Can Free Information Really Accelerate Technology Diffusion?

Can Free Information Really Accelerate Technology Diffusion? RICHARD D. MORGENSTERN and SAADEH AL-JURF

ABSTRACT Significant environmental benefits are often associated with the rapid diffusion of new energy-saving technologies. Over the past decade, electric and gas utilities, as well as the federal government, have provided free technical information and, in some cases, direct cash subsidies, to owners of existing commercial buildings to stimulate investment in specific energy-saving technologies. Yet little is known about the effectiveness of the information component of these programs. Can information itself, without explicit cash subsidies, actually increase investment in new technologies? To examine these issues, a model of retrofit investment in highefficiency lighting technologies is developed. Estimates are based on a sample of commercial buildings, rather than the more common comparison of program participants to a synthetic pairing with another population. The principal finding is that information programs appear to make a significant contribution to the diffusion of high-efficiency lighting in commercial office buildings. Additionally, there is some evidence that the programs are more effective in encouraging retrofits by those who have already invested in advanced lighting technologies than for first-time purchasers.  1999 Elsevier Science Inc.

Introduction Public policies which accelerate the development and dissemination of new technologies may, over the long term, be one of the most important tools for environmental protection. Such policies have the potential to lessen the trade-off between conventionally measured economic well-being and environmental quality. For example, they may lower the costs of reducing greenhouse gas emissions. While it is politically expedient to support policies and programs that lower economic costs by appealing to advanced technologies, it is less clear how, even after apparently cost-effective technologies are developed, they diffuse throughout the economy. Because such technologies are typically embodied in new investment capital, will firms routinely and promptly make these new investments? Can public policies accelerate this process? If so, how? This research examines the effectiveness of one particular policy, namely, the provision of free technical information by electric utilities about environmentally friendly technologies. “Free” means that the recipient incurs no direct costs to obtain the information. RICHARD D. MORGENSTERN is Visiting Scholar, Resources for the Future, and Associate Assistant Administrator for Policy, U.S. Environmental Protection Agency (on leave), Washington, DC. SAADEH AL-JURF is a student at the University of Virginia School of Law, Charlottesville, Virginia. Address correspondence to Dr. R. D. Morgenstern, Resources for the Future, 1616 P Street, N.W., Washington, DC.

Technological Forecasting and Social Change 61, 13–24 (1999)  1999 Elsevier Science Inc. All rights reserved. 655 Avenue of the Americas, New York, NY 10010

0040-1625/99/$–see front matter PII S0040-1625(98)00059-6

14

R. D. MORGENSTERN AND S. AL-JURF

Over the past decade electric utilities, along with gas utilities and federal and state governments, have been providing information to potential buyers to stimulate private investment in certain energy conservation technologies. Utility-sponsored demand-side management (DSM) programs, which often combine free technical information and cash subsidies, have been available to both commercial and residential customers in many states since the mid-1980s.1 High-efficiency lighting is the most popular element of these programs. The “Green Lights” program, sponsored by the U.S. Environmental Protection Agency (EPA), has provided similar information, albeit not tailored to sitespecific conditions, to corporations and other institutions since 1991. Although many recipients of free technical information do retrofit advanced lighting technologies, little is known about whether they are motivated solely by this information or by other factors as well. The most transparent way to measure the relation between technology adoption and the provision of free technical information would be to focus on two groups of buildings, identical in all respects, including building user characteristics, only one of which (randomly) receives the free information. One could then examine whether there is a difference in the rates of technology adoption between the two groups of buildings. Because the groups would be otherwise identical (due to randomization), this would yield an unbiased estimate of the effect of information programs on technology adoption. Unfortunately, survey data on such randomized samples of buildings and buildings users are generally unavailable. Program evaluations typically compare program participants to some synthetic control group. Given the inherent difficulties of finding a suitable control group the results of such studies are frequently called into question. The present article is based on a random sample of buildings—in this case a Department of Energy survey of commercial buildings—that is not specifically directed at participants in energy conservation programs. In the absence of a set of randomized buildings users, we construct a model of technology adoption. This model incorporates other factors besides information programs which affect the investment decision. How many firms, for example, would adopt the technologies anyway, perhaps because they also receive subsidies, face high electricity prices, are intensive appliance users, or have a preference for high tech solutions, but decide to accept the free information because it is available? Failure to account for such factors may cause us to overestimate the influence of information programs on technology adoption. Additionally, to test the hypothesis that the acceptance of technical information is itself an endogenous determinant of technology retrofit, they are modeled as jointly determined dichotomous variables. Our findings lend support to the notion that the free dissemination of technical information can speed the diffusion of new energy-saving technologies.

Literature Review Research by economists on the subject of technology diffusion dates back more than four decades. The single most important conclusion of this work is that diffusion of new, economically superior technologies is a gradual rather than instantaneous process (for literature reviews, see [1, 2]). Diffusion is often portrayed as a classic sigmoid curve over time (i.e., the rate of adoption begins slowly, speeds up, and eventually slows down). One justification for the sigmoid curve is that due to a lack of knowledge or confidence in its performance, the probability that a non-user will adopt a new technology 1 DSM programs provide site-specific information to customers on performance, costs, and availability of energy-saving equipment; some also provide financial assistance to those undertaking retrofits.

FREE INFORMATION AND TECHNOLOGY DIFFUSION

15

increases with its growing popularity. From this intuition, it makes sense that the rate of adoption will be slow in the beginning (before it becomes popular) and in the end (when there are few non-users). The pioneering work of Griliches [3] established that the diffusion of a new technology can be understood in an economic framework by allowing it to be partly determined by the (expected) economic returns to adopting the technology. Mansfield [4] demonstrated that the rate of diffusion can depend on the size of adopting forms, the perceived risk of the new technology, and the size of the required investment. Such factors do not, however, distinguish the rate of adoption by different firms. Individual firms are considered to be homogeneous with respect to their rate of adoption. An alternative approach would focus on inherent differences or heterogeneity among firms [5–7]. In such a model the gradual diffusion process depends on differences among potential users that affects the value of the innovation to them (e.g., the costs of learning about a new technology, adapting existing processes, and acquiring and operating the equipment). One can think of there being a threshold above which it pays to adopt the new technology and below which it does not. Over time, the cost of the innovation may fall and/or the quality may improve, thereby lowering the threshold. Use of this approach depends on the ability to identify differences that affect a firm’s valuation of the innovation (for environmental applications using both homogeneous and heterogeneous models, see [8–11]). The empirical literature on the effectiveness of utility-sponsored DSM programs draws, in part, on the studies of technology diffusion. Most studies focus on programs involving direct subsidies and compare them to some synthetic control group (see, for example, 12–22]). Those that also include information programs generally fail to distinguish between utility customers accepting information only and those accepting explicit subsidies. In contrast, the present study focuses on electric utility-sponsored information-only programs that are designed to lower the cost of learning about new energy-saving technologies. With a random sample of buildings we apply a simple economic model that focuses on the heterogeneity of users.

Model and Data Economic models of firm decision making are typically based on rational choice. Firms are assumed to make decisions which minimize the present discounted value of expected net costs. In the case of retrofit investment in high-efficiency lighting, the primary benefit is the reduction in electricity costs. The firm chooses whether or not to adopt the new technology by calculating the difference between the present discounted value of the operating savings versus the installed costs of each proposed investment. Technology adoption occurs when expected energy savings associated with adoption are calculated to be greater than or equal to the expected installation and equipment cost of the new lighting technologies. Although we do not have building-level information on these energy savings or equipment/installation costs, we do know whether or not the new technologies were adopted during a specified period. Thus we use a stylized reduced form probit model of technology adoption: Pr(Yi 5 1) 5 F[S(XS,i) 2 C(Xc,i)]

(1)

where F is a cumulative standard normal density function, S(Xs,i) is a proxy for operating savings from reduced electricity costs, and C(Xc,i) is a proxy for the installed cost of the new technology. Xc,i and Xs,i are data vectors determining operating savings and installed costs, respectively.

16

R. D. MORGENSTERN AND S. AL-JURF

This specification assumes that the probability of technology adoption rises with the modeled net savings (i.e., the difference between S and C). Actual adoption also depends on an unobserved disturbance. Specifying S(Xs,i) and C(Xc,i) as linear functions we estimate the coefficients (as and ac) using maximum likelihood.2 To test the hypothesis that the acceptance of technical information is an endogenous (as opposed to exogeneous) determinant of technology retrofit, they can both be modeled as jointly determined dichotomous variables (see Appendix 1). Our data set contains several variables, Xc,i, relevant to electricity savings: the average price paid by the building for electricity, the number of hours per week the building operates, building size, and whether or not time of day pricing is applied. As proxies for the cost of adopting the new technologies, Xs,i, we have information on the age of the building and the prevailing wage rate in the area. Although a utility-sponsored information program would not be expected to reduce the direct cost of a new technology, it is posited to reduce certain “trial and error” costs associated with finding and assessing the new technologies. Both the provision of free information and the presence of time of day pricing are postulated to directly increase the probability of technology adoption. They are both defined as simple categorical variables (1 5 yes; 0 5 otherwise). Our principal data set is the Department of Energy’s 1992 Commercial Buildings Energy Consumption Survey (CBECS) [23], a probability sample, representing the 4.8 million commercial buildings in the United States as of spring 1992. Information in this survey is drawn from building owners/managers/tenants as well as from local utilities.3 In order to focus on retrofits and avoid the possibility that the high-efficiency lighting was installed at the time of new construction, the sample is limited to buildings constructed prior to 1986. And because of the varied patterns of energy demand in different types of commercial buildings (e.g., restaurants versus warehouses), the sample is further restricted to office buildings, the largest and probably most homogenous category of commercial buildings. The three most common lighting upgrades, compact fluorescents, occupancy sensors, and specular reflectors, are selected as indicators of high-efficiency lighting for this study. Three dichotomous variables are defined according to whether lesser or greater amounts of advanced lighting technologies are present in the building. TECH1 indicates whether a building contains one or more of the three technologies; TECH2 and TECH3 indicate whether a building contains at least two, or all three technologies, respectively. As shown in Table 1, 64.3% of the commercial office buildings had one or more of these technologies in place, 27.5% had two or more, and 6.5% had all three. CBECS asks about DSM participation over the three previous years and categorizes positive responses into “site-specific information,” and “financial assistance.”4 These two types of DSM assistance involve very different activities. Because receipt of financial 2

All continuous variables are modeled in logs. Cross-checking information from two separate sources is a desirable but unusual practice. In CBECS the building respondents (owners/managers/tenants) reported only about one-fourth as much participation in DSM programs as did utility respondents. The authors of the CBECS indicate that one of the reasons this discrepancy may arise is that the utilities had more detailed records of DSM participation [23]. Examination of the individual responses indicated that the discrepancies went in both directions: that is, utilities reported buildings had participated in DSM programs when the building respondent indicated they had not participated, and vice-versa. Accordingly, on the assumption that errors of omission were more likely than errors of commission, participation in DSM programs is defined on the basis of a positive indication from either the utility or the building respondent or both. 4 There was also a category for “general information.” However, that is most likely a bill insert and not part of any systematic transfer of information. 3

FREE INFORMATION AND TECHNOLOGY DIFFUSION

17

TABLE 1 Variable Means and Standard Deviations (SD) Variable Adopted lighting technology (TECH1) Adopted lighting technology (TECH2) Adopted lighting technology (TECH3) Wages (dollars/hour) Electricity price (P, cents/kwh) Hours Year of construction Size of building (in thousands of square feet) Time of day pricing (TOD) Infromation provided (INFO) Owner occupied Heating-cooling DSM Annual electricity bill (cents/square feet) Sample size

Mean

SD

0.643 0.275 0.065 13.814 7.931 90.200 1967.0 73.428 0.124 0.209 0.733 0.272 1.572 990

(0.479) (0.447) (0.247) (2.709) (3.270) (44.195) (21.441) (52.599) (0.329) (0.407) (0.443) (0.445) (1.104)

Source: [23]. TECH1 One if at least one of three lighting technologies in use. Zero otherwise. TECH2 One if at least two of three lighting technologies in use. Zero otherwise. TECH3 One if all three lighting technologies in use. Zero otherwise.

assistance is predicated on actually installing specified equipment, one cannot use this information to examine the effect that accepting financial assistance has on the decision to retrofit lighting systems. In contrast, receipt of site-specific information does not bind the recipient to any action. Although the building owner/tenant has to request or at least accept the offer of site-specific information from the utility, there is no requirement to actually install the equipment. Accordingly, the sample is defined to include only those buildings which received site-specific information from the utility but did not receive financial assistance.5 All other variable definitions are relatively straightforward.6 Table 2 shows the distribution of technology adoption outcomes depending on whether they received information from the local utility. In this and subsequent tables all data are weighted by building size. Thus the percentages shown here represent fractions of the total square footage in the sample, not the proportion of building structures. Of those receiving information from the utility, 20.7% have none of the three technologies, 36.5% have one, 29.6% have two, and 13.2% have all three technologies. Among those not receiving utility-supplied information, a higher percentage have none or one of the technologies and a lower percentage have two or more of the technologies. Although these tabulations do not account for any of the economic and use factors likely to effect the probability of technology adoption, they do suggest two tentative conclusions: (1) information programs appear to increase the probability of 5 To test for the possibility that eliminating those buildings which received financial assistance (n 5 340) introduces selectivity bias into the sample, the models presented in Table 3 and Appendix 1 were re-estimated on the full sample by coding “financial assistance” as if it were the same as “information.” The basic results were unchanged; although, as expected, the coefficient on the newly defined information variable was slightly larger. 6 Average electricity prices are defined as the annual electricity bill divided by the annual kilowatt-hours, as reported by the utility. The presence of time of day pricing (as reported by the utility) is defined as a dichotomous variable. As shown in Table 1, it is used in 12.4% of the buildings. Average hourly wages in the region are constructed by dividing annual earnings of electrical equipment installers by their hours worked per year. Wages were averaged for metropolitan and non-metropolitan statistical areas in each census division (Census Bureau’s Data Extraction System).

18

R. D. MORGENSTERN AND S. AL-JURF TABLE 2 Technology Adoption According to Information Provided: Gross Data

INFO provided INFO not provided INFO relative to no INFO

Percent not adopting any of three technologies

Percent adopting one of three technologies

Percent adopting two of three technologies

Percent adopting all three technologies

20.7 39.7 219.0

36.5 36.9 20.4

29.6 18.6 11.0

13.2 4.8 8.4

Source: [23].

high-efficiency lighting and (2) such programs seem to shift non-adopters to the two or more technologies category, leaving the percent adopting only one technology virtually unchanged. The model estimates presented in the next section assess whether these tentative conclusions withstand further scrutiny.

Results Using TECH1, TECH2, and TECH3 as (alternative) dependent variables, the maximum likelihood results for the exogenous model are presented in Table 3. Most of the independent variables are significant in at least one of the equations and the coefficients generally conform to expectations. The coefficients on average electricity prices are positive and significant in all technology adoption equations. The coefficients for time of day pricing are also positive in all equations and significant in two of the three equations. As expected, buildings that operate longer hours per week and those which were constructed more recently have a greater likelihood of adopting highefficiency lighting technologies, although the coefficients on these variables are not TABLE 3 Probit Results: Technology Adoption with Exogenous Program Participation Exogenous program participation Independent variables Intercept Wages (dollars/hour) Electricity prices Hours Year of construction Building size (in thousands of square feet) Time of day pricing Information provided Log likelihood

TECH1 Technology adoption 251.871* (30.436) 0.659*** (0.222) 0.687*** (0.161) 0.160 (0.101) 6.037 (4.006) 0.191*** (0.038) 0.657*** (0.155) 0.594*** (0.116) 2971.04

TECH2 Technology adoption

TECH3 Technology adoption

228.455 (35.585) 20.299 (0.225) 0.465*** (0.136) 0.018 (0.119) 3.228 (4.683) 0.227*** (0.053) 0.217 (0.137) 0.530*** (0.108) 2935.10

n 5 990 observations. Standard errors in parentheses. *** Significantly different from zero at the 0.01 level; ** at the .05 level; * at the .10 level.

87.061 (57.704) 21.459*** (0.466) 1.510*** (0.256) 0.224 (0.242) 212.229 (7.554) 0.255* (0.145) 0.924*** (0.246) 0.820*** (0.210) 2585.59

FREE INFORMATION AND TECHNOLOGY DIFFUSION

19

TABLE 4 Technology Adoption According to Information Provided: Modeled Results Percent not adopting any of three technologies INFO provided INFO not provided INFO relative to no INFO

Percent adopting one of three technologies

Percent adopting two of three technologies

Percent adopting all three technologies

14.8 32.6

39.0 39.4

38.3 26.3

7.8 1.6

217.8

20.4

12.0

6.2

Source: Calculated from Table 3.

significantly different from zero. Larger buildings are more likely to adopt high-efficiency lighting, at least for TECH1 and TECH2. Wages are a more complex story.7 Of key interest in this study is the information variable which is positive and highly significant in all three technology adoption equations. Thus, even after adjusting for a complex set of factors that a rational model suggests would influence adoption, the provision of technical information by local utilities remains an important determinant of technology choice.8 One way to display these results is to consider how the probability of adopting advanced lighting technologies varies depending on whether the firms received information from the local utility. To develop such estimates, we apply the basic model to various subsamples of buildings in order to compare the effect of information programs on those firms which have adopted some advanced lighting technologies versus the effect of these programs on first-time buyers.9 As shown in row one of Table 4, 14.8% of those who received information from the utility have none of the three technologies, 39.0% have one, 38.3% have two, and 7.8% have all three. Among those who did not receive utility-supplied information, a higher percentage have none or one of the technologies and a lower percentage have two or more of the technologies. These estimates, based on the economic and use factors included in the model confirm the basic findings shown in the simple sample distributions calculated in Table 2, (i.e., information programs appear to increase the probability of technology adoption, and they appear to be somewhat more important in encouraging building owners to adopt 7

If labor were only a factor in the installation of the high-efficiency lighting, one would expect the coefficient to be negative. However, since labor is also required to change lightbulbs and high-efficiency lighting tends to require less frequent bulb replacement, there is also a labor-saving element associated with highefficiency lighting. For TECH3 the negative and highly significant coefficient indicates the first effect dominates. For TECH1 the reverse is true. This suggests that marginal labor costs for installation are a more important factor than marginal labor savings from reduced maintenance in the presence of a larger number of highefficiency lighting appliances. 8 The results from the endogenous model are not statistically different from the single equation approach (see Appendix 1). 9 The probability of adopting no new technologies is derived from calculating the value of TECH1 using average values for all variables except INFO, which is set equal to 0. The probability of adopting one new technology is derived from the same equation, where INFO 5 1, multiplied by one minus the probability of adopting two or more technologies, given that you have already adopted one appliance. (The latter probability is derived by re-estimating the basic model on the subset of buildings which have adopted at least one technology. For that subsample, the calculated effect of the information programs is to raise the probability of adoption from 41.5% to 54.2%.) The probabilities of adopting two or all three technologies are derived from similar calculations. For the subsample of buildings with two or three technologies in place, the effect of the information programs is to raise the probability of adoption of all three technologies from 5.9% to 16.8%.

20

R. D. MORGENSTERN AND S. AL-JURF

multiple technologies than for encouraging non-adopters to purchase a single technology).

Conclusions This article examined the use of information subsidies as a means of accelerating the diffusion of new technologies. Estimates are based on a random sample of commercial buildings, rather than the more common comparison of program participants to a synthetic pairing with another populations. In the context of an economic model which emphasizes heterogeneity among building users, the provision of technical information by electric utilities is found to be a significant determinant of the adoption of highefficiency lighting technology in commercial office buildings. Additionally, information programs appear to be more effective for encouraging building owners who have already invested in some advanced lighting technologies compared to first-time purchasers. These findings are based on a relatively simple specification where information programs are modeled as directly increasing the probability of technology adoption. Other specifications may also be of interest (e.g., whether the information programs are particularly effective in the presence of high energy prices or long hours of building use). It is instructive to make cross-comparisons among policy tools—in this case electricity prices and information programs—to determine their relative effectiveness in influencing technology choice.10 Of course, the preferred way to compare these policy tools is in the context of a more complete economic model (e.g., one which accounts for both the benefits and costs of the different tools). Such an accounting would recognize that price increases would have the additional effect of reducing electricity use in new and existing installations as well as in nonlighting uses of electricity. Notwithstanding those differences, our estimates suggest that electricity prices would have to rise by more than 25% in order to bring about an increase in technology adoption equivalent to that induced by the utility-sponsored information programs.11 EPA’s Green Lights program, which has some similar features to utility-sponsored DSM programs, measures its success in terms of the amount of its participant’s building space committed to install new technology.12 It does not estimate how much retrofit is likely to have occurred in the absence of the program [24, 25]. It is certainly plausible to interpret our findings as consistent with general claims of effectiveness made by providers of technical information, including the Green Lights program. However, our results clearly suggest the importance of accounting for heterogeneity among users when developing quantitative estimates. Absent detailed cost information it is not possible to address the cost-effectiveness or net economic benefits of information-based programs. However, it is clear the provision of technical information has a positive effect on the decision to adopt high-efficiency 10 Changes in average prices can be brought about by some form of energy taxes (which would increase end-use prices), or, alternatively, by policies that increase competition in the electricity industry, such as restructuring (which would reduce end-use prices). 11 This estimate is derived from the TECH1 equation by calculating the increase in average electricity prices necessary to bring about an increase in technology adoption equivalent to the use of a utility-sponsored information program. 12 Although it does not directly provide for site visits by lighting experts, Green Lights is similar to utility DSM programs in many respects. Typically Green Lights works with managers of large amounts of commercial floor space and provides evaluation tools for the firm to undertake a set of detailed assessments. These assessments, in turn, provide the basis for the firm to make its own site-specific decisions to determine which lighting upgrades are most appropriate in particular applications.

FREE INFORMATION AND TECHNOLOGY DIFFUSION

21

lighting technologies, particularly so when firms had prior experience with related technologies. The authors acknowledge helpful comments from Howard Gruenspecht, Winston Harrington, Richard Newell, William Pizer, and David Ribar.

References 1. Jaffe, A. B., and Stavins, R. N.: The Energy Paradox and the Diffusion of Conservation Technology, Resource and Energy Economics 16, 91–122 (1994). 2. Jung, C., Krutilla, K., and Boyd, R.: Incentives for Advanced Pollution Abatement Technology at the Industry Level: An Evaluation of Policy Alternatives, Journal of Environmental Economics and Management 30(1), 95–111 (1996). 3. Griliches, Z.: Hybrid Corn: An Exploration in the Economics of Technological Change, Econometrica 15, 501–522 (1957). 4. Mansfield, E.: Industrial Research and Technological Innovation. W.W. Norton, New York, 1968. 5. David, P. A.: The Mechanization of Reaping in the Ante-Bellum Midwest, in Industrialization in Two Systems. H. Rovosky, ed., Harvard University Press, Cambridge, MA, 1969, pp. 3–39. 6. David, P. A.: Technology Diffusion, Public Policy, and Industrial Competitiveness, in The Positive Sum Strategy: Harnessing Technology for Economic Growth. R. Landau and N. Rosenberg, eds., Washington, D.C., National Academic Press, 1986, pp. 373–391. 7. Stoneman, P.: The Economic Analysis of Technical Change. Oxford University Press, Oxford, UK, 1983. 8. Downing, P. B., and White, L. J.: Innovation in Pollution Control, Journal of Environmental Economics and Management 13, 18–27 (1986). 9. Thirtle, C. G., and Ruttan, V. W.: The Role of Demand and Supply in the Generation and Diffusion of Technical Change. Harwood Academic Publications, New York, 1987. 10. Malueg, D. A.: Emission Credit Trading and the Incentive to Adopt New Pollution Abatement Technology, Journal of Environmental Economics and Management 16, 52–57 (1989). 11. Oster, S.: The Diffusion of Innovation Among Steel Firms: The Basic Oxygen Furnace, The Bell Journal of Economics 13, 45–56 (1982). 12. American Council for an Energy Efficient Economy (ACEEE): Proceedings of the ACEEE Summer Study on Energy Efficiency in Buildings. Washington, D.C., 1992, 1994. 13. Braithwait, S., and Caves, D.: Three Biases in Cost-Efficiency Tests of Utility Conservation Program Evaluation, Energy Journal 15(1), 95–120 (1994). 14. Eto, J., Vine, E., Shown, L., Sonnenblick, R., and Payne, C.: The Total Cost and Measured Performance of Utility-Sponsored Energy Efficiency Programs, Energy Journal 17(1), 31–51 (1996). 15. Hirst, E.: Price and Cost Impacts of Utility DSM Programs, Energy Journal 13(4), 72–90 (1992). 16. Hirst, E.: Effects of Utility Demand-Side Management Programs on Uncertainty, Resource and Energy Economics 16(1), 24–45 (1994). 17. Joskow, P. L., and Marron, D. B.: What Does a Negawatt Really Cost? Evidence From Utility Conservation Programs, Energy Journal 13(4), 41–74 (1992). 18. Malm, E.: An Actions-Based Estimate of the Free Rider Fraction in Electricity Utilitly DSM Programs, Energy Journal 17(3), 41–48 (1996). 19. Ozog, M. T., and Waldman, D. M.: Weighting Nonrandom Samples in Voluntary Energy Conservation Program Evaluation, Energy Journal 15(1), 129–141 (1994). 20. Train, K. E.: Incentives for Energy Conservation in the Commercial and Industrial Sectors, The Energy Journal 9(3), 113–128 (1998). 21. Train, K. E., and Atherton, T.: Rebates, Loans and Consumers’ Choice of Appliance Efficiency Level: Combining Stated- and Revealed-Preference Data, Energy Journal 16(1), 55–69 (1995). 22. Wirl, F.: On the Unprofitability of Utility Demand-Side Management, Energy Economics 16(1), 46–53 (1994). 23. U.S. Department of Energy (USDOE): Commercial Buildings Energy Consumption and Expenditures, 1992. Washington, D.C., 1995. 24. U.S. Environmental Protection Agency: EPA Green Lights Program Snapshot. April 30, Washington, D.C., 1995. 25. U.S. General Accounting Office (USGAO): Global Warming: Information on the Results of Four of EPA’s Voluntary Climate Change Programs Washington, D.C., 1997.

22

R. D. MORGENSTERN AND S. AL-JURF

26. Maddalla, G. S.: Limited-Dependent and Qualitative Variables in Econometrics. New York, N.Y., Cambridge University Press, 1983. 27. Ribar, D. C.: Teenage Fertility and High School Completion, Review of Economics and Statistics 76(3), 413–424 (1994). Received 10 July 1998; revised and accepted 12 December 1998

Appendix 1 TEST OF POSSIBILE ENDOGENEITY

A simple empirical model can be used to analyze the possibility that acceptance of technical information is an endogeneous determinant of technology retrofit [26, 27].13 Let INF*i denote the (unobserved) benefit to the ith firm of receiving technical information from the utility. Assume that INF*i is a linear function of economic and institutional determinants. Specifically, let INF*i 5 Zid 1 hi

(A1)

where Zi. is a vector of observed variables which includes both Xi and other variables, and hi is a random error term.14 Although INF*i is not observable, it is related to INFi (which is observable) as follows:

5

1 (if firm receives/accepts technical information) INFi 5 0 (if firm does not receive/accept technical information)

if INF*I > 0 otherwise. (A2)

The error terms ei and hi are assumed to be distributed bivariate standard normal with means and standard deviations as follows:15

3he 4 z N13004,31 r142. i

(A3)

i

Equation A1 with the distribution assumption (A3) specify the probability of technology adoption as a probit model with information as an endogenous determinant. The purpose of the distribution assumption in equation A3 is to allow for the measurement of correlation (r). (The case of r 5 0 is the model used in the text, i.e., the acceptance of information is exogenous to the decision to retrofit.) The effect of information on technology adoption is only identified subject to exclusion or covariance restrictions, in this case on the vector Xi. The variables which are excluded from Xi should be theoretically and statistically related to obtaining information from the utility but unrelated to technology adoption. In fact, it is both logically and empirically difficult to identify factors that influence one but not the other. Notwithstanding, this research considers three such variables—a dichotomous variable indicating whether or not the building has participated in utility-sponsored DSM programs on heating and 13 The theoretical foundations of this model are found in Maddala [26]. An interesting empirical (albeit non-energy) example can be found in Ribar [27]. Train [20] constructs an econometric model to deal with the endogenous choice issue by assuming a nested logit distribution. 14 The determination of TECH* and INFO* are not treated symmetrically (INFO* does not include TECH as an explanatory variable). Unfortunately, the likelihood function for the symmetric specification does not, in general, integrate to one [26]. 15 Maximum likelihood estimation of this specification is straightforward. However, the coefficients and error variances in equations 9 and 11 are only identified up to their proportions, b/se, A/se, and d/sh. We apply the standard normalization se 5 sh 5 1.

0.112 (0.156) 2970.78

251.630 (30.332) 0.637*** (0.225) 0.668*** (0.163) 0.149 (0.104) 6.019 (3.991) 0.197*** (0.039) 0.657*** (0.157) 0.440* (0.230)

Technology adoption

20.420*** (0.125) 1.433*** (0.111) 20.159*** (0.053)

27.186 (47.722) 20.414 (0.290) 20.419** (0.206) 20.232* (0.137) 23.453 (6.290) 0.087* (0.052) 20.198 (0.165)

DSM program participation

20.004 (0.159) 2935.10

228.503 (35.595) 20.298 (0.229) 0.465*** (0.138) 0.018 (0.119) 3.234 (4.685) 0.226*** (0.053) 0.217 (0.138) 0.536** (0.237)

Technology adoption

TECH2

20.410*** (0.125) 1.437*** (0.110) 20.147*** (0.055)

28.152 (47.653) 20.417 (0.292) 20.422** (0.210) 20.239* (0.137) 23.578 (6.281) 0.087* (0.052) 20.204 (0.169)

DSM program participation

Endogenous Program Participation

n 5 990 observations. Standard errors in parentheses. *** Significantly different from zero at the 0.01 level; ** at the 0.05 level; * at the 0.10 level.

Log likelihood

•

Electricity expend/sq ft.

Heating-cooling DSM

Owner occupied

Information provided

Building size (in thousands of square feet) Time of day pricing

Year of construction

Hours

Electricity prices

Wages (dollars/hour)

Intercept

Independent variables

TECH1

TABLE A1 Probit Results: Technology Adoption with Endogenous Program Participation

0.227 (0.237) 2584.69

21.491 (0.467) 1.447*** (0.287) 0.213 (0.238) 212.120 (7.827) 0.264* (0.145) 0.913 (0.253) 0.493 (0.483)

86.468

Technology adoption

TECH3

20.369*** (0.134) 1.446*** (0.115) 20.143*** (0.052)

25.630 (48.114) 20.419 (0.298) 20.418** (0.211) 20.238* (0.138) 23.247 (6.347) 0.084 (0.053) 20.218 (0.166)

DSM program participation

FREE INFORMATION AND TECHNOLOGY DIFFUSION 23

24

R. D. MORGENSTERN AND S. AL-JURF

cooling; a dichotomous variable indicating whether or not the building is owner occupied; and a variable for the electricity bill (per square foot) of the entire building (not just the lighting system).16 The endogenous results, which are similar to the exogenous model in many respects, are shown in the Table A1. Despite the theoretical appeal of the endogenous specification, the empirical grounds for rejecting the exogenous model are not strong.17,18

16 Participation in another utility-sponsored DSM program is taken as an indicator that the building owner is familiar with DSM programs. Assuming the firms’ experience with the other program was positive, one would expect a positive sign on this variable (for a positive assessment of DSM programs see Hirst [15, 16]; for a contrary view see Wirl [22]). Owner occupancy is a more complicated factor. Inasmuch as owneroccupants are more likely to reap the full benefits from lighting retrofits, one would expect a positive sign on this variable. If, however, many owner occupants had already retrofit their lighting systems, then they might be expected to participate less often in utility DSM programs than non-owner occupants. Electricity expenditures per square foot is also a complicated factor. High electricity expenditures per square foot, even if it is unrelated to lighting, could represent a wake-up call to acquire information about lighting retrofits. On the other hand, low expenditures per square foot in a building which did not already use efficient lighting could indicate a strong preference for energy efficiency and thus could account for an interest in technical information on lighting retrofits. None of these three variables is assumed to directly affect the probability of lighting retrofit. (Some might challenge that assumption, for example, owner occupancy may also affect the ability to capture savings from retrofitting lighting systems.) Several other types of DSM programs were also examined for inclusion as independent variables in this equation. However, none altered the basic results reported here and none were statistically significant. 17 The excluded variables in the endogenous model are highly significant and of the expected signs in all three equations. However, the values of the correlation coefficient which measure endogeneity (r) are all quite small (0.112, 20.004, and 0.227, respectively). In no case is the t-value of these correlation coefficients greater than 0.9. Similarly, statistical comparison of the log likelihood ratios indicates there are no significant differences between the exogenous and the endogenous specifications. These findings suggest, contrary to expectations, extremely weak evidence to support use of the endogenous model. Note also that the only other variable which is statistically significant in the estimation of DSM program participation is electricity prices and it has a negative coefficient. Sensitivity analysis showed that estimating this equation separately without the three excluded variables also produced a significant and negative coefficient. The most obvious explanation for this finding is that many buildings in areas with high electricity prices had previously installed high-efficiency lighting and were thus less likely to get involved in DSM lighting programs. 18 One possible explanation for the lack of statistical significance between the exogenous and endogenous models is that some firms have a preference for high technology solutions and that preference is an important determinant of both the decision to adopt high-efficiency lighting and to participate in the lighting DSM program. To check for the possibility of such a misspecification, a total of 22 different measures of “high technology” equipment, relating to energy management equipment, heating and cooling, shell measures, and others were examined as possible variables. Few of these variables were significant and none of them statistically altered the size or significance of the coefficient on the information variable.