Big picture
• In epidemiology, we often compare measures of
disease frequency between different groups
– Building on measures of disease module
• Typically comparing disease frequency between
groups with different exposures
• The goal is to ascertain associations between
exposures and outcomes and, ultimately, effects
of exposures on outcomes
Big picture
• Reminder on the distinction between
associations and effects
• An association tells us about probabilities of past
events
– Carrying matches is associated with lung cancer
Big picture
• An effect is causal and it tells us how
probabilities change if conditions change
– If you remove matches from pockets in the
population, does the rate of lung cancer decrease?
?
Big picture
• Measures of association quantify the strength
and direction of associations between exposures
and outcomes
• Strength of association:
– Is exposure weakly related to outcome? Strongly?
Big picture
• Direction of association:
– Does exposure increase risk of disease (risk factor)?
– Does exposure decrease risk (protective factor)?
– Is there no association (null result)?
Big picture
• Recall counterfactual framework illustrated by an
“ideal experiment”
• A hypothetical study which, if we could actually
conduct it, would allow us to infer causality
– Population experiences one exposure and observed
for outcome over a given time period
– Roll back the clock
– Change the exposure but leave everything else the
same, observe for outcome over the same time period
– Compare the outcomes under both exposures: this is
the causal effect
Big picture
• In reality, the “ideal experiment” cannot be conducted
• We evaluate the associations between exposures and
outcomes by comparing outcomes between groups that
experienced different exposures
Big picture
• Measure of association does not necessarily (or
even typically) quantify a causal effect
• Associations may be explained by various
sources:
– Causal relation
– Chance (random error)
– Bias (systematic error)
Big picture
• Throughout course will enumerate the methods
for:
– Designing studies to minimize error (systematic and
random)
– Assessing the roles of systematic and random error
once a study is conducted (qualitatively and
quantitatively)
– Incorporating knowledge or assumptions about the
causal process
– Analytically removing or adjusting for error to the
extent possible
Big picture
• These are key steps to moving towards inferring
causality from associations

2.1 big picture

  • 1.
    Big picture • Inepidemiology, we often compare measures of disease frequency between different groups – Building on measures of disease module • Typically comparing disease frequency between groups with different exposures • The goal is to ascertain associations between exposures and outcomes and, ultimately, effects of exposures on outcomes
  • 2.
    Big picture • Reminderon the distinction between associations and effects • An association tells us about probabilities of past events – Carrying matches is associated with lung cancer
  • 3.
    Big picture • Aneffect is causal and it tells us how probabilities change if conditions change – If you remove matches from pockets in the population, does the rate of lung cancer decrease? ?
  • 4.
    Big picture • Measuresof association quantify the strength and direction of associations between exposures and outcomes • Strength of association: – Is exposure weakly related to outcome? Strongly?
  • 5.
    Big picture • Directionof association: – Does exposure increase risk of disease (risk factor)? – Does exposure decrease risk (protective factor)? – Is there no association (null result)?
  • 6.
    Big picture • Recallcounterfactual framework illustrated by an “ideal experiment” • A hypothetical study which, if we could actually conduct it, would allow us to infer causality – Population experiences one exposure and observed for outcome over a given time period – Roll back the clock – Change the exposure but leave everything else the same, observe for outcome over the same time period – Compare the outcomes under both exposures: this is the causal effect
  • 7.
    Big picture • Inreality, the “ideal experiment” cannot be conducted • We evaluate the associations between exposures and outcomes by comparing outcomes between groups that experienced different exposures
  • 8.
    Big picture • Measureof association does not necessarily (or even typically) quantify a causal effect • Associations may be explained by various sources: – Causal relation – Chance (random error) – Bias (systematic error)
  • 9.
    Big picture • Throughoutcourse will enumerate the methods for: – Designing studies to minimize error (systematic and random) – Assessing the roles of systematic and random error once a study is conducted (qualitatively and quantitatively) – Incorporating knowledge or assumptions about the causal process – Analytically removing or adjusting for error to the extent possible
  • 10.
    Big picture • Theseare key steps to moving towards inferring causality from associations