The data Directory UMM :Data Elmu:jurnal:E:Energy Economics:Vol22.Issue6.2000:

K. Vaage r Energy Economics 22 2000 649]666 657

3. The data

Our data are based on an extensive survey on household energy consumption, organised by the Norwegian Central Bureau of Statistics in late spring 1980. A revision of the collected data together with an econometric analysis is documented Ž . in Hem 1983 . The survey covers 2289 households equally distributed throughout the country. It contains data on the following groups of variables: 3.1. Energy consumption and expenditures for each energy source Electricity is the dominating energy source, representing approximately 65 of Ž . the households’ energy consumption. Other fuels are oil 25 and solid fuels Ž . Ž . mostly burning wood 10 . The data do not allow separation of consumption into heating vs. other purposes, but of a total consumption of 23.4 TWH in 1979, 13 70]80 represented heating. In addition to the respondents’ own answers, our data base contains information from the respective electricity utilities, giving the exact tariff type 14 electricity consumption and electricity expenditure for each interviewee. 3.2. Heating technology The survey gives detailed information about existing installations. A distinct characteristic of the Norwegian household consumption is the high degree of combined heating technology. Of the households in the survey, 80 combine two or more fuels, while 11 are based solely on electricity, and only 9 on oil or wood. 15 The interviewees are also asked to give some subjective characteristics on quality, plans regarding new installations, etc. Unfortunately, there are no objective measures of quality, nor any data on capital costs, data that are highly recom- mended when modelling the technology choice. 3.3. Building characteristics Detailed information is available about age, type, floor area, number of rooms, ownership, etc. Obviously, these are important variables to include in the conditio- 13 Ž . The fossil fuels are converted to KWH by converting factors reported in Hem 1983, p. 25 . 14 Ž . Some 60 of the households belong to the group with increasing block price tariff ‘H3’ . It is well Ž . known, see, e.g. Taylor 1975 , that block pricing tariff might cause problems when calculating average Ž prices from expenditure data. However, we have reason to believe that this is a minor problem see . exogeneity tests in Vaage, 1995 , and continue as if electricity were available for each household at a fixed price. 15 We omitted households living in flats without a chimney from the sample, since they are left with no other alternatives but electricity for heating purposes. K. Vaage r Energy Economics 22 2000 649]666 658 nal demand model, but their potential influence on appliance choice will be tested for as well. 3.4. Inter ¨ iewees and household members Data are available on geographic location, age, sex, occupation and education of the interviewed persons. The survey also reports on the number of household Ž . members, the number of children, gross household income before taxes , and the number of persons with income. This provides us with a database quite suitable for modelling of the conditional energy demand, but with severe limitations with respect to a satisfying model of the appliance choice. The first problem is lack of data on appliance attributes, b . The i k Ž . Ž . quality index in Eq. 7 , therefore, has to be reduced to c s exp a q « . Hence, i i i Ž . quality cleanness, comfort, controllability, flexibility, reliability, etc. is represented by appliance dummies, a , plus the random component, « . i i The second problem is the insufficient price data for the different choice alternatives. The consumer choice of appliance purchase, replacement, and retire- ment is based on operating and capital costs of the alternative technologies. However, in the present survey only contemporaneous data on operating costs are available. 16 Nevertheless, our view is that our cross-sectional data contains impor- tant information: prices on all the different energy sources show large regional variation in our sample. As for oil prices, this is probably explained by the transportation costs. This also applies to wood; in addition, in rural areas it is common to have access to cheap burning wood through relatives and friends, a phenomenon that probably explains the huge price variation for this fuel. The Norwegian electricity pricing policy, which in 1980 was mainly based on each electricity utility’s historical costs, ensures a massive variation in electricity price as well. Hence, our data confirm large regional variation in relative operating costs for the different heating alternatives. The question is whether it makes sense to explain the dynamic phenomenon of choosing heating technology by contempora- neous relative prices. After all, we must expect that the majority of the households in our sample have based the choice of appliance on relative prices in a more or less distant past. Our point is that regional differences in relative prices are fairly stable over time, since they are explained by topography, historical costs, etc. Conditional on no migration, then, contemporaneous energy prices might actually be a good proxy for operating cost from the time in the past when the decision Ž . about appliance choice was taken. However, of course, other non-regional price movements, for example the oil price changes in the 1970s, weaken the quality of the current prices as a proxy for historical operating costs. 16 Ž . Nesbakken and Strøm 1993 manage to complement the 1990 Energy Sur ¨ ey by collecting capital cost data. Armed with this improved database they present a discretercontinuous model in the spirit of Ž . Dubin and McFadden 1984 . Unfortunately, such extra data have not been available in the present study. K. Vaage r Energy Economics 22 2000 649]666 659 In addition, to justify the use of contemporaneous operating costs to explain the appliance choice we have to assume static expectations. As noted by Dubin and Ž . McFadden 1984 , whose sample is also a cross-section, this assumption is at best only approximately true, and ideally should be tested against a more complete dynamic model, using panel data on consumer behaviour.

4. Results and discussion