Lecture 2

Question 1

Parameters are (?), but at least we know it exists as a (?)

Answer 1

unknown; fixed constant

Question 2

Using statistics, we can only make a statement about parameters (population characteristics), but not about each individual. We acknowledge that all individuals are (?)

Answer 2

different

Question 3

a p< 0.05 criterion explains that the rate of Type 1 error is controlled at (?)

Answer 3

5%

Question 4

P(Type 1 error) = α =

Answer 4

(# of experiments rejected by H0) / (# of all experiments)

Question 5

Power = 1 - P(Type 2 Error) = 1 - β =

Answer 5

(# of experiments rejected by H0 when H0 is false) / (# of all experiments)

Question 6

Power depends on your:

Answer 6

sample size and the decision rule

Question 7

If the null hypothesis (H0) is really true, the proportion of p values less than the 0.05 (α) threshold is:

Answer 7

5%. In other words, Type 1 error is controlled by setting a reasonably low α threshold

Question 8

If the alternative hypothesis (H1) is really true, the proportion of p values less than 0.05 (α) will be:

Answer 8

greater than 5% (which is a good thing). It is actually called 1 − β (power)

Question 9

Power is increased by 3 the following factors:

Answer 9

Large sample size : it is intuitive as data close to population would lead improved replicability
Liberal decision rule. It comes with an increased Type 1 error rate (tradeoff). Discouraged in most cases
Increased signal-to-noise ratio also called as effect size. Need to get noiseless measure whenever possible