* Diagram is not drawn to scale.
Independent Variables
The primary variables of interest include utility ownership types and state policies and regulations. Beginning with the former, Form EIA-861 includes seven utility
ownership models: private, cooperative, municipal, Federal Power Marketing
Administration, state power authority or organization, municipal marketing authority (MMA), and county-level subdivision, irrigation district, or utility district. Each of these variables is transformed into a dichotomous variable, equal to one if it appropriately represents the utility’s ownership model and equal to zero otherwise. Throughout the remainder of this study, I refer to all utilities that are independently owned as “private” and all government- or cooperative-owned utilities as “public”.
Electricity Market: the total sample (n=3277)
Utilities and their customers that do not have DG capacity (n=3083) C. Utilities with utility-owned DG capacity (n=153) B. Utilities with consumer- owned DG capacity (n=41)
The policy instrument variables include net metering standards, RPS policies, and interconnection standards. I include a binary net metering variable in the two-part model, equal to one if the utility has net metering customers and equal to zero otherwise. This variable is not included in the consumer-owned versus utility-owned probit models. Instead, as explained above, this variable is used to help distinguish between the two ownership types. The RPS policy and interconnection standard variables are both dichotomous, coded as a one if the policy is active in 2005. Both variables are compiled from the DSIRE database. Any state that enacted a policy during or after November, 2005 is considered to have an inactive policy during the period of analysis and is coded to equal zero.
Additional utility-level characteristics are included as covariates. We control for summer peak power output, measured in megawatts. Since peak load shaving plants are one of the primary DG units included in this analysis, we assume that higher peak capacity will be associated with greater DG deployment. Peak power is the maximum amount of power that was sold in the summer of 2005 during the month and the specific day of highest electricity demand. Total sources of power, another utility-level variable, is the total megawatt-hours of power sold in retail markets over all of 2005. This variable is re-scaled by a factor of 10,000 MW. Therefore each 1 MW of total sources in the summary statistics and regression outputs represents 10,000 MW.
State-level electricity characteristics include the price of electricity and the state’s status of electricity market restructuring. The price of electricity, extracted from EIA electricity data, is the average of electricity prices from all electricity sources per state in 2005. The deregulation variable indicates whether a state has restructured its electricity
market. States that have either partially or fully restructured their electricity market are coded to equal one; states that have kept their market regulated or have “destructured” their market after a period of deregulation have a regulation variable equal to zero.
Finally, I control for the following state demographics: average household income, measured in $1,000 of U.S. dollars; population, measured in 100,000 citizens; and binary regional dummy variables. We include the region variables to control for location-specific dynamics that may affect DG deployment rates. Population and household income data come from U.S. Census Bureau data.
Results
Results of the collinearity diagnostics revealed that multicollinearity between variables did not exist; variance inflation factors were all small and below standard threshold levels and all bivariate correlations were well below .8. The White test did detect heteroskedasticity, however, and so we estimated robust standard errors in the final version of our two-part model to correct this problem.13
Before turning to the empirical results, it is informative to consider dependent variable, and note how well it conforms to the working definition of DG systems, as defined above. Beginning with a discussion of DG applications, Form EIA-861 does not
13 Additional specification tests revealed somewhat conflicting information about the most appropriate form of the utility-type variable. I first conducted a Wald test on both parts of the two-part model to test whether the public utility parameters were equal to each other. The resulting chi-squared test statistics, with four degrees of freedom, were .77 for the first part model and 159.62 for the second part model. I therefore could not reject the null hypothesis of equality in the first part, which estimates likelihood of adoption, but could reject the null hypothesis in the second part, which estimates total capacity conditional on having any at all. A two-part model, however, should ideally include the same set of parameters in both parts of the model. I similarly conducted an LM test on both parts of the model to further explore this issue of
specification. The NR2 from the second part equation, with five degrees of freedom, was 60.16 and thereby significant at all conventional significance levels. For the LM calculation in the first part equation, I adjusted for non-linearities and heteroskedasticity as part of the NR2 calculation. The resulting NR2 estimate was 12.62, which was significant at the 10 percent significance level. I concluded that the utility parameters should not be clumped into one public utility variable but should remain separate variables.
explicitly distinguish between different applications of DG power—for instance, between peak load shaving and micro-generation—with one exception: CHP units are excluded from the DG classification. I can draw additional conclusions about the DG system attributes from EIA-861 supporting data. Figure 3.2 presents the percentage distribution of DG units by fuel type. Distillate fuel and natural gas are the most common types of fuel, which collectively contribute 72.6 percent of the total units. Water and other
renewables provide fuel for 8.5 percent of the DG units in the sample. Figure 3.3 presents the percentage distribution by technology type. The majority of the DG units are internal combustion engines or combustion turbines. Roughly 11 percent of the DG units in the sample come from wind or hydroelectric power. Although it is not demonstrated in either graph, roughly 32 percent on average of the DG capacity is used for back-up power, the majority of which comes from internal combustion engines using distillate fuel. This information does not allow one to fully classify the technology distribution of DG units in this sample according to the above definition, although it does provide a rough picture of DG type and fuel source.14 Based on these attributes, I conclude that the DG variable in this sample primarily represents peak load shaving and backup power, some localized conventional DG plants, as well as an occasional remote power system that is connected to the grid. CHP is not included and micro-generation is hardly included, if at all. 15
14
Figure 3.2. Percent Distributed Generation by Fuel Type
Figure 3.3. Percent Distributed Generation by Technology Type
I further divide DG fuel types according to whether the system is owned by a utility or a customer. Figures 3.4 and 3.5 reveal that utility DG systems are primarily fossil fuel-based, with roughly 78 percent of the systems powered by distillate oil or natural gas. The customer-owned DG systems, on the other hand, include a greater share of renewable fuel types. Customers appear more inclined to adopt renewable-based DG systems than utilities.