Flashcards
Hypothesis testing B
Cards in this deck (8)
- 1t-testp-value doesn't tell us how big/significant effect is
use t-test with test stat = Signal / Noise
t = sample mean - miu 0 / std. dev./sqrt(n) - 2Significance level (False Positive rate) -> Type 1 Error [false rejection of H0]α = P(decide θ ≠ 0 | θ = 0)
i.e. probability of deciding drug is dangerous when it isn't
want α to be SMALL-> don't want false alarms - 3False negative rate (β) -> Type 2 Error [false non-rejection of H0]β = P(decide θ = 0 | θ )
i.e. probability of deciding drug is not dangerous when it is
want β to be SMALL -> don't want to put dangerous drug onto market - 4Power (Gamma, γ)γ = 1- β = P( decide θ ≠ 0 | θ)
i.e. Probability of deciding drug is dangerous when it is
- we want power to be LARGE - 5Power increases with:1. significance (other things being =)
2. sample size (other things being =)
3. effect size (other things being =) - 6Statistical significance filterreject H0 at 5% if abs|t|>2
mean > 2*s / sqrt(n)
mean < - 2*s / sqrt(n)
->This means for small sample/study, need big observed effect to declare a result
->if H0 rejected in small study, effect will be exaggerated - 7Publication biasphenomenon where studies more likely to be published if report a positive result
-> a form of survivorship bias - 8criticisms of null hypothesis significance testing recipe1. Hard to deal with multiple comparisons, esp if combined with publication bias
2. p-values NOT directly reflect effect size
-> can only be interpreted with sample size
Information
- Description
- Flashcards on Hypothesis testing B, created by avery pumpkin on 11/01/2019.
- Created by
- avery pumpkin
- Views
- 0
- Rating