ASMASYS

Income elasticity of willingness to pay for marine and coastal ecosystem services

Progress of the meta-analysis

Julian Sagebiel, Marlin Riede, Malin Wiese, Felix Sutermeister, Martin Hänsel, Marjella Ernst
Leipzig University · Work Group Nature’s Values · 11 November 2026

A global meta-analysis of valuation studies

We collect all non-market valuation studies that use a survey to value a marine or coastal ecosystem service and report the income of the respondents
From these studies we estimate the change in the value of marine and coastal ecosystem services with income, for all services together and for 4 groups of services
The estimates feed into the welfare function of the ASMASYS integrated assessment model, a model of climate, economy and ecosystems for assessing marine carbon dioxide removal

We published our plan for the search, eligibility rules, variables and models before the analysis, as a preregistration on the Open Science Framework (osf.io/z8kgx).

The income elasticity of willingness to pay

The income elasticity measures the response of willingness to pay for an environmental good to income.

Income elasticity =% change in willingness to pay% change in income

Above 1

Willingness to pay rises faster than income. The good behaves like a luxury good.

Below 1

Willingness to pay rises more slowly than income. The good behaves like a necessity.

With an elasticity of 0.5, people with 10% more income are willing to pay 5% more.

Progress since the 2025 ASMASYS Annual Meeting

Done

Preregistration

Literature search3,788 records

Double screening2,226 records

Full texts911 retrieved

Codebook

Ongoing

AI extraction685 of 911 studies coded

Blind human checks15 studies so far

First modelson the data coded so far

Freeze the dataafter all checks

Preregistered analysis

Integrationinto the climate-economy model

Paperwith open data and code

Aug 2024
start

Today

Milestone 3.2.1
Dec 2025
Overview of the studies in the meta-analysis

Milestone 3.2.2
Dec 2026
First meta-analysis results in the model

Milestone 3.2.3
Jul 2027
Meta-analysis complete, final results in the model

At the 2025 meeting we had 82 studies from the Ecosystem Services Valuation Database and Drupp et al. (2025). The new search replaces that sample.

From 3,788 records to 947 eligible studies

The purple bars are still growing. 226 full texts wait for coding.

374 studies from 75 countries

7,534 willingness-to-pay values so far. The US has the most studies (75), ahead of China (22), Australia (21) and the UK (20).

Estimating the income elasticity

Each willingness-to-pay value is one observation, together with the income of the respondents
We relate willingness to pay to income, both on a log scale. The slope of that line is the income elasticity
Values from the same study form a group, so a study with many values does not dominate (multilevel model)
Values from larger samples count more, and we control for 15 study characteristics, such as sample size, payment and ecosystem service
The main sample holds comparable values only (1,940 values from 213 studies)

This is the primary model we registered in advance. All other models are robustness checks.

Three samples of willingness-to-pay values

All values

5,976 values
338 studies

Every value we could convert to 2020 US dollars, whatever the study asked for

Comparable values main sample

1,940 values
213 studies

Only values that
• are mean willingness to pay
• are for an improvement
• are paid per year, once or monthly, not per visit
• are for the smallest change a study offers
• come from the model the authors prefer

Strict sample

1,009 values
127 studies

Comparable values that also
• state whether they are per household or per person, for both willingness to pay and income
• have well measured income
• are not very high outliers

The samples are nested. Each step removes values that are harder to compare, and the sample gets smaller.

1,440 model variants for one question

3Samples

all values
comparable values
strict sample

2Model types

multilevel model
simple regression

3Weightings

all values equal
by sample size
by sample size, no study dominates

2Study characteristics

without
with

2Top income class

lower bound × 1.5
lower bound × 2

2Negative values

dropped
made positive

2Extreme incomes

kept
top and bottom 1% dropped

5Very high values

kept
dropped, 4 thresholds

3 × 2 × 3 × 2 × 2 × 2 × 2 × 5 = 1,440 model variants, each a full model on the data coded so far

The 2.5 rule for very high values is the one we registered, and 1.5, 2 and 3 are its checks. We also estimate 24 models for each group of ecosystem services and 7 by type of payment.

Willingness to pay rises more slowly than income

Income elasticity

0.19 to 0.42

middle half of 1,440 model variants

Mean 0.32
Below 1 in all of them
2025 poster data: 0.53

Preliminary, based on the AI extraction before the human check.

Estimates by modelling choice

	Options	Mean elasticity
Sample	All values (5,976 values, 338 studies)	0.46
	Comparable values (1,940 values, 213 studies)	0.31
	Strict sample (1,009 values, 127 studies)	0.19
Model type	Multilevel model, values grouped by study	0.30
	Simple regression	0.34
Weights	All values count equally	0.29
	Values from larger samples count more	0.38
	Larger samples count more, no study dominates	0.29
Study characteristics	None	0.24
	15, such as sample size, payment and service	0.40
Data handling	Top income class, negative values, extreme incomes, outliers	all within 0.05

Purple marks the primary model. The mean elasticity is the average over all 1,440 variants with that option.

Estimates by sample and model

Our main result is the primary model on comparable values. All values give the narrowest intervals but mix values that are not comparable, such as fees per visit. The strict sample is the most homogeneous, but with 127 studies its intervals are wide.

Four groups of ecosystem services

Recreation and aesthetics covers tourism, recreation and scenic views
Biodiversity covers species and habitats
Regulating covers water quality, protection from erosion and floods, and climate
Non-use covers the value of knowing that an ecosystem exists and is kept for future generations

One value can belong to several groups.

Willingness to pay and elasticity by ecosystem service

	Studies	Mean willingness to pay	Median willingness to pay	Mean income	Elasticity, primary model (95% confidence interval)	Mean elasticity, 24 alternative models
Recreation and aesthetics	94	$93	$21	$48,700	0.76 (0.28 to 1.25)	0.41
Biodiversity	130	$108	$44	$49,700	0.55 (0.24 to 0.86)	0.42
Regulating	111	$137	$28	$54,600	0.43 (0.05 to 0.82)	0.27
Non-use	65	$111	$53	$57,300	0.66 (0.30 to 1.02)	0.48
All services	183	$94	$24	$50,700	0.53 (0.28 to 0.78)	0.40

Comparable values only, without one-off payments. Willingness to pay per household or person and year for the change each study values, in 2020 US dollars adjusted for purchasing power. The median willingness to pay is about 0.06% of income.

Elasticities by ecosystem service

The primary model uses comparable values. Each group has 24 alternative models from 2 model types, 3 weightings, with or without study characteristics, and 2 samples (all values and comparable values). The groups differ less than the uncertainty of each estimate.

Elasticities by type of payment

Primary model on all mean values for the smallest change, so that fees per visit are included. No study asks for a payment per hectare. Market products such as food are not in the data, because our eligibility rules exclude them.

Open challenges

Studies value different changes, for example a whole programme, 1 more km of beach or a fee per visit. The main analysis uses comparable values only (213 studies so far)
Willingness to pay refers to the change each study describes, not to a common unit such as a hectare. Only 126 of 374 studies report the size of the site
49% of the values report income only in classes, such as 1,000 to 2,000 a month, and we use the midpoint of each class
The AI flags 65% of the values for a human reviewer
We have few studies from Africa and South America

Thank you for your attention!

Contact: Julian Sagebiel, julian.sagebiel@idiv.de

https://asmasys.cdrmare.de/en

Backup

Screening and full-text retrieval

Screening

78%

of the 1,530 records screened by both teams got the same decision from each team

A machine-learning tool (ASReview) sorts the records, and 2 teams make the decisions
338 disagreements settled in a joint re-review

Full texts

96%

of the 947 eligible studies are available as full text

We used publisher interfaces, open access services and university licences
36 studies are still missing

Income reporting is the main reason for exclusion

209 of 311 excluded studies report no income of their sample. Many use income only inside a regression model.

Survey methods and years of publication

The years 2017–2020 are thin because those studies come late in the coding queue. 136 of the 226 studies still to code were published then.

All 1,440 model variants

Conversion to 2020 US dollars

Income classes to a mean, using the midpoint of each class (lowest class: upper bound × 0.75, open top class: lower bound × 1.5)
Monthly × 12, weekly × 52, yearly and one-off payments as they are
Adjust for inflation in local currency with national consumer prices (World Bank)
Convert with 2020 purchasing power parities
Drop income measured on a scale and incomes below 1,000 US dollars a year

	Values	Studies
Extracted values	7,534	374
Income in money terms	7,076	349
Willingness to pay positive	6,078	348
Converted to 2020 US dollars	6,048	344
Income at least 1,000 US dollars a year	5,976	338