we assumed that the ratio of coal impacts to oil or gas impacts is equal to the ratio of these impacts found in the ExternE study.

In ExternE, coal and oil externalities were both calculated at 6.27 cents/kwh (2009 US dollars). So oil is equal to 100% of coal, and Epstein's full 12.88 cents/kwh for air pollution and climate change costs were assigned to oil generation. Natural gas, however, was calculated at 1.98 cents/kwh, only 32% of coal's 6.27 cents in the ExternE study. Therefore 32% of the 12.88 cents/kwh for "mining/transport," "air pollution," and "climate change" was assigned to natural gas generation (4.07). Subsidies were not included in the scaling because the original ExternE study already incorporated subsidies.

There are inevitable uncertainties and inaccuracies in the externality calculations, both in the original data and in the assumptions we made to combine them. However, the resulting numbers represent our best estimate– far better than assuming that externalities are in all cases 0. The relative magnitudes of coal, oil, and gas externalities are broadly similar to other studies, and nuclear externalities would only matter if ExternE had under-estimated this number by an order of magnitude.

In the interactive figure, the viewer can use the "Externalities" menu to toggle between our "Best Estimate" (calculated as described here), "ExternE Average" (ExternE data with no input from the Epstein study), or remove externalities entirely by selecting "None."

Levelized Costs:

Levelized costs are an attempt to calculate future energy prices for a not-yet-built plant of a given energy type. Levelized costs take into account the expected costs of power plant construction, maintenance, transmission, as well as fuel costs over a designated period of time. The levelized costs of future power plants are nearly always higher than the costs from already-built generation capacity of a given type. These numbers have a significant amount of uncertainty, particularly for fuel costs, but are the most relevant numbers to look at when planning future generation capacity.

The Energy Information Administration (EIA) publishes estimates of levelized costs for different types of power plants, looking at the expected 30-year cost per kilowatt hour from a power plant constructed today (coming online in 2016). This cost is then annualized and adjusted for inflation to provide a comparative measure between different generation types. While levelized cost does take into account subsidies received by different sources of energy generation, it doesn't account for externalities.


For our analysis, we simplified the EIAs numbers slightly to allow us to compare them to current generation portfolios (Figure 4).


"Advanced Coal" (10.9 cents/kWh) and "Advanced Coal with CCS" (13.6 cents/kWh) were ignored since there are only two of the former in the US and none of the latter.  Therefore these types of facilities don't represent significant options at the present time, particularly in the case of CCS which has never been tested on  large scale at a coal facility.  In our analysis "Coal" is the same as "Conventional Coal" (9.5 cents/kWh) in the EIA's table.


The EIA lists a number of different types of natural gas plants. We ignored ACT and ACG since these technologies are not in widespread use. Our estimate based on utility profiles is that US natural gas plants are approximately 75% CCC plants and 25% CCT plants. Scaling the costs with those numbers, we arrived at an approximate levelized cost for natural gas of 7 cents per kWh.


No value for petroleum generation is given by the EIA, so we were forced to estimate one. We assumed it to be 11 cents/kWh, on the grounds that an oil-fired plant likely costs slightly more than a coal-fired plant. It may be that the EIA did not include this number because of the volatility of oil prices, which would add considerable uncertainty.

Even if this number is significantly inaccurate, it does not change our conclusions.


"Offshore Wind" was ignored as representing a negligible part of current wind generation, and the conventional wind number of 9.7 cents/kWh was used.


In 2009, photovoltaic solar represented around 3/4 of total solar capacity, while thermal solar accounted for the remaining 1/4. We used a weighted average of these two types of solar capacity to arrive at a general levelized cost for solar energy of 23 cents/kWh.

Geothermal, Hydro, Nuclear, and Biomass:

These values were all taken straight from the EIA tables.

Further reading:



All of our statistical results are accessible via the interactive at right.  As we conducted our analyses, we replicated them in this figure, which is directly linked to our database.  For instructions about how to use this interactive figure, see our interactive figure section.

To detect significant correlations between changes in generation portfolio and electricity prices, we tested the significance of simple linear regression associating price in a given state and a given year (Y-axis) with the percentage of electricity generated by a give source (X-axis). For example, we tested whether the percentage of coal could predict price in the state of CO. Each datapoint corresponded to a year. Our statistics were computed using a custom JavaScript library available here.

To segregate significant correlations from those more likely to have arisen coincidentally we applied a cutoff of p<0.05. In many cases, we were applying a number of statistical tests in parallel, which increases the likelihood of coincidentally "significant" correlations (multiple testing problem). To highlight correlations that were so strong that they remain significant in light of the multiple testing problem, we corrected our significance tests by dividing by the number of tests (Bonferroni correction). States that never used a given generation type were not counted in the number of tests (e.g. we only counted 48 tests for coal generation, since two states and DC did not use any coal in any of the years we examined).

The Bonferroni correction is appropriate for filtering out individual states that have significant correlations and for making inferences about the states that do exhibit a significant correlation. However, this correction is conservative and most probably underestimates the number of states that did exhibit significant correlations. Our results primarily emphasize the lack of correlation, in which case the Boneferroni correction is not conservative, since it increases the false negatives (correlations are rejected despite being meaningfully significant.)  Therefore, the interactive figure reports the uncorrected p<0.05 significant results, but highlights the Bonferrroni corrected significant correlations.

Looking at the correlation between coal use and consumer electricity prices, when a weaker p<0.05 threshold for significance was applied, the relationship between coal use and electricity prices depended on the state (Figure 1). Seven of thirteen significant correlations show rising prices with increased coal burning, the other six show the opposite. The remaining 35 coal-burning states show no correlation between coal generation and price. This latter category includes states such as Florida and Colorado, which reduced their use of coal by over 20% and for which correlations should have been readily apparent, had they existed. Results using the stronger (Boneferroni corrected) significance cutoff led to similar conclusions. Finally, when externalities were considered in the analysis of coal use and total costs, the number of positive correlations was 30, with no significant negative correlation (Figure 5). In other words, for 30/48 states, higher percentage of coal use resulted in higher total costs. 

In addition to the statistics presented in this report, we explored a number of avenues that did not yield usefully different results. In particular:


<< Previous Section <<