The implications of empirical and 1:1 substitution ratios for consequential LCA: using a 1 % tax on whole milk as an illustrative example

Life cycle assessment (LCA) is an impact assessment tool for “supporting decisions on the substitution between two product systems” (Weidema et al. 2009, p.6), and generally, two distinct types of LCA are identified: attributional LCA (ALCA) and consequential LCA (CLCA). ALCA aims to model the environmental impacts produced by the processes used in the life cycle of the product studied (Ekvall and Weidema 2004; Earles and Halog 2011). In contrast, CLCA aims to model the system-wide change in environmental impacts caused by a change in production volumes or product configuration (Ekvall 2002; Ekvall and Weidema 2004; Curran et al. 2005; Finnveden et al. 2009; Plevin et al. 2014). The development and use of CLCA is largely motivated by the principle that in order to make rational decisions, information is required on the consequences of the decision at hand (Weidema 1993; Ekvall 1999; Wenzel 1998; Plevin et al. 2014). Nevertheless, because CLCA has evolved out of ALCA, there are still a number of methodological features that remain in common, such as the assumption of a 1:1 substitution ratio between the two product systems being compared.

Plevin et al. (2014) argue that the assumption of a 1:1 substitution ratio is a feature of attributional LCA (though the recent development of advanced ALCA (AALCA) means this may not always be the case (Andrae 2015)) and that CLCA should model the actual rate of substitution that occurs. However, much of the existing guidance for CLCA adheres to the principle of functional equivalence between the two product systems in question, to ensure that the comparison is on a like-for-like basis (Weidema 2003; ISO 2006; European Commission et al. 2010). Similarly, there are many examples of CLCA studies that adhere to a 1:1 substitution ratio, for example, Ekvall and Andræ (2006) calculate the change in emissions achieved by the ban on the use of lead in solder by subtracting the LCA result for lead-based solder from the LCA result for a functionally equivalent quantity of lead-free solder. Even in CLCA studies that explicitly consider the price difference between competing production systems, such as Thiesen et al. (2008), the 1:1 substitution assumption is still present.

There are a number of studies that depart from this convention and model non 1:1 substitution ratios between competing products, particularly in the literature on biofuel policy (Smeets et al. 2014; Rajagopal 2013; Drabik and De Gorter 2011; Taheripour and Tyner 2013; Hochman and Rajagopal 2010). For example, Smeets et al. (2014) use a computational general equilibrium model to estimate a range of substitution ratios between 0.78 and 0.66 (i.e. for every one unit of biofuel produced, between 0.78 to 0.66 units of fossil fuel is substituted). These non 1:1 substitution ratios occur because the supply of biofuels reduces the price of oil, which in turn increases demand for oil elsewhere in the world. In addition, the increased production of biofuels increases GDP in supplying regions, which in turn increases demand for oil. Such studies use partial equilibrium (PE) or general equilibrium (GE) models, and it is precisely because these models tend not adhere to the assumption of functional equivalence that some authors suggest there is a distinction between traditional “biophysical” CLCA and CLCA based on equilibrium modelling (see for example Brandão et al. 2014, p.462). However, this distinction does not appear to be entirely clear-cut as there are examples of CLCA studies that use partial equilibrium modelling but also adhere to a 1:1 substitution ratio, e.g. Ekvall (2000) and Ekvall and Andræ (2006).

Given this background context, the goal of the present study is to further explore the importance of modelling the actual substitution ratio between competing product systems and the implications for CLCA guidance and practice. This is done using the illustrative example of a hypothetical tax on whole milk. The impact category considered is global warming potential, though the implications of the study are equally relevant to other impact categories.

A tax on whole milk was selected as milk is a major food commodity (Cederberg and Stadig 2003) and has received a correspondingly high level of attention in the LCA literature. Cederberg and Stadig (2003) provide an LCA of milk production and explore alternative methods for dealing with beef co-products from the dairy system. They suggest that “prospective” LCA studies (what would now normally be described as consequential LCA) should use the method of “substitution” when dealing with co-products, as this method provides information on “the environmental consequences of manipulating product systems” (Cederberg and Stadig 2003, p.350). Substitution involves identifying the product systems that are displaced by the co-products and crediting the displacement of those product systems to the decision studied, as the avoidance of those systems and their associated impacts are a consequence of the decision (Ekvall and Weidema 2004; Brander and Wylie 2011).

Thomassen et al. (2008) report the results from both an attributional and a consequential LCA for fat and protein-corrected milk. They also demonstrate the way in which the consequential approach aims to capture the system-wide effects of milk production by including the substitution effects of beef co-products from the dairy sector on dedicated beef and pork production. A similar study is provided by Flysjö et al. (2011), which models six alternative approaches for dealing with beef co-products from milk production, including substitution effects. Dalgaard et al. (2014) present a model for calculating the carbon footprint of energy corrected milk, with a fat content of 4.1 %. The consequential modelling option includes the substitution effects of both beef and manure co-products from the dairy system.

One co-product from the dairy system that is generally absent from the existing LCA literature is milk fat from the manufacture of low fat milk. This appears to be because existing studies tend not to differentiate between types of milk with different levels of fat or have selected whole milk as the functional unit, and therefore, milk fat co-products do not arise. In the case of Hospido et al. (2003), milk fat is identified as a co-product but is omitted from the analysis due to its small proportion of total production (>2.5 %).

One study that does consider milk fat in more depth is Flysjö (2012), where it is argued that skimmed milk and cream (i.e. milk fat) are both “determining” co-products, i.e. an increase in demand for either product will increase the production of both, and in such situations, the treatment of co-products in consequential LCA will be equivalent to allocation by economic value (Weidema et al. 2009). However, it is plausible that milk fat is not a co-determining product as alternative products exist, at least for some of the applications of milk fat (Ong and Goh 2002), and therefore, the long-run marginal production costs of the alternative products will constrain the price of milk fat and its contribution to the revenue from the co-producing unit process (Weidema et al. 2009). Although the primary contribution of the present article is concerned with the assumption of a 1:1 product substitution ratio in CLCA, a subsidiary contribution is an investigation of the displacement effects of milk fat co-products, which is generally absent from the existing literature on milk.

The condition of negativity is met since all the own price elasticities are negative and statistically significant. Table 2 shows the Marshallian cross-price elasticities of the different milk products with regard to a change in the price of whole milk. The results suggest that a 1 % tax on whole milk is likely to decrease demand for whole milk by 1.48 % and increase demand for low fat milk by 3.44 %. The elasticities for the other alternative milk products are not statistically significant and are not used in the remainder of the analysis.

Table 3 shows the CLCA results for low fat and whole milk, with a credit given to the avoided emissions from the milk fat co-product from the low fat milk. The highest estimate of palm oil emissions (36.15 tCO₂e/tonne of palm oil) gives in a net negative result (−190 gCO₂e/L of low fat milk), as the avoided palm oil emissions are greater than the rest of the life cycle emissions for low fat milk. The high estimate of palm oil emissions is due to deforestation and peatland drainage associated with palm cultivation. Depending on the size of the credit given for avoided palm oil production, the difference between the CLCA emissions for low fat and whole milk varies between a 10 % difference and a 120 % difference.

Table 4 shows the estimated change in consumption caused by a 1 % tax on whole milk in litres (derived from the volume data in Table 4 and the elasticities for low fat and whole milk in Table 2). Whole milk consumption decreases by a greater quantity than the increase in low fat milk consumption, indicating that there will be an overall decrease in the total amount of milk consumed.

Table 5 provides a comparison of the two different approaches to the substitution ratio between whole milk and low fat milk. The empirical substitution ratio is 0.52 (derived by dividing the change in low fat milk consumption by the change in whole milk consumption), i.e. for every 1 L reduction in whole milk consumption an additional 0.52 L of low fat milk is consumed. The estimated effects of the 1 % tax based on the 1:1 ratio approach and empirical ratio approach differ greatly, depending on the change in emissions caused by the production of low fat milk. In the extreme case, the 1:1 ratio may underestimate the reduction in emissions caused by the tax by 416 %.

As with any assessment of system-wide consequences, there are numerous sources of uncertainty with the estimated effects of the decision (Plevin et al. 2014). Some of these are indicated in the analysis of the tax on whole milk, e.g. by the magnitude of difference between the results from different parameter values for the emissions associated with palm oil (Table 5). One source of uncertainty that is worth highlighting, as it could be considerable, is the possibility of other consequences from the 1 % tax (which are not included in the model). The additional tax revenue could allow the government to reduce taxes elsewhere in the economy, causing an increase in demand in the affected markets, with associated increases in emissions (however, further econometric modelling would be required in order to support this claim). Alternatively, the government could spend the additional tax revenue itself, either increasing or decreasing environmental impacts, depending on whether the expenditure is on environmental protection measures or other items.

The main limitation of the conditional demand model (an almost ideal demand system) is the modelling of only milk products. The model’s assumption is that milk expenditures are fixed and thus remain constant during the modelled time period. Klonaris and Hallam (2003) have criticised this form of demand modelling and emphasised the need for converting to unconditional demand models. However, unconditional demand models require more data as they model the various food groups, and the problem of group expenditure being treated as exogenous in the model may still exist (Thompson 2004). The potential of using the conditional model with endogenous group expenditure could result in the violation of demand theory (Thompson 2004). Therefore, this paper accepts that there are limitations and benefits to using the conditional almost ideal demand system.

Despite these limitations, if the goal of a CLCA is to model the actual consequences from the decision at hand then modelling the actual substitution ratio does appear highly important. To add further illustrative detail to the milk example and assuming the Schmidt (2010) figure for palm oil, if a policy-maker were given the 1:1 emissions reduction estimate of 97 gCO₂e/L of milk substituted, then he/she may decide not to proceed with the tax. In contrast, the actual reduction may be over four times greater than the 1:1 estimate, and an effective mitigation opportunity would be missed. Adherence to a 1:1 substitution ratio may undermine the aim of estimating the actual consequences of the decision at hand.

However, there are at least three possible responses from adherents to the 1:1 ratio convention:

A slightly different argument in favour of empirical elasticities is that they are highly useful for identifying the comparator product, as well as the substitution ratio. The cross-price elasticities for skimmed, semi-skimmed, and soya milk were not statistically significant, which suggests that these products would not be affected by the tax on whole milk, and these products can therefore be excluded from the analysis of comparator systems. An alternative method for identifying the comparator product involves a three-step procedure for identifying obligatory and positioning properties, market segments, and the products within those market segments with the identified properties (Weidema et al. 1999; Ekvall and Weidema 2004). The use of empirical data to model the cross-price elasticity offers a more objective assessment of what the substitute product(s) will be.

A final point on the use of empirical substitution ratios is the importance of modelling the specific decision in question or type of intervention used to implement the change. An apparent assumption under-pinning the 1:1 ratio convention is that the decision in question is always a straightforward choice between two functionally equivalent products, whereas in reality, this is an oversimplification (as illustrated by the example of the tax). In the case of the regulation banning lead-based solder (Ekvall and Andræ 2006), the choice of a 1:1 substitution ratio may be a reasonable estimate, as the intervention itself does not involve the manipulation of price, in contrast to a tax, although there could still be a change in price and a reduction in total demand for solder, if lead-free solder has higher cost than lead-based solder. The different substitution effects from different types of intervention, such as taxes, regulations, and information campaigns, could be investigated in future research.

In terms of the subsidiary purpose of this paper, i.e. to contribute to the LCA literature on milk, a clear finding is that milk with different levels of fat content is likely to have very different greenhouse gas emissions consequences. In the extreme case, the additional production of low fat milk may create a net reduction in emissions, due to the substitution of palm oil. These effects have not previously been explored in the LCA literature.

A guiding principle of CLCA is that it should model the actual consequences of the decision at hand; however, the convention of assuming a 1:1 substitution ratio between comparator products does not necessarily adhere to this principle. This paper provides evidence to suggest that CLCA should model empirical substitution ratios as these may differ significantly from the 1:1 assumption. In the case of the ratio between whole milk and low fat milk, this was found to be 0.52 and not 1 as convention would suggest. The 1:1 assumption could lead to a large underestimation in the emission reductions caused by a 1 % tax on whole milk.

Based on the analysis presented above, it is recommended that existing guidance and standards, such as ISO 14044 and the ILCD Handbook are amended to allow the use of alternative substitution ratios and that the assumption of a 1:1 ratio should be viewed as a heuristic or default value in the absence of other information, rather than as a methodological principle.

Allowing the use of non 1:1 substitution ratios between competing products is a departure from what may be described as “biophysical” CLCA but appears to be wholly consistent with the aim of modelling the actual consequences of the decision at hand.