4. Hypothesis Testing Flashcards
What is a null hypothesis?
Takes the sceptical viewpoint
Crafted as ‘no difference’ where a difference in a parameter is being measured
What is an alternative hypothesis?
An alternative parameter value that is proposed by the H0
Crafted for the sake of testing something meaningful to an investigator
What is a 1-sided test?
A parameter being greater than or less than its value predicted by H0
What is a 2-sided test?
A parameter being different H0
What is a critical value?
A values used to compare to a test static to avoid errors
What is a p-value? what does it tell you?
A value that determines the probability of an observation
What is Type I error? What level of Type I error is considered standard in science?
Rejecting the null hypothesis when the null hypothesis is true
5% is an acceptable type 1 error
hence Fisher proposed CV is alpha = 0.05
What is Type II error? How does Type II error relate to statistical power?
Failing to reject the null hypothesis when the null hypothesis is false
The critical value beta is used as an acceptable type 2 error.
Statistical power = 1-beta
(used to calculate the sample size needed in an experiment)
What are some reasons we would fail to reject H0 when it is false (type 2 error)?
The size to the effect we are measuring is small
The sample size is too small
The variance (σ ) is too large
Explain what is meant by: Correlation is not causation
the appearance of correlation does not mean the explanatory variable can be used to predict the response variable
What is a ‘hidden’ variable?
a variable in a belief network whose value is not observed for any of the examples
there is no column in the data corresponding to that variable
What is Simpson’s Paradox?
the existence of data for which a
statistical association holds for a population but is reversed in a subpopulations
What is Occam’s razor?
among the theories that fit the data equally well, choose the simplest
theory
