Valuing the Environment

Why we value the environment

The motivation

What: The DICE damage function (Ch. 2) needs a dollar (or "GDP-equivalent") value for non-market climate impacts — lost national parks, premature mortality, ecosystem services. So we need methods to VALUE the environment.

Why: Climate and biodiversity affect things people care about that have no market price; to weigh them in cost-benefit analysis and the SCC we must monetize them via willingness to pay (WTP).

Remember: Example: a Brown University thesis found ~11–17 US national parks that will LOSE the feature they're named for (e.g. Glacier National Park losing its glaciers) due to warming. For such losses we estimate a WTP-equivalent and add it to literal GDP impacts → GDP-equivalent loss.

Five types of economic value

What/Remember:

1. Direct use value — e.g. whale watching.
2. Indirect use value — e.g. a forest storing carbon.
3. Option value — preserving the option of future use.
4. Existence value — value of simply knowing whales exist (a NON-USE value).
5. Bequest value — value of leaving them for future generations (a NON-USE value).

Why: Use values often have a market trace; non-use values (existence, bequest) do not — they can only be captured via surveys/stated preferences.

Correlation, causation, and bias

Three interpretations of a correlation

What/Remember: If $A$ and $B$ correlate, three possibilities: (1) $A$ causes $B$; (2) $B$ causes $A$ (reverse causation); (3) a third factor $C$ causes both (common cause / omitted variable).

Why: The foundational caution before any causal claim in environmental economics.

Omitted Variable Bias (OVB)

What: OVB (Omitted Variable Bias) occurs when the observed association between $A$ and $B$ misrepresents the causal effect because an unmeasured variable correlates with both.

Why: The single biggest empirical pitfall in environmental economics.

Remember (classic): Across countries, PM2.5 (fine Particulate Matter under 2.5 micrometers) is NEGATIVELY correlated with health spending — NOT because pollution lowers spending, but because RICH countries have BOTH less pollution AND more health spending (income is the omitted common cause). To identify causal effects you need quasi-experimental variation (wildfires, election timing, disease outbreaks, wind shifts) plausibly UNCORRELATED with the unobserved drivers of the outcome.

The four valuation methodologies

Overview — pros and cons

Remember:

• Statistical (cross-sectional / time-series / panel): uses real-world data; risk of OVB; can't capture non-use values or counterfactuals beyond the historical range.
• Experiments (lab / field): clean causal interpretation; limited scale; may miss general-equilibrium/behavioral spillovers.
• Structural models: comprehensive; can simulate counterfactuals not seen in data; depend on uncertain assumptions/inputs.
• Surveys (contingent valuation): the ONLY feasible tool for non-use (existence/bequest) values; suffer from hypothetical bias.

What's needed for causal inference (statistical)

What: The variation in the explanatory variable must be plausibly UNCORRELATED with unobserved drivers of the outcome (conditional on controls).

Why/Remember: Chang et al. (2016) — pear-packing plant: PM2.5 driven by DISTANT wildfires is plausibly EXOGENOUS to local productivity drivers (like local heat), so it identifies the causal effect of air pollution on worker productivity. The trick is finding quasi-random variation.

Why structural models add value — the China PM2.5 example

What: A structural model captures channels a single regression/experiment cannot.

Remember: A 50% cut in Beijing PM2.5 raises GDP per worker by ~15%, of which only ~5.8 percentage points is the DIRECT health-productivity channel; the rest comes from high-skill-worker RETENTION + AGGLOMERATION effects. This is the flagship "which method could estimate THIS?" example — only a structural model captures the indirect retention/agglomeration channels.

Value of a Statistical Life

VSL definition

What: VSL (Value of a Statistical Life) = the additional cost individuals would collectively bear for safety improvements that, in aggregate, reduce expected fatalities by ONE:

$$\text{VSL} = \frac{\text{WTP for a risk reduction}}{\text{size of the risk reduction}}$$

Why: Lets us monetize mortality reductions for cost-benefit analysis (clean-air policy, climate damages).

Remember: It is NOT the value of any actual identifiable life — it's the value of a SMALL risk reduction times the many people exposed. Worked example: Job A = 2% fatality risk at $60k, Job B = 1% risk at $50k → workers accept $10k less for a 1% lower risk → $\text{VSL} = \$10{,}000 / 0.01 = \$1{,}000{,}000$.

Real-world VSL numbers

Remember: Switzerland's government CBA uses a central VSL ≈ CHF 7 million (2022) — which EXCEEDS per-capita lifetime income (people value life beyond their earnings). VSL income elasticity ≈ 0.6 → a 1% rise in income raises VSL ~0.6% (a classic short-answer/calc item).

Behavioral responses and structural modeling

Hoy no Circula (Mexico City)

What: A license-plate-based driving ban removing ~20% of cars from the road on workdays.

Why/Remember: It FAILED — Davis (2008) found NO decrease in air pollution, because households bought additional, older, dirtier cars to circumvent the ban (and shifted driving to off-peak hours/weekends). Canonical case for why STRUCTURAL / behavioral-response modeling matters: naive policy ignores how people adapt.

Stated vs. revealed preferences

Hypothetical bias

What: People OVERSTATE willingness to pay in hypothetical surveys relative to settings with real financial stakes.

Why: It means contingent valuation may OVERSTATE environmental benefits.

Remember: Barrage & Lee (2010) (study run in China): ~80% said they would donate hypothetically, but only ~30–50% (roughly half) actually paid when real money was on the line. The DIRECTION (hypothetical ≫ real) is the robust point; don't over-rely on the exact citation/percentages.

Contingent Valuation (CV)

What: CV (Contingent Valuation) = a survey method asking people their stated WTP for an environmental good — the only feasible way to estimate NON-USE (existence/bequest) values.

Why: Established as a legally-recognized valuation tool through major liability cases.

Remember:

• 1989 Exxon Valdez oil spill: CV estimated ~$3 billion in non-use damages; Exxon paid >$1 billion.
• 2010 BP Deepwater Horizon: CV estimated $17.2 billion non-use loss; BP paid $8.8 billion in natural-resource damages, >$65 billion total.

These cases established CV as a legally-recognized valuation tool, despite hypothetical-bias concerns.

Reading treatment/control graphs (causal-inference skill)

The counterfactual

What: The counterfactual = the outcome the TREATED unit would have had if the treatment had NOT happened.

Why: Without it, before/after comparisons confound the treatment with everything else changing over time.

Remember (recipe): The causal effect at any horizon = (actual outcome) − (counterfactual/control outcome) read straight off the graph AT that time point — NOT (after − before) for the treated group alone. Read the gap between the ACTUAL line and the COUNTERFACTUAL line both SHORT-RUN (just after treatment) and LONG-RUN (end of chart), since answers often give two numbers, e.g. "(−40, +10)." "Negligible" short-run effect = the two lines sit on top of each other right after the intervention.

Key formulas & one-line takeaways

Key formulas

$$\text{VSL} = \frac{\Delta \text{WTP}}{\Delta \text{risk}} \qquad \left(\text{e.g. } \frac{\$10{,}000}{0.01} = \$1{,}000{,}000\right)$$

$$\text{VSL income elasticity} \approx 0.6 \quad\Rightarrow\quad \%\Delta\text{VSL} \approx 0.6 \times \%\Delta\text{income}$$

$$\text{Causal effect}(t) = \text{actual outcome}(t) - \text{counterfactual outcome}(t)$$

One-line takeaways

• Five types of value: direct use, indirect use, option, existence, bequest (the last two are NON-USE values, capturable only by surveys).
• A correlation between $A$ and $B$ has three readings: $A\to B$, $B\to A$, or a common cause $C$ (omitted variable).
• OVB is the biggest pitfall: PM2.5 vs. health spending is negatively correlated across countries only because rich countries have both less pollution AND more spending — income is the omitted variable.
• Four methods: statistical (OVB risk), experiments (clean but small/no spillovers), structural (counterfactuals, but assumption-dependent), surveys/CV (only tool for non-use values, but hypothetical bias).
• Causal inference needs variation uncorrelated with unobserved drivers — Chang et al. (2016) used distant-wildfire PM2.5 at a pear-packing plant.
• Structural models capture indirect channels: a 50% Beijing PM2.5 cut raises GDP/worker ~15%, only ~5.8 pp of it direct health-productivity; the rest is high-skill retention + agglomeration.
• VSL = WTP / risk reduction; $10k for 1% lower risk ⇒ $1M. NOT the value of an actual life. Swiss CBA uses VSL ≈ CHF 7M (2022); VSL income elasticity ≈ 0.6.
• Hoy no Circula (Davis 2008) failed — households bought extra dirty cars; shows why behavioral/structural modeling matters.
• Hypothetical bias (Barrage & Lee 2010, China): ~80% donate hypothetically, only ~half with real stakes ⇒ CV may overstate benefits.
• Contingent Valuation underpinned major liability awards: Exxon Valdez 1989 (~$3B non-use est.), BP Deepwater Horizon 2010 ($17.2B non-use est.; $8.8B resource damages, >$65B total) — establishing CV as a legally-recognized tool.
• Read treatment/control graphs as actual − counterfactual at each horizon (short- AND long-run); a falling treated line can still mean zero/positive effect.