How it works
This page explains where the numbers come from. It starts in plain language, then gives the detail a level down for analysts who want it.
In one paragraph
People in Australia, France and Italy answered a survey in which they chose between different vaccine mandate designs and the option of no mandate. A two-class latent-class choice model uses those choices to estimate how likely a person is to support any given design. The tool combines that support estimate with an assumption about lives saved, a money value for each life saved, and any costs you enter, to show benefits, costs and how they compare.
Public support
Support is estimated using a two-class latent-class choice model. The estimate combines class-specific support probabilities using the estimated size of each class for the selected country and outbreak scenario. The model groups people into two preference classes, a supporter class and a resister class, each with its own preference weights. For a selected design the tool works out support within each class, then averages the two using the class shares.
For a selected policy bundle, support is calculated as:
P(support) = Σc πc × exp(Vpolicy,c) / [exp(Vpolicy,c) + exp(Vno mandate,c)]
where πc is the estimated class share for class c, Vpolicy,c is the class-specific utility of the selected mandate bundle, and Vno mandate,c is the class-specific utility of the no-mandate alternative. The Policy A display constant is set to zero because the tool predicts support for a generic policy bundle rather than a left-versus-right experimental alternative. The same fixed coefficients are used every time, so the support figure for a given design is always the same and results are reproducible.
The percentage is an estimate of acceptability from stated preferences. It is not a forecast of real world behaviour, compliance or political feasibility. Treat it as one piece of evidence among several.
Estimated class shares
The class shares are the estimated size of each preference class for the selected country and outbreak. They sum to one within each scenario.
| Scenario | Supporter class | Resister class |
|---|---|---|
| Australia, mild | 0.747 | 0.253 |
| Australia, severe | 0.781 | 0.219 |
| France, mild | 0.724 | 0.276 |
| France, severe | 0.771 | 0.229 |
| Italy, mild | 0.712 | 0.288 |
| Italy, severe | 0.761 | 0.239 |
Class-specific utility coefficients
Each class has its own weights for the no-mandate constant and the design features. The reference design is high risk workers only, medical exemptions only, and lifting at 50 percent, so its features are coded zero. Lives saved is multiplied by the selected value in lives saved per 100,000.
| Scenario | Class | No mandate | All occupations | Medical or religious | Medical, religious or personal | Lift at 70% | Lift at 90% | Per life saved |
|---|---|---|---|---|---|---|---|---|
| AU mild | Supporter | -1.010 | -0.193 | -0.179 | -0.210 | 0.101 | 0.167 | 0.039 |
| AU mild | Resister | 2.948 | -0.266 | 0.108 | 0.153 | -0.095 | -0.261 | 0.015 |
| AU severe | Supporter | -0.792 | 0.116 | -0.152 | -0.233 | 0.162 | 0.241 | 0.045 |
| AU severe | Resister | 2.725 | -0.012 | -0.094 | 0.053 | 0.104 | 0.055 | 0.010 |
| FR mild | Supporter | -0.627 | -0.112 | -0.161 | -0.149 | 0.119 | 0.186 | 0.034 |
| FR mild | Resister | 2.779 | -0.192 | 0.069 | 0.180 | -0.024 | -0.025 | 0.009 |
| FR severe | Supporter | -0.439 | 0.059 | -0.124 | -0.176 | 0.149 | 0.260 | 0.036 |
| FR severe | Resister | 2.565 | -0.234 | -0.115 | -0.024 | 0.148 | 0.208 | 0.001 |
| IT mild | Supporter | -0.867 | -0.177 | -0.133 | -0.229 | 0.132 | 0.172 | 0.028 |
| IT mild | Resister | 2.716 | -0.256 | -0.162 | 0.033 | -0.135 | -0.194 | 0.008 |
| IT severe | Supporter | -0.633 | 0.170 | -0.120 | -0.224 | 0.195 | 0.354 | 0.033 |
| IT severe | Resister | 2.749 | -0.089 | -0.169 | 0.107 | -0.074 | -0.029 | 0.003 |
Reading example: in Australia under a severe outbreak the supporter class makes up about 78 percent of people and strongly favours mandates, while the resister class makes up about 22 percent and has a large positive no-mandate constant, so it rarely supports a mandate. Overall support is the share-weighted average of the two.
Lives saved and money value
You set how many lives are saved for every 100,000 people. The tool scales that up to the population you choose. For example, 25 lives per 100,000 across one million people is 250 lives. That total is multiplied by the money value you give to each life saved to get the health benefit in money terms.
The survey covered 10 to 40 lives saved per 100,000. If you choose a value outside that range, the tool shows a clear extrapolation warning, because the model is being used beyond the data it was built on.
Four ways to value a life saved are offered: the value of a statistical life, the value of a life year, a value per quality adjusted life year, and health system savings per life. Pick the one that matches your guidance. The other health effects shown, hospital admissions, intensive care admissions and working days, are rough conversions from lives saved and are reported separately. They are not added to the money value, to avoid double counting.
Costs and the ratio
Costs are optional. If you enter them, the tool sums six categories: digital systems, public information, enforcement, monitoring and compensation, administration, and other costs. Net benefit is the health benefit minus total cost. The benefit to cost ratio is the health benefit divided by total cost. A ratio above 1 means benefits are larger than costs under your assumptions.
The typical costs button fills these fields with indicative figures scaled to your population, time period and outbreak situation. They are starting points, not official estimates, and should be replaced with country guidance before any formal use.
Indicative cost figures, per one million people per year
| Category | Australia | France | Italy |
|---|---|---|---|
| Digital systems | 1,200,000 | 1,000,000 | 900,000 |
| Public information | 800,000 | 700,000 | 600,000 |
| Enforcement | 1,800,000 | 1,500,000 | 1,400,000 |
| Monitoring and compensation | 2,200,000 | 1,800,000 | 1,600,000 |
| Administration | 800,000 | 700,000 | 600,000 |
| Other | 500,000 | 400,000 | 400,000 |
These figures are multiplied by your population in millions, by the number of years, and by an outbreak factor of 0.8 for a mild situation or 1.3 for a severe outbreak.
Predicted support by preference class
The support chart shows support within each preference class and the share-weighted overall figure. Because the two classes can pull in different directions, the overall estimate reflects both how each class views the design and how large each class is in the selected scenario.
What this tool does not do
It does not decide whether a mandate is right. It does not capture fairness across groups, legal questions, enforcement practicality, trust, or politics. Those need separate judgement and stakeholder input. The figures are a structured starting point for discussion, not an answer.
Version
eMANDEVAL Future version 4.4.0. Predicted public support now uses a class-share-weighted two-class latent-class choice model. The benefit and cost calculations, interface, accessibility features and exports are otherwise unchanged.
Back to the tool