Carnegie Mellon University

Assumptions, Uncertainty, and other Considerations with the EIO-LCA Method


The EIO-LCA method is a linear model.  Thus, the results of a $1,000 change in demand or level of economic activity will be 10 times the results of a $100 change in demand.  

The results represent impacts through the production of output by the sector with increased demand. For the most part then, the use phase and end-of-life phases are not directly included in the results.  However, additional analyses using the EIO-LCA method can model these life cycle stages. 

For example, modeling a $1 million increase of demand from the industry sector that produces automobiles represents the impacts from materials extraction, materials manufacturing, parts manufacturing, assembly, transport of good between these stages, as well as product design and testing of vehicle models - all activities prior to the final vehicle from the assembly line getting driven out the manufacturing facility gates.  That analyses of $1 million in the automobile manufacturing sector does not include impacts from the fuel used to drive the car during its useful life or the impacts of salvaging parts or landfilling materials from an end-of-life vehicle.  One could estimate the upstream impacts from the fuel consumption with the EIO-LCA method by doing an analysis for an increase in demand from the petroleum manufacturing sector.  Emissions from the use phase would need to be estimated using other methods.

Many assumptions go into creating the impact vectors (the values for the environmental effects and materials consumption).  Most data that we use are categorized by industry sectors using the North American Industry Classification System (NAICS) or other generic categories (e.g., the USDA categorizes farms by crop type).  These data do not directly map onto the IO sectors in the economic models.  We allocate values using weighted averages, or information from data sources or other publications.  See the documentation associated with the model of interest for information on specific assumptions made in creating the impact vectors.   

The IO models used for the various EIO-LCA models represent economies of a single nation.  Imports and exports, though, are a major part of any economy's transactions.  Imports are implicitly assumed to have the same production characteristics as comparable products made in the country of interest.  Thus, if a truck is imported and used by a U.S. company, the environmental effect of the production of the truck is expected to be comparable to those made in the U.S.  To the extent that overseas production is regarded as more or less of an environmental concern, then the results from the EIO-LCA model should be modified by adding additional transportation and logistics (e.g., for overseas delivery) as well as possibly adjustment for different production processes.


We are uncertain as to all the uncertainty in the EIO-LCA models available on the site.  Here are some of the most important:

  • Old Data: The data associated with each model are representative of the year of the model.  Thus, data for the 1997 U.S. Benchmark  model are from 1997, including the economic input-output matrix and the associated environmental data.  Care should be taken in using a model to replicate current conditions.  The changes in these data over time vary widely.  Economic input-output coefficients for stable industries (e.g., steel making, which has had similar processes for years) may be similar to past coefficients; however EIO coefficients for rapidly changing industries (e.g., computer manufacturing, which has rapid development of products and processes) may be very different over time.  Similarly, environmental data can change over time due to changes in process efficiency, regulations for pollutants, or production levels.
  • Uncertainty Inherent in Original Data:  All data incorporated into an EIO-LCA model is originally compiled from surveys and forms submitted by industries to governments for national statistical purposes.  The uncertainty in sampling, response rate, missing/incomplete data, estimations to complete forms, etc. from the original data remain as underlying uncertainty in the EIO-LCA models.  See the model documentation for references to the original data sources and refer to the documentation provided with the original data source for more information of uncertainty within a given data source. 
  • Incomplete Original Data:  Related to the uncertainty in the original data sources, some data used in the EIO-LCA models are incomplete, in that they underestimate the true values.  A good example of this is toxic release data.  In the U.S., only facilities which emit above a certain threshold of toxics or which fall into certain industry classifications are required to report their toxic emissions.  So, the actual value of toxic emissions reported is known to be lower than the actual level of emissions.   See the model documentation for references to the original data sources and refer to the documentation provided with the original data source for more information of uncertainty within a given data source. 
  • Aggregated Original Data:  As mentioned above, most data are categorized in a way that does not directly correspond to the economic input-output sectors used in the IO matrix.  For example, electricity use for commercial buildings is aggregated by the type of building (e.g., office space, retail space, etc.), not by sector (e.g., engineering consulting offices, accounting, etc.).  We make assumptions to allocate aggregated data to the most appropriate sector.  See the model documentation for more information about how aggregated data is allocated. 
  • Aggregation of Sectors:  The results of an EIO-LCA analysis represent the impacts from a change in demand for an industry sector.  Depending on the model chosen, an industry sector represents an collection of several industry types, and this aggregation leads to uncertainty in how well a specific industry is modeled.  For example, in the U.S. models, one sector represents Power Generation and Supply, which would include coal-fired plants with high levels of CO2 and particulate emissions as well as hydropower plants with virtually no CO2 or particulate emissions.  The results for impacts from the Power Generation and Supply sector thus represent the "average" impacts for generating electricity.  (Yet, we like to point out that the U.S. models designate one sector entirely for Tortilla Manufacturing, so the impacts for making tortillas are well-represented.)  Non-U.S. models are more aggregated, with up to only 100 sectors representing all industries.  See the model information for the number of sectors represented  in the economy of a given model.

Other Issues and Considerations

As an LCA tool, the EIO-LCA models are incomplete as only a limited number of environmental effects are included.  The EIO-LCA models use as the basis for data only those data which are publicly available (i.e., no proprietary data is included, all data sources are provided).  While industry specific data is available for a number of environmental effects, we do not have data for impacts such as habitat destruction, non-hazardous solids wastes, or non-toxic pollutants to water. Some data used in earlier models (e.g., fertilizers) are no longer collected at the national level due to efforts to minimize reporting burden of companies. Other sources and LCA methods will need to be consulted to account for a full range of environmental impacts.

The EIO-LCA method, models, and results represent the inventory stage of the LCA.  The results estimate the environmental emissions or resource consumption associated with the life cycle of an industry sector, but do not estimate the actual environmental or human health impacts that these emissions or consumption patterns cause.  For example, the U.S. models estimate the emissions of particulates to the air, but do not estimate the increased number of hospitalizations or deaths due to these emissions.  

Each EIO-LCA model uses economic data as the user-defined parameter of analysis.  Each model uses the currency of the country of origin (i.e., U.S. models should have $US as input, Germany model should have €  as input, etc.).  Similarly, the monetary values represent the value of the currency in the year of the model.  So, the 1997 U.S. Benchmark model is based on 1997 U.S. dollar values.  If current prices are used, they should first be converted to the model year with an appropriate economic index. The Statistical Abstract of the United States provides historical price indexes for the U.S. for the overall economy and for major commodity groups such as food, energy, and transportation.  

    For example, if you found prices for hospitalization for 2006 but wanted to use the 2002 U.S. Benchmark model, you would need to convert the prices.  The Statistical Abstract of the United States lists the consumer price index for medical care in 2006 as 336.2 and in 2002 as 285.6.   Dividing the 2002 medical CPI by the 2006 medical CPI results in a ratio of 0.85.  All 2006 prices should be multiplied by 0.85 for use in the model.  

Another consideration is the correct use of producer versus purchaser prices.  Most of the economic input-output models that form the basis for the EIO-LCA models represent the producer prices - the price a producer receives for goods and services (plus taxes, minus subsidies), or the cost of buying all the materials, running facilities, paying workers, etc.  The purchaser price includes the producer price plus the transportation costs of shipping product to the point of sale, and the wholesale and retail trade margins (the profit these industries take for marketing and selling the product).  For many goods, the producer prices can be far less than what a final consumer would pay (e.g., the producer price for leather goods in U.S. is approximately 35% of the final purchaser price).  For many services, where no goods are transported and wholesale/retail trade is limited, the producer price and purchaser price are often the same (e.g., barber shops and childcare).