ENVR 3400 - association and causation

what is an essential starting point for evaluating cause-effect relationships?

a statistically significant association between an exposure and an outcome

associations can be 3 things…

spurious, non-causal, causal

what is the most important thing you do in science?

statistical analysis

choosing the right stats will tell you…

• how to design an experiment

• if your data means ANYTHING

hypothesis

a scientific question you set out to answer using Socratic principles

null hypothesis

there is no statistical difference between treatments in your experiment

• assumes there is no effect, difference or everything observed is random noise

what can lead you to reject your null hypothesis, but most likely confirm your scientific hypothesis?

finding statistically significant differences between treatments

alpha error (false positive + type I error)

the error of concluding there is a statistically significant difference, change, or an effect when, in reality, there is not

beta error (false negative)

the error of concluding there is no statistically significant difference, change, effect when, in reality, there is

t statistic

is a mean of your experiment actually difference from a population (real) mean

• testing of null hypotheses in an experiment

ANOVA (analysis of variance)

tells you if any treatment in your experiment statistically significantly affects the outcome

basic regression model / regression

a practical model that is used to…

• explain outcomes (cause and effect?)

• test hypothesis about possible causes

how strong a correlation is depends on….

the value of the R² value — the closer the value is to 1, the stronger the correlation

statistical association in epidemiology

implies that the exposure is associated with a change in PROBABILITY of the outcome; it does NOT mean that the exposure CAUSES the outcome

• e.g. smoking is considered a cause of lung cancer, but not everyone who smokes gets lung cancer… if everyone stopped smoking, there would still be lung cancer… therefore, the probability of lung cancer goes down, so technically this is an association

spurious associations

false associations that usually result from sampling error or bias

alpha, or a P, of 0.05 says that there is a ___% chance that the results are correct

95% … but there is still 5 out of 100 people that would show false results. this is strictly due to sampling error

braiding knowledge

western knowledge is not “better” than traditional knowledge, nor vice versa

• there are always ways of knowing that are not limited to having the most statistically validated study

• based on centuries of observations about the ecology and geography of the environment and the organisms that exist within

bias

primary cause of spurious associations ; it is a type of systematic non-random error in conduct, design, or analysis of epidemiological studies

• e.g. case control study of EMR and the association with childhood leukemia → positive but spurious association

non causal associations

these are real associations but not causal

• usually represent secondary associations due to confounding factors

• the association exists because the exposure is associated with another factor that is in turn associated with the outcome

non-causal example: prospective cohort study that examined the incidence of gambling and cancer

nevada residents had higher rate of cancer than utah

but utah has high proportion of Mormons who abstain from smoking and drinking (major factors)

so even though the association is real, it is non-causal

causal associations

changes in the exposure produce changes in the outcome

• results need to be in context with statistics as well as weight of evidence

what other factors and uncontrolled variables in epidemiological studies can affect the outcome?

humans have different epigenetics and respond differently to different exposures/stimuli

lab studies rely on very well-defined strains of mice to reduce the amount of variability in experimental studies