Methodology Guide · Vendor-Neutral

Attributional vs Consequential LCA: Which Method, and When?

Published August 26, 2026 · 13 min read

Figure 1. The same 2,000 L CHO fed-batch batch, modelled attributionally against 26 US grid subregion averages and consequentially against a single natural gas combined cycle marginal unit. Computed with the Bioprocess LCA Calculator.

Quick verdict

Attributional LCA asks what share of existing emissions belongs to your product; consequential LCA asks what emissions change because your product exists. Use attributional for reporting, declarations and hotspot analysis, where average data must be auditable. Use consequential for decisions that change production volumes. The choice rarely reorders your process improvements, but it can decide whether a siting or procurement claim is worth anything at all.

Key differences at a glance

Different questions, not different accuracy. One allocates a share of what already happens. The other models what would change. Neither is a more correct version of the other.
Average versus marginal data is the practical fault line. Almost every disagreement between the two methods traces back to whether a factor describes the system as it is or the part of it that responds.
On our worked batch the method is worth 5.0%. A 2,000 L CHO fed-batch case moves from 3,980 to 3,783 kg CO₂e per kg when average electricity is swapped for a marginal unit. That is small.
But it is worth everything to a siting decision. Attributionally, upstate New York beats the dirtiest US subregion by 5.1×. Consequentially the same choice is worth nothing, because gas responds to new load in both places.
Standards let you pick, and require you to say which. ISO 14044 is method-agnostic. Most compliance frameworks built on it are attributional in practice because auditors need reproducible data.

Side-by-side comparison

Dimension	Attributional LCA	Consequential LCA
Question answered	What share of emissions is mine?	What emissions change if I do this?
Data type	Average, system-wide	Marginal, responding units only
Multi-output handling	Allocation by mass, energy or economic value	System expansion and substitution credits
Electricity factor	Grid or subregion average	Short- or long-run marginal mix
Boundary behaviour	Fixed, physical supply chain	Expands to any market that responds
Reproducibility	High: published factors, traceable	Lower: depends on the market model chosen
Auditability for reporting	Accepted by GHG Protocol, PEF, EPDs	Rarely accepted for disclosure
Decision support	Weak for volume-changing decisions	Designed for exactly that
Uncertainty	Data uncertainty, mostly quantifiable	Model uncertainty, often not
Typical bioprocess use	Product carbon footprint, hotspot ranking	Siting, capacity, procurement, policy
Our 2,000 L worked case	3,980 kg CO₂e/kg (400 g/kWh avg)	3,783 kg CO₂e/kg (379 g/kWh marginal)

Both columns come from the same inventory and the same open factor set. Only the electricity model differs. The full derivation is below.

Attributional LCA in detail

Attributional modelling draws a boundary around the physical supply chain that delivers your product and asks how much of the world's existing environmental burden sits inside it. Every input gets an average factor: the mean carbon intensity of the grid you draw from, the mean burden of producing a kilogram of polyethylene resin, the mean impact of treating a cubic metre of effluent. Where a process makes more than one saleable output, the burden is split between them by a declared allocation rule, usually mass, energy content or economic value.

The defining property is that these results are additive. If you summed the attributional footprints of every product made on Earth, you would recover global emissions exactly once, with nothing double-counted and nothing missed. That property is what makes the method suitable for reporting. It is also what makes it useless for certain decisions, because a share of a fixed total cannot tell you what happens when the total moves.

This is the mode almost everyone works in without naming it. Our guide to emission factors for bioprocessing is an attributional data set, the worked life cycle assessment example is a study of exactly this kind, and the Bioprocess LCA Calculator ships average factors by default because eGRID subregion values and the eLCI stationary combustion set are both averages.

When attributional wins

Any time the output has to survive an auditor. Product carbon footprints, Environmental Product Declarations, Scope 1, 2 and 3 inventories and supplier disclosures are all attributional by construction, because the reviewer needs to trace every number to a published source and get the same answer twice. It also wins for hotspot analysis, where the goal is to find the largest contributor rather than to predict a change. On our worked batch, electricity is 95.6% of the total under either method, and you do not need a causal model to see that.

Consequential LCA in detail

Consequential modelling starts from a decision rather than a product. It asks what physically changes in the world if the decision is taken, and it follows those changes wherever they go, including into markets that have no physical connection to your supply chain at all. Instead of allocating a multi-output process, it expands the system: if your process makes a co-product that displaces something else, that displaced production is subtracted as a credit.

The mechanism that matters most in bioprocessing is the electricity model. Attributional accounting charges you the average intensity of your grid. Consequential accounting charges you the intensity of whatever generator ramps up when your load appears, which is a very different plant. Nuclear and hydro run flat out whenever they are available and do not respond to a marginal megawatt-hour. Wind and solar are constrained by weather, not demand. The unit that follows load on most of the US system is a gas turbine, so the marginal factor is close to the intensity of gas generation regardless of how clean the regional average looks.

The idea was formalised by Weidema, Frees and Nielsen (1999), who set out how to identify a marginal production technology, and extended by Ekvall and Weidema (2004) into a general procedure for drawing inventory boundaries this way. Plevin, Delucchi and Creutzig (2014) put the argument at its sharpest, contending that using attributional results to estimate the climate benefit of a policy actively misleads decision-makers, because a share of current emissions is not a prediction of future ones.

When consequential wins

When the decision changes how much gets produced, or where. Siting a new facility, adding a suite, signing a renewable tariff, switching a feedstock at scale, or advising on policy are all questions about change, and a method built to apportion a fixed total cannot answer them. The cost is honesty about uncertainty: you have swapped data uncertainty, which can be quantified, for model uncertainty about how markets respond, which frequently cannot. Zamagni and colleagues (2012) survey that trade-off carefully, and Suh and Yang (2014) are blunter, warning that a consequential model can be built to produce almost any result if its market assumptions go unexamined.

The same batch, both ways

Arguments about method are cheap. Here is the difference measured on one inventory, using the 2,000 L CHO fed-batch case that the rest of this cluster is built on: 2,000 L vessel at 80% working volume, 3 g/L titer, 70% downstream yield, 14-day batch, 90% success rate. That produces 3.024 kg of product and consumes 28,759 kWh, of which cleanroom HVAC is 93.5%.

Only one thing changes between the two runs: the electricity factor.

The attributional factor is the eGRID subregion average, published by the US EPA and redistributed CC0 through the NETL ElectricityLCI project. The consequential factor has to be constructed, and the construction is worth showing because it is where a consequential study either earns or loses its credibility:

Step	Value	Source
Marginal generator	Natural gas combined cycle	Load-following unit on most US grids
Average operating heat rate	7,146 Btu/kWh (HHV)	US EIA, 2020 fleet average
Converted	7.5394 MJ/kWh (HHV)	1 Btu = 1055.056 J
Natural gas combustion	0.05030157 kg CO₂e/MJ (HHV)	eLCI stationary combustion, CC0
Marginal grid factor	379.2 g CO₂e/kWh	Product of the two rows above

Both the heat rate and the combustion factor are on a higher-heating-value basis, so they multiply directly. Mixing an HHV heat rate with an LHV emission factor is the single most common arithmetic error in this calculation and it inflates the answer by about 10%.

One caveat, stated plainly. That 379.2 g CO₂e/kWh is combustion only. It excludes upstream gas extraction, processing and transmission, including methane leakage, which a complete consequential study would include and which would raise the figure. Treat it as a lower bound. At a 20% upstream adder the marginal factor becomes 455.1 g/kWh and the batch result rises to 4,504 kg CO₂e/kg, which is higher than the attributional answer rather than lower. The direction of the method effect is not robust to that assumption, and any such study that does not state its upstream treatment should be read sceptically.

Result

Quantity	Attributional (400 g/kWh avg)	Consequential (379.2 g/kWh marginal)
Product per batch	3.024 kg	3.024 kg
Electricity	28,759 kWh	28,759 kWh
Batch footprint	12,036 kg CO₂e	11,439 kg CO₂e
Per kg of product	3,980 kg CO₂e/kg	3,783 kg CO₂e/kg
Difference	−5.0%

Five percent. Against a published mAb range of 4,000 to 20,000 kg CO₂e/kg, that is well inside the noise. If you had spent a week arguing about which method to use for a hotspot study, you wasted the week.

Where the method decides everything: siting

The 5% above is a national-average artefact. Run the same batch against each of the 26 eGRID subregions and the picture changes completely, because the attributional answer moves with the region while the consequential answer does not move at all.

Subregion	Grid average	Attributional	Consequential	Method effect
NYUP — Upstate NY	134.3 g/kWh	1,453	3,783	+160.3%
CAMX — California	240.4 g/kWh	2,462	3,783	+53.6%
NWPP — Northwest	297.3 g/kWh	3,003	3,783	+25.9%
SRMV — Mississippi Valley	382.0 g/kWh	3,809	3,783	−0.7%
ERCT — ERCOT Texas	460.0 g/kWh	4,551	3,783	−16.9%
RFCW — RFC West	567.7 g/kWh	5,575	3,783	−32.1%
MROE — MRO East	761.8 g/kWh	7,421	3,783	−49.0%

All values kg CO₂e per kg of product. Seven of the 26 subregions shown; the full set spans NYUP at the clean end to MROE at the dirty end.

Three things fall out of that table.

The consequential column is constant. One number, 3,783, for all 26 regions. That is not a modelling shortcut, it is the substantive claim: if a gas turbine follows your load in every one of those regions, then the regional average mix is a fact about the electricity already being generated and has nothing to do with the electricity you caused.

The method effect changes sign at 379.2 g CO₂e/kWh. Eight of the 26 subregions are cleaner than the marginal unit and get worse under the marginal method; the other 18 are dirtier and get better. Anyone on a clean grid who adopts consequential modelling is volunteering for a higher number, which is one honest reason the method is unpopular in disclosure.

Siting is worth 5.1× or nothing, depending only on the method. Attributionally, moving this batch from MROE to NYUP takes it from 7,421 to 1,453 kg CO₂e/kg, an 80.4% reduction, and no physical change to the process whatsoever. Consequentially, the same move is worth exactly zero. Two defensible methods, one decision, and answers that could not disagree more.

This is the practical reason the choice matters. It is also why a company can honestly report a low Scope 2 figure after relocating while a consequential analyst insists no emissions were avoided. Both are right about their own question. The failure mode is quoting one as if it answered the other, which is precisely the confusion Plevin and colleagues set out to name.

What the method does not change

It would be easy to leave with the impression that nothing in an LCA is stable. The opposite is true for the decisions most process engineers actually make.

Titer	Attributional	Consequential	Ratio
1 g/L	11,940	11,348	0.950
3 g/L	3,980	3,783	0.950
5 g/L	2,388	2,270	0.950
10 g/L	1,194	1,135	0.950

The ratio is constant to three decimal places. Swapping the electricity model rescales the whole result by a fixed factor and leaves every internal comparison intact. Raising titer from 1 to 10 g/L is a 90.0% reduction under both methods; it is still the largest process lever available, and it was never in doubt.

Ranked against the other choices available on this batch, at a fixed grid region:

Grid region, NYUP to MROE: +410.7%
Titer, 1 to 10 g/L: −90.0%
Remove cleanroom HVAC from the boundary: −89.3%
Attributional to consequential: −5.0%
LCIA characterisation method vintage: −0.005%

The method sits fourth, an order of magnitude below the physical levers. The twist is that it is the method choice which determines whether the item at the top of that list is a lever you are entitled to claim at all. That interaction, not the 5%, is the reason to get it right. If you want to see the bottom of that list derived, the life cycle impact assessment guide covers why characterisation is the least consequential choice in the whole study.

Pros and cons

Attributional strengths

Reproducible: published average factors, same answer twice
Accepted by GHG Protocol, PEF and EPD programmes
Additive, so results can be summed across a portfolio
Data is largely free and open for bioprocess inputs
Excellent for hotspot ranking, which is most of the value

Attributional weaknesses

Cannot predict the effect of a change in volume
Allocation choices are conventions, not physics
Credits siting and tariff choices that may avoid nothing
Encourages relocating a burden rather than removing it

Consequential strengths

Actually answers “what changes if we do this?”
Exposes claims that shuffle burden without cutting it
Handles co-products without an arbitrary allocation rule
The right basis for policy and capacity decisions

Consequential weaknesses

Market response models are assumptions, often unverifiable
Results are sensitive enough to be steered, deliberately or not
Marginal inventory data is mostly commercial, not open
Not additive: two studies cannot simply be summed
Rarely accepted for regulatory disclosure

Which should you use?

You have to publish the number

Customer questionnaire, EPD, CDP response, Scope 3 category 1. The reviewer needs to trace every factor to a source and reproduce the total.

Attributional

You are looking for the hotspot

Which stage dominates? On this batch, electricity is 95.6% under both methods and cleanroom HVAC is 93.5% of that. A causal model adds nothing.

Attributional

You are choosing a site or a tariff

The whole question is whether the choice avoids emissions or relocates them. Average-data accounting cannot distinguish the two; that is exactly what marginal modelling is for.

Consequential

You are adding or removing capacity

A new suite, a switch to perfusion, a tenfold volume increase. Production volumes move, so a fixed-share method is the wrong instrument.

Consequential

If you genuinely cannot tell which applies, run attributional first. It is cheaper, the data is open, and it will find your hotspot. Reach for a consequential model only when a specific decision hinges on a market response, and when you are prepared to defend the market assumption in writing.

Run both methods on your own batch

Every figure on this page came out of the free calculator. Pick a grid subregion for the attributional answer, then type 379 into the custom grid field for the consequential one. The gap you see is your method effect, on your process.

Open the Bioprocess LCA Calculator

Real-world use cases

CDMO

Customer asks for a product footprint

Average-data modelling, with the eGRID subregion of the actual site. Declare the boundary and the allocation rule. Anything else will fail a customer audit.

Site selection

Two candidate locations, different grids

Run both. Report the attributional pair because that is what will appear in disclosures, but make the decision knowing the consequential answer may be that the grids are equivalent.

Process development

Ranking improvement projects

The average-data method is sufficient. The ratio between options is method-invariant, so titer, yield and HVAC will rank identically either way.

Procurement

Evaluating a renewable tariff

The one case where the methods disagree about whether you did anything. Ask whether the contract causes new capacity; if it does not, a consequential model credits it with nothing.

System expansion: the other place the methods split

Electricity is the clearest divergence but not the only one. The second is what happens to multi-output processes, and in biomanufacturing that usually means incinerating single-use plastic with energy recovery.

Attributionally, burning polymer is a burden and nothing else. Polyethylene releases 3.137 kg of fossil CO₂ per kg burned, derived stoichiometrically from the repeat unit. Across the 82 kg of polymer in our single-use batch, incineration contributes 243 kg CO₂e.

Consequentially, the recovered electricity displaces generation somewhere else and earns a credit. Whether that credit is large enough to matter is a question with an arithmetic answer, and it is a useful one because it needs no market model at all. Setting the credit equal to the burden gives the net electrical efficiency at which incineration becomes carbon-neutral:

Polymer heating value	Break-even net electrical efficiency
40 MJ/kg	74.4%
43 MJ/kg	69.3%
46 MJ/kg	64.7%

Real waste-to-energy plants reach roughly 20 to 25% net electrical efficiency. The break-even is nowhere near reachable, so the conclusion is robust across any plausible heating value: energy recovery reduces the end-of-life burden by about a third, and cannot eliminate it. At 43 MJ/kg and 25% efficiency the net becomes 2.005 kg CO₂e per kg of polymer, taking the batch waste term from 243 to roughly 155 kg CO₂e.

That is a 0.7% change on a 12,036 kg batch. Worth knowing, not worth arguing about, and a good illustration of why the single-use versus stainless comparison comes out as close to a tie as it does.

What the standards and the software actually support

The modelling approach is a property of your data and your goal statement, not of the program you run it in. Every major package can do both; what varies is whether a suitable inventory exists.

ISO 14040 / 14044 — method-agnostic. Describes both approaches and requires only that the goal and scope statement declares which one is in use and that boundary, allocation and data selection are consistent with it.
GHG Protocol Product Standard — attributional in practice. Built for disclosure, so it demands traceable average data and allocation rules rather than market models.
EU Product Environmental Footprint — average-data based, with a prescribed dataset hierarchy that leaves little room for a bespoke marginal model.
openLCA — free and open source, runs either approach; the constraint is which system model your imported database ships.
SimaPro — commercial, supports both, widely used for attributional compliance work.
Brightway — open-source Python framework, the most flexible option for building a custom consequential model, and the usual choice in academic work.

The practical obstacle here is data, not tooling. Databases published with a consequential system model and a marginal electricity mix are mostly commercial and licensed in ways that restrict redistribution. That is why our own LCA calculator ships average factors: eGRID and the eLCI combustion set are open and freely redistributable, and no equivalent open marginal dataset exists. If you need a consequential electricity result, the custom grid field takes any figure you can justify, including the 379.2 g/kWh derived above. For a fuller treatment of the packages themselves, see our comparison of SimaPro, GaBi and openLCA.

Frequently asked questions

What is the difference between attributional and consequential LCA?

Attributional LCA asks what share of global emissions belongs to your product. It uses average data and allocates existing burdens. Consequential LCA asks what emissions change because your product exists. It uses marginal data and models the physical response of affected systems. The first is an accounting question, the second is a causal one, and they routinely give different numbers for the same process.

Which is better, attributional or consequential LCA?

Neither is better in general. They answer different questions. Use attributional for product declarations, corporate reporting, hotspot analysis and anything that has to be auditable, because average data is traceable and reproducible. Use consequential when the point of the study is to inform a decision that will change production volumes, such as a siting choice, a capacity expansion or a policy. ISO 14044 does not mandate either, but it does require you to declare which one you used.

Should I use average or marginal grid emission factors?

It depends on the question. An average factor tells you the carbon intensity of the electricity system you are part of, which is what you report. A marginal factor tells you the carbon intensity of the generation that responds when you add load, which is what actually changes when you switch a machine on. For a US site the two can differ by a factor of three in either direction: eGRID subregion averages range from 134 to 762 g CO₂e/kWh, while a natural gas combined cycle marginal unit sits at roughly 379 g CO₂e/kWh.

Does the choice of LCA method change which process improvements are worth doing?

Usually not. Modelled on a 2,000 L CHO fed-batch case, switching from an average to a marginal grid factor rescales the whole result by a constant 5.0 percent and leaves the ranking of process levers untouched: raising titer from 1 to 10 g/L is a 90 percent reduction under both methods. What the method does change is the value of choices that act through the electricity system rather than through the process, and siting is the clearest example.

Why does consequential LCA make my grid region irrelevant?

Because consequential LCA prices the generation that responds to new load, not the generation already on the system. Under attributional accounting the same batch is 1,453 kg CO₂e per kg in upstate New York and 7,421 in the Midwest Reliability East region, a 5.1-fold saving for siting well. Under a marginal model both come out at 3,783, because in both places the plant that ramps to meet one more megawatt-hour of biotech load is a gas turbine. The hydro and nuclear already on the New York system are fully dispatched and do not respond.

Can consequential LCA give a negative or avoided-burden result?

Yes, through system expansion, where a co-product or recovered energy is credited with displacing something else. Incinerating single-use plastic with energy recovery is the common case in biomanufacturing. It does not go negative here: polyethylene releases 3.137 kg fossil CO₂ per kg burned, and at a 43 MJ/kg heating value the recovered electricity would have to be generated at 69.3 percent net electrical efficiency to cancel that. Real waste-to-energy plants achieve roughly 20 to 25 percent, so recovery reduces the end-of-life burden by about a third rather than eliminating it.

Does ISO 14040 require attributional or consequential LCA?

Neither. ISO 14040 and 14044 are method-agnostic and describe both approaches without naming them as a mandatory choice. What the standards do require is that the goal and scope statement makes the modelling approach explicit and that the system boundary, allocation procedure and data selection are consistent with it. Most compliance frameworks built on top of ISO, including the GHG Protocol Product Standard and the EU Product Environmental Footprint, are attributional in practice because auditability demands reproducible average data.

Which LCA software supports consequential modelling?

The modelling approach is a property of the database and the system model rather than the software. openLCA, SimaPro and Brightway can all run either. The practical constraint is the inventory: consequential work needs a database published with a consequential system model and a marginal electricity mix, and those are mostly commercial. Our own calculator ships open average factors from eGRID and the eLCI stationary combustion set, which makes it attributional by default, though you can enter a marginal figure in the custom grid field and get a consequential electricity result.

Resources and references

Ekvall & Weidema (2004), The International Journal of Life Cycle Assessment 9:161–171. System boundaries and input data in consequential life cycle inventory analysis. The paper that turned the idea into a usable procedure, and still the clearest statement of how such a boundary is drawn.
Weidema, Frees & Nielsen (1999), The International Journal of Life Cycle Assessment 4:48–56. Marginal production technologies for life cycle inventories. The original method for identifying which technology actually responds to a change in demand.
Plevin, Delucchi & Creutzig (2014), Journal of Industrial Ecology 18:73–83. Argues that using attributional results to estimate climate-change mitigation benefits misleads policy makers. The strongest published statement of the case made in the siting section above.
Zamagni et al. (2012), The International Journal of Life Cycle Assessment 17:904–918. Lights and shadows in consequential LCA. A balanced review of where the method delivers and where its uncertainty becomes unmanageable.
US EPA eGRID. Public-domain source of the 26 subregion average grid factors used for every attributional figure on this page.