GEA1000 chapt 2 simpson's paradox and confounders

bro why do i have to take this

what is simpson’s paradox

• a trend appears in the majority of several groups of data

• but disappears or reverses when the groups are combined

why is a sliced bar graph used

• gives a way to compare 3 categorical variables

what is the important thing to note about 2 variables that may be correlated

correlation DOES NOT imply causation

define a confounder

• a third variable associated to 2 other variables

• which give rise to a distorted conclusion

how do you define a confounder provided there is sufficient data

let C be the confounder

let A be the independent variable

let B be the dependent variables

• if there is a positive/negative association between A and C

  • rate (C | A) > / < rate (C | NA)

• if there is a positive/negative association between A and B

  • rate (B | C) > / < rate (B | NC)

as long as there is an association between the third variable, the primary independent and dependent variables, the third variable is a confounder

how do you determine if a variable is a confounder

• measure and collect data on it

• if there is no measurement on the variable, there is no knowledge on whether it is a confounder

what can be done to control confounders

• non-randomised studies

  • eg observational studies

what are some limitations on non-randomised studies

• unsure if all confounders are controlled for

• limited conclusion

• evidence of ‘association’ not causation

what can be done to control confounders

random assignment

• acts as a general solution across all confounders

why do confounders occur due to association

• it arises due to unequal rates / 2 unequal proportions

what are the advantages about random assignment on confounders

• can control for confounders

what are some disadvantages for random assignment for confounders

not always possible

• due to ethical reasons or other constraints

what is a solution for confounders

• slicing method to control confounders