Final Report: Households, Consumption, and Energy Use: The Role of Demographic Change in Future U.S. Greenhouse Gas Emissions

EPA Grant Number: R829801
Title: Households, Consumption, and Energy Use: The Role of Demographic Change in Future U.S. Greenhouse Gas Emissions
Investigators: ONeill, Brian , Dalton, Michael , Leiwen, Jiang , Pitkin, John , Prskawetz, Alexia
Institution: Brown University , California State University - Monterey Bay , Max Planck Institute for Demographic Research
EPA Project Officer: Michaud, Jayne
Project Period: September 1, 2002 through September 30, 2005 (Extended to September 30, 2006)
Project Amount: $279,015
RFA: Futures: Research in Socio-Economics (2001) RFA Text
Research Category: Economics and Decision Sciences

Objective:

The primary objective of the proposed research was to determine whether expected changes in the composition of the U.S. population by household type over the next 25-100 years will have a substantial influence on total energy demand and carbon dioxide emissions. Specific aims for achieving this overall goal were to: (1) develop a set of long-term household projections that characterize plausible ranges of the future distribution of households by size, age, composition, and other demographic characteristics, including nativity; (2) quantify how consumption patterns vary across households of different types; (3) introduce disaggregated household types into an existing energy-economic growth model of the United States to test the effect of accounting for demographic heterogeneity in energy and emissions projections, including the potential effects of alternative immigration scenarios.

Summary/Accomplishments (Outputs/Outcomes):

Our findings are summarized for each of the specific aims described above.

Develop a Set of Long-Term Household Projections That Characterize Plausible Ranges of the Future Distribution of Households

During the project we produced two sets of household projections for the United States—a preliminary set in Year 1 to use as input for initial work on other specific aims and a final set in Year 3—to explore the scope for future changes in household size and age structure. To produce these projections, we used the dynamic household projection model ProFamy. We first assessed the sensitivity of future living arrangements to various demographic events. Our sensitivity analysis demonstrates that the most important determinants of household size composition of the population are fertility and union formation and dissolution rates. While the effect of other factors is small, we found some non-obvious results. For example, delayed childbearing interacts with the age structure of the population to produce first an increase in average household size, and later a decrease. In addition, we find that increased life expectancy acts in the direction of smaller households, which was not obvious a priori given the expectation of a decline in single person elderly households when mortality is reduced.

We then assessed the outlook for future changes in households by developing three scenarios aimed at exploring a wide range of outcomes for household size and age. These scenarios are based on new scenarios for fertility, life expectancy, and migration that we derived by averaging across existing scenarios in the literature, an approach that has been suggested but that has not previously been used in long-term projections. We also produced the first long-term scenarios for household formation and dissolution rates that go beyond mechanical assumptions and are based on reasoning grounded in past trends, experience in other countries, and current theoretical perspectives. We anticipate that marriage rates could plausibly double or decline by half; cohabitation rates could double; and divorce rates could increase by 25% or decline by half.

Results indicate that average household size declines over the next few decades in all scenarios, due mainly to changes in population age structure and secondarily to momentum in household formation and dissolution processes. By the second half of the century, the range of plausible household size outcomes is 2.0 to 3.1, with this result being driven by tradeoffs in the proportion of the population living in one and two person households on the one hand, and households of size 4+ on the other. The proportion living in households headed by the elderly (65+) doubles in the youngest scenarios, and nearly quadruples in the oldest scenario, to 40% of the population. Conversely, the proportion living in households headed by the young (< 45) declines by nearly half, from 60% to 35%.

Taken together, these results give a first look at the range of plausible outcomes for living arrangements over the next 50-100 years. We were relatively conservative in defining our high and low scenarios by not choosing the most extreme scenarios in the literature for the components of population change and by grounding our scenarios for union formation and dissolution rates in past experience in the United States and other countries. In this way, we argue that the range of outcomes presented here is a minimum plausible range of uncertainty. It is possible that unprecedented rates of demographic events could be experienced in the future, in which case this range would be expanded even further.

Quantify How Consumption Patterns Vary Across Households of Different Types.

We analyzed Consumer Expenditure Survey (CES) data for patterns of consumption across households of different types. As a first step, we aimed to identify the most appropriate household types and aggregate consumption categories to use within our general equilibrium model of the U.S. economy. Our quantities of interest were budget shares for aggregate expenditure categories and shares of income, labor, assets, savings, and taxes by household type. Expenditure means cover all expenditures calculated by integrating data from the Interview and Diary Surveys (the two components of the CES). We analyzed results using a set of 12 household types (three age categories and four size categories). Based on differences in expenditures, income, savings, and assets across these household types, we concluded that there was indeed significant age heterogeneity, and that in particular our modeling approach should capture large differences between households headed by young (< 45) householders and those headed by middle-aged householders, in addition to well known differences between older households (> 65) and the rest of the population. Regarding size, we concluded that differences in economic behavior would justify differentiating size up to at least the size 4+ category in the young and middle-age groups, and up to at least the 3+ category in the old age group. In terms of consumption categories, we began with the 17 categories already included in our general equilibrium model and have concluded that, in the future, reducing this number to 6–9 categories would be sufficient.

We then estimated the specific parameters needed for the economic model. For example, mean expenditure shares for 17 categories of consumer goods were calculated and used as benchmark data for the Population-Environment-Technology (PET) model and converted to share parameters, μijt, that calibrate the model’s household demand system. Results show that older households spend a substantially larger share of income than younger households on utilities, services, and health care and a substantially smaller share on clothing, motor vehicles, and education.

The PET model requires additional data on age-specific economic activities. Government transfers, for example, include social security, workers compensation, unemployment benefits, and other kinds of public assistance, and these favor older households in per capita terms by a wide margin. Savings include retirement contributions, down payments on purchases of property, mortgage payments, capital improvements, and investments in own businesses or farms. Assets include the value of financial accounts and securities plus the equity share of property. Our results show clear variation over the lifecycle and drive key results in our emissions scenarios.

Introduce Disaggregated Household Types into an Existing Energy-Economic Growth Model of the United States

To incorporate age heterogeneity into the PET model, we developed a “multiple dynasty” structure that disaggregates the population not by age groups per se, but by dynasties, that is, groups that contain households of a given age today and that track those households, and the households of their children, as they age over time. This approach allows demographic heterogeneity while maintaining consistency with the overall model structure of decentralized, forward-looking households over an infinite planning horizon. The dynastic structure assumes intergenerational altruism in the form of parents caring about the welfare of their children. This form of altruism is implicit in the dynastic structure of neoclassical growth models; the multiple dynasty structure makes it an explicit means of linking cohorts into three heterogeneous, infinitely lived dynasties. Each dynasty contains households separated in age by the average length of a generation, which is about thirty years, so that on average, younger households are descendents of the older households.

We use the PET model to estimate effects of population aging by comparing emissions baselines from simulations with age-specific heterogeneity to baselines without aging and a representative household. To isolate demographic effects, the first set of simulations does not include technical change. Our results compare two types of heterogeneous households to a representative household. The first type has heterogeneity only in expenditure shares for different consumer goods that depend on the age of the household head. The second type has heterogeneity in expenditure shares and also in sources of household income, including capital and labor. The first type of heterogeneity affects only the composition of demand, but our results show that these effects are negligible. In contrast, age-specific heterogeneity in labor income reduces CO2 emissions by 11%, 18%, and 37% per year by 2100 in the high, medium, and low population scenarios, respectively. In our reference case, a labor scale effect accounts for about 85% of these reductions, and the other 15% is from capital dynamics and general equilibrium effects. However, sensitivity analysis indicates that simply scaling the labor supply of a single representative dynasty to account for population aging has ambiguous effects that either underestimate or overestimate emissions reductions from population aging, depending on values of household substitution parameters, about which we are uncertain.

Figure 1. Future U.S. CO2 Emissions Assuming No Technical Change for Three Population Scenarios, Modeled Using a Representative Household (Solid Lines) or Allowing for Demographic Heterogeneity by Accounting for Aging (Dashed Lines), from Dalton et al. (2006).

A second set of simulations compares emissions baselines with population aging to a representative household in the presence of technical change. Assumptions about technical change are based on the Special Report on Emissions Scenarios (SRES) A1 Scenario for Organisation for Economic Co-operation and Development (OECD) countries. For our reference values of household substitution elasticities, effects on emissions from aging and decreases in carbon intensity from technical change are additive in the long run. The most interesting result is that effects of aging on emissions are as large, or larger, than effects of technology in some cases. The main trade-off in this result is the amount of aging in the household projections on the one hand, and the nature of the technical change on the other.

Results support further consideration of demographic factors in emissions projections, and suggest that these factors may be critical to the development of new emissions scenarios, particularly those based on low population projections for the United States, because effects of aging are most important in this scenario.

Test the Effect of Alternative Immigration Scenarios on Future U.S. Carbon Emissions

Another potentially important source of heterogeneity in the U.S. population is immigration. Households that are headed by the foreign-born, or by second-generation immigrants, can differ from the rest of the population in terms of various demographic and economic characteristics. Immigration is a key determinant of future demographic outcomes in the United States, and therefore we investigated whether explicitly accounting for heterogeneity by nativity status might be worthwhile in projections of energy use and carbon emissions.

The three population scenarios used in our PET model analysis approximately indicate the effects of different levels of immigration on the size and age composition of the population and, in turn, on future greenhouse gas emissions. However, they do not reflect the potential effects of differences in average earnings between foreign-born and native-born workers. These gaps, which are believed to reflect differences in productivity or efficiency rather than the effects of discrimination, indicate that there is immigration-related heterogeneity in the population that is not reflected by the energy-economic model. These differences in labor productivity are potentially important, since we found that the aging effect on emissions acted primarily through changes in labor supply that affected the economic growth rate and hence emissions.

We carried out a first-order estimate of the likely immigration-related labor supply’s effects on the model’s projections of carbon emissions. Performing a full PET model analysis would require substantial new work on household projections, model development, and data analysis. Our aim was therefore to make a rough estimate of the likely effects based on economic data describing differences between immigrant and native-born households, and existing demographic projections. These results could then be used as a basis for deciding whether further analysis is warranted.

In summary, we found that, while the scale effects of immigration are, of course, very important to future emissions, the compositional effects are likely to be quite modest, lowering projected emissions by at most 10% over the course of the century. This estimate was based on three components: (1) estimated differences, by nativity of the householder, in earnings and consumption; (2) assumptions about the speed at which the foreign-born economic characteristics will converge, or assimilate, to the native-born characteristics; and (3) quantifications of the future nativity composition of the population (foreign-born and their descendents). These different elements were combined to estimate the aggregate effect of this compositional change on overall consumption and hence emissions.

Data availability was an important limitation, since household-level economic data by nativity are scarce. We used a combination of the CES, the Current Population Survey (CPS), and Census data (Public-Use Microdata Samples [PUMS] 5%) in our analysis. We found that large nativity effects occur only for Hispanic households (Asian households, for example, show only weak effects), and therefore focused our analysis on Hispanic immigration. We find that there are large gaps among the mean per person pre-retirement earnings in foreign-born Hispanic households, those in native-born Hispanic households, and those in non-Hispanic households, whether native or foreign-born. For those living in households where the householder is under age 45, the mean earnings in foreign-born households are below those in native-born third-plus generation Hispanic households by 27 %, in native-born third-plus generation non-Hispanic households by 49 %, and in foreign-born non-Hispanic households by 52 %. Similar gaps are seen in the next older age group—householder age 45 to 64 years.

Whether the cross-sectional differences in the earnings across immigrant generations are a reliable indicator of future differences has been a subject of many scholarly inquiries and much debate. The competing theories in the scholarly literature about immigrant adaptation have radically different implications for the long-term future of immigrants and their descendents, and no scholarly consensus has emerged. Therefore, our estimates incorporate a range of assumptions about future gaps in earnings and consumption for the descendents of Hispanic immigrants, corresponding to the range of impacts implied by the major theories. We develop two estimates that are intended to span the range of foreseeable gaps in mean per capita earnings of second-and-higher generation Hispanic households: one based on the assumption that their earnings will converge to the mean for the entire population (a rapid assimilation scenario) and the other based on the assumption that the earnings gap remains at the current (2002–2005, CPS) level for first-generation Hispanic households for all their descendents (an impeded assimilation scenario).

For estimating the projected foreign-born share of the total Hispanic population that would be appropriate to apply to our own household projections (and PET model analysis), we draw on U.S. Census Bureau results. This share declines steadily from the early years of the 21st century through 2100 in all three Census 2000 projection series, from 36% in 2000 to a range of 14 to 25% at mid-century and 6 to 12% at the end of the century. Disregarding the Census low series, in which the assumed levels of future Hispanic immigration are well below the lowest of the assumptions used in our own population and household projections, we can infer from the Census Bureau 2000 results that our projections imply a foreign-born share of the Hispanic population in the range of 20–25% in 2050, declining to 9–12% in 2100, with the higher figure applicable to the Large-Young scenario and the lower end of the range to the Small-Old scenario.

These ranges of the first-generation Hispanic population can be used to estimate the impacts of immigration under the rapid assimilation scenario. For the impeded assimilation scenario, we need to know not only the number of first-generation Hispanics, but also the number of second-generation plus the future number who are descendents of these cohorts, that is, in the third- and higher generations. Current data from the CPS, combined with projections of numbers of foreign-born Hispanics from the Census, indicate that there would be a total of between 81.8% and 83.0% potentially unassimilated Hispanics in the low and high series, respectively, in 2100.

Table 1 shows the results of combining our estimates of differences in earnings across households by nativity, our assumptions about convergence in earnings for two scenarios (rapid vs. impeded assimilation), and our projections of the fraction of the Hispanic population that is foreign-born, or second- or third-generation immigrants. We find that the largest impacts occur if assimilation is impeded and if immigration is high (high population scenarios). Even in this case, the reduction in earnings is less than 10%. Assuming that this reduction applies roughly to consumption as well (as established by our previous data comparison), a first-order estimate is that accounting for heterogeneity by nativity would have at most a 10% impact on emissions associated with household consumption and only by the end of the century, according to our most extreme assumptions. It is also worth noting that the results indicate that the most important assumption is how fast assimilation takes place, not how high the level of immigration is.

Table 1. Projections of Impacts of Immigration on Average Earnings, 2050 and 2100, for Three Population Scenarios

2000

2050

2100

Impacts on average earnings with current per household gap = –30.5%

Rapid assimilation scenario

Low

–1.3%

–1.0%

–0.5%

impact on total consumption relative to baseline

Middle

–1.3%

–1.5%

–0.9%

High

–1.3%

–2.1%

–1.2%

Impeded assimilation scenario

Low

–2.0%

–5.0%

–8.2%

impact on total consumption relative to baseline

Middle

–2.0%

–5.6%

–8.8%

High

–2.0%

–6.2%

–9.2%

Thus, we concluded from this analysis that a full PET model study of the composition effect of immigration is unwarranted at this time.

Policy Implications Report

Summary of Policy Implications. Our key finding is that, in a first analysis, aging of the U.S. population could reduce projected CO2 emissions by up to 37% over the second half of this century, relative to what they would otherwise be if aging were not accounted for. The maximum effect occurs in a scenario in which the most aging occurs, that is, a low population growth scenario with low fertility and low immigration. It should also be noted that in this first analysis, we did not consider potential adjustments, such as changes in labor force participation or of productivity at later ages. Nonetheless, we believe the result indicates that aging of the population may be an important factor to consider when assessing the outlook for long-term emissions in the United States.

There are several policy implications of this finding. First, since previous emissions scenarios have not considered this factor explicitly, it may be that those scenarios overestimate the magnitude of future emissions growth. The scale of future emissions is an important determinant of both the amount of climate change that might be experienced in the absence of mitigation policies and of the costs of achieving any particular emissions reductions. A firm conclusion on the effect of aging would be premature, but our results raise the possibility that it could have a numerically significant downward effect on emissions and point toward the desirability of further analysis. There is a growing literature on the influence of aging on economic growth, and this work should be better integrated with the long-term energy and emissions scenario modeling community to address this question. In particular, it indicates that there are potentially strong linkages among aging, social security systems, and economic growth. If aging is also important to energy use and emissions, as our work suggests, then there may be a link between policies affecting the social security system and those relevant to energy and emissions.

We also point out that a modeling framework which differentiates between subpopulations offers the possibility of modeling the distributional impacts of energy or emissions policies. We did not carry out emissions policy scenarios in this project, but we did develop a modeling framework that could separately assess the impacts of, for instance, a carbon tax or emissions cap on older versus younger households.

There were also two negative findings that are important. First, the effect of aging is not primarily through its effect on the composition of consumption. These changes turn out to have only minor effects on aggregate emissions. Rather, the aging effect is on economic growth, through its effect on labor supply.

Second, we found that explicitly accounting for differences between immigrant and non-immigrant households is probably unnecessary when assessing aggregate emissions. This is not to say that the effect of immigration on future emissions is not important. In fact it is quite important in that it is a key determinant of future population growth, which has an important scale effect on future emissions. However, some have argued that if it were taken into account that immigrant households typically have lower incomes and different consumption patterns relative to the rest of the population, the outlook for future U.S. emissions might be changed. We found this not to be the case. The impact of explicitly accounting for the nativity of households would probably lower long-term emissions projections by less than 10% at the most.


Journal Articles on this Report : 1 Displayed | Download in RIS Format

Other project views: All 14 publications 5 publications in selected types All 2 journal articles
Type Citation Project Document Sources
Journal Article Jiang L, O'Neill BC. Impacts of demographic trends on US household size and structure. Population and Development Review 2007;33(3):567-591. R829801 (Final)
  • Abstract: Wiley Online-Abstract
    Exit
  • Supplemental Keywords:

    population, consumption, emissions, climate change, households,, RFA, Economic, Social, & Behavioral Science Research Program, Scientific Discipline, Air, climate change, Economics, decision-making, Social Science, Economics & Decision Making, anthropogenic stress, atmospheric carbon dioxide, demographic, environmental monitoring, carbon emissions, energy generation, human population growth, population environment technology model, socioeconomic indicators, socioeconomics, greenhouse gases, human dimension, population abundance, demographics, global warming , energy consumption, ecosystem sustainability, behavior change, climate variability

    Progress and Final Reports:

    Original Abstract
  • 2003 Progress Report
  • 2004 Progress Report
  • 2005 Progress Report
  • 2006