Uncertainty and framing in science

1
Uncertainty and framing in science

• Like social situations, frames indicate how we
  see the world.
• Difference between Newtonian bodies at rest vs.
  Aristotles notion of what makes bodies moves
• Many scientific frames are two value
• Either false or provisionally accepted as true.
• Either Chlorine (35) has an atomic mass of
  34.968, or it doesnt.

2
Why science uses a two-value frame

• Reason 1 The question is precise enough that it
  does not require elaboration.
• Reason 2 Attempting (and failing) to falsify
  precisely testable hypotheses can provide strong
  support for them (at least over time).
• Reason 3 Even if a hypothesis has deficiencies,
  scientists may provisionally accept it if it has
  not been falsified, because no better alternative
  exists.

3
What is a null hypothesis?

• A conservative approach to research.
• Relies upon repeated falsification for validity.
• Inductive research.
• Claims no relationship between variables, and
  then attempts to find a relationship.
• For example No cancers will result after
  continuous exposure to polychlorinated biphenyls
  at 5 ppb over a five year period.

4
Type I Errors- False Positives

• From a scientific view, Type I errors are very
  bad
• No desire to reject a hypothesis for wrong
  reasons
• If it is a null hypothesis, that means finding a
  relationship when none exists.
• Scientists will therefore be conservative in
  attributing relationships between variables.

5
Type II Errors- False Negatives

• This changes the burden of proof
• A Type II error is when one assumes that no
  relationship exists, when in fact it does.
• Eg, assuming no danger from PCBs when they are
  actually dangerous.
• This means that one would assume greater danger
  until risks are shown to be acceptable.

6
Uncertainty and choice in methodology

• Do you choose to minimize Type I or II risks?
• Cannot do both
• Do we assume that things are safe until they are
  shown to be harmful (traditional approach), or do
  we assume risk exists until shown otherwise
  (Precautionary Principle)?
• Methodological value choice
• Has implications for health and economics

7
Interpretation of evidence

• People may agree on risk numbers, but not the
  interpretation of them.
• When a range of uncertainty exists, do you take
  the lower (less likely) occurrence
• or assume that more likely events will happen?
• Known as Maximin rule
• This is the difference between Mill and Rawls

8
Magnitude of uncertainty

• It is not uncommon for up to SIX orders of
  magnitude to exist in risk assessment.
• This is the difference between 1 in 1 chance of
  death, or 1 in 1,000,000.
• Most causal pathways cannot be known until after
  the fact.
• To take no action is unethical
• to take complete action is impossible.
• Legal system favors the former. Why?

9
How to deal with uncertainty in science

• Ethical considerations should be used in science
• Decisions on choice of variables can have effects
• Areas that can be researched may depends upon
  politics

10
Example of risk analysis

• Variables related to assessment are interrelated
  mathematically
• One can decrease false positives and false
  negatives, but only at the expense of detection
  level or increase in sample size
• For example a disease with incidence of 8/10,000
  and 95 confidence interval, with relative risk
  of 3 as significant.
• This would require a sample size of 13,500
• Rarer diseases require larger samples

11
We cannot control all variables

• Despite desires for greater accuracy, resources
  do not allow for unlimited samples
• Ethical questions there, as well
• When faced with uncertainty in making a
  prediction, what course of action?
• Do we assume that predictions must be airtight
  before making them?
• Or can scientists make pronouncements without all
  possible (95) certainty?
• Can scientists be neutral in politics?

12
Wildlife protection uncertainty

• Wildlife studies are never certain
• Data interpretation is always a consideration
• Example What is the minimum viable population
  size (MVP)?
• Important questions, but no easy models

13
Example of the Florida Panther
14
FLA Panther is highly endangered

• Perhaps only 40 left
• Questions
• What is the MVP?
• How will suitable habitat be found?
• Will wildlife corridors (left) work properly?
• Is this worth the investment?
• Should we be directing energies elsewhere?

15
Example Racial bias in health research

• Ethical questions about samples are more than
  size
• Demographic groups must also be chosen
• Drug testing tends to be done on white males
• Automatically screens out effects on
• Children
• Racial minorities
• Women, esp. pregnancy
• Physiology can differ between sexes and ethnic
  groups

16
What form of rationality do we use?

• Ethical rationality
• Assesses the moral goodness or badness of a
  situation/alternative
• Can conflict with epistemic rationality
• May focus more upon reducing false positives
• Can we say that one is better than another?

• Epistemic rationality
• Rationality of belief
• Looks at outcomes and consequences
• Tries to indicate probability costs
• Often relies upon utility
• Requires the burden of proof be on the person
  risking false positives

17
Considerations of scientific rationality

• Credibility and scientific predictions
• People may worry that making incorrect
  predictions may later harm credibility
• Ethically and morally, however, this may not be a
  sufficient argument for not taking action
• Political pressures to give information
• Professional pressures to achieve certainty and
  reduce false positives
• The 95 rule is meant to ensure growth of
  knowledge and peer review

18
Why is peer review not more widely used?

• Resources
• Time and money must be devoted to ensuring a good
  peer review process
• This may admittedly conflict with political
  expediency
• Nature of ecological research
• One can either ensure certainty or provide
  politically useful information it is very
  difficult to do both

19
Arguments for pro-environment

• Remember that withholding judgment in a political
  environment pro development
• This is because the null hypothesis is generally
  a claim of no impact from development
• This is misleading, because not all costs and
  subsidies are considered equally
• Consider costs to destroying wetlands
• 50-80K/acre/year in water filtration
• Such costs are borne by taxpayers

20
Science in the courts

• If science is not always of the best quality in
  regulation, are the courts any better?
• The answer is no burdens of proof in court are
  different for civil and criminal law
• Tort law (civil) only requires a preponderance of
  evidence in favor of a causal connection
• This is different from the beyond a reasonable
  doubt of criminal law
• Criminal law limits false positives, civil false
  negatives

21
This means that science in courts

• need not be peer reviewed
• Only needs to be consistent with widely
  accepted theories
• Ferebee vs. Chevron Chemical Company 736 F.2d
  1529 (DC Cir. 1984)
• This is meant to keep out fringe theories, but
  its enforcement is haphazard, at best
• In the end, one only need convince a jury of
  causality, and the judge of admissibility.