Experiments (Exam 1)

Question 1

experiment

Answer 1

• statistician sets the levels of the explanatory variable
• more time consuming & complicated than observational study
• ONLY way to test cause & effect relationships

Question 2

what are the 4 principles of experimental design?

Answer 2

control, replication, blocking, randomness

Question 3

control

Answer 3

• we need a baseline for comparison
• in order to demonstrate the benefit of the treatment we care about, we need a baseline to compare it to (the control)

Question 4

replication

Answer 4

• the act of applying the explanatory variable levels to more than 1 subject
• NOT the same as repeated measurements

Question 5

block

Answer 5

• group of subjects known to be similar in some way that is thought to influence the response variable
• group assignment is carried out separately within each block to ensure that the blocking variable is balanced between the 2 groups

Question 6

lurking/nuisance/confounding factors

Answer 6

other variables that we can expect to affect our response variable

Question 7

randomness

Answer 7

• assignment of subjects to groups should be carried out randomly
• randomization should “average out” all of the remaining sources of variability in our response variable that have not been blocked
• not as precise as blocking, but allows
• for unbiased estimation of the group effects
• “block what you can and randomize what you cannot”

Question 8

experimental unit

Answer 8

one member of the initial sample that will be subjected to the experiment

Question 9

factors

Answer 9

• the explanatory/independent variables
• each has multiple levels:
• levels/treatments of a factor (the chosen values of the factor that are being varied in the experiment)
• treatment group (the group of experimental units assigned to a treatment)

Question 10

randomized comparative experiments

Answer 10

• the sample of experimental units is assigned to treatment groups using a purely randomized approach with NO blocking
• treatment groups don’t have to be the same exact size
• with increasing sample sizes, there are diminishing returns (growing the sample size is always important, but most crucial when the sample sizes are smaller)
• balance is really important when sample sizes are smaller (in 1’s or 10’s)

Question 11

randomized block experiments

Answer 11

• the sample of experimental units is divided into blocks, and treatments are randomized to experimental units separately within each block (done proportionately)
• need to split each blocking group evenly
• group together people who share 1 common characteristic

Question 12

matched pairs experiments

Answer 12

• the sample of experimental units is grouped into closely matched pairs of 2
• treatments are randomized within pairs such that one experimental unit in each pair is assigned the treatment and the other is assigned the placebo
• very difficult in practice, especially with humans
• benefit: can make individual-level comparisons, which is stronger than comparing 2 groups on average
• ex: a pre-test (control) and post-test (treatment) taken by the same individual, psychological studies performed on twins/children from the same home