Scribd
Upload
21 views12 pages

Power Power: Statistical Inference

Power is the probability of correctly rejecting the null hypothesis. It depends on factors like the sample size (n), difference between the null and alternative values, and standard deviation. Power increases with larger n, greater difference between null and alternative, and smaller standard deviation. Power can be calculated for t-tests and other common statistical tests using functions like power.t.test() that incorporate relevant distributions.

Uploaded by

nitin gupta
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
21 views12 pages

Power Power: Statistical Inference

Power is the probability of correctly rejecting the null hypothesis. It depends on factors like the sample size (n), difference between the null and alternative values, and standard deviation. Power increases with larger n, greater difference between null and alternative, and smaller standard deviation. Power can be calculated for t-tests and other common statistical tests using functions like power.t.test() that incorporate relevant distributions.

Uploaded by

nitin gupta
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

Power

Statistical Inference

Brian Caffo, Jeff Leek, Roger Peng


Johns Hopkins Bloomberg School of Public Health
Power
· Power is the probability of rejecting the null hypothesis when it is false
· Ergo, power (as it's name would suggest) is a good thing; you want more power

· A type II error (a bad thing, as its name would suggest) is failing to reject the null hypothesis
when it's false; the probability of a type II error is usually called β

· Note Power = 1 − β

2/12
Notes
· Consider our previous example involving RDI

· H0 : µ = 30 versus Ha : µ > 30

· Then power is

ˉ − 30
> t1−α,n−1 | µ = µa )
X
P(
s/√n

· Note that this is a function that depends on the specific value of µa !


· Notice as µa approaches 30 the power approaches α

3/12
Calculating power for Gaussian data
Assume that n is large and that we know σ

ˉ − 30
> z1−α | µ = µa )
X
1 − β = P(
σ/√n
ˉ − µ + µ − 30
> z1−α | µ = µa )
X a a
= P(
σ/√n
ˉ−µ
| µ = µa )
X a µ − 30
= P( > z1−α − a
σ/√n σ/√n

| µ = µa )
µa − 30
= P(Z > z1−α −
σ/√n

4/12
Example continued
· Suppose that we wanted to detect a increase in mean RDI of at least 2 events / hour (above 30).
· Assume normality and that the sample in question will have a standard deviation of 4;
· What would be the power if we took a sample size of 16 ?

- Z1−α = 1.645
- µa −30
−−
σ/√n
= 2/(4/√16 ) = 2
- P(Z > 1.645 − 2) = P(Z > −0.355) = 64%

pnorm(-0.355, lower.tail = FALSE)

## [1] 0.6387

5/12
Note
· Consider H0 : µ = µ0 and Ha : µ > µ0 with µ = µa under Ha .
√n(X ˉ −µ )
· Under H the statistic Z = 0
is N(0, 1)
0 σ
√n(µa −µ0 )
· Under H
a Z is N( σ , 1)
· We reject if Z > Z1−α

sigma <- 10; mu_0 = 0; mu_a = 2; n <- 100; alpha = .05


plot(c(-3, 6),c(0, dnorm(0)), type = "n", frame = FALSE, xlab = "Z value", ylab = "")
xvals <- seq(-3, 6, length = 1000)
lines(xvals, dnorm(xvals), type = "l", lwd = 3)
lines(xvals, dnorm(xvals, mean = sqrt(n) * (mu_a - mu_0) / sigma), lwd =3)
abline(v = qnorm(1 - alpha))

6/12
7/12
Question
· When testing Ha : µ > µ0, notice if power is 1 − β, then

| µ = µa ) = P(Z > zβ )
µa − µ0
1 − β = P(Z > z1−α −
σ/√n

· This yields the equation

√n(µa − µ0 )
z1−α − = zβ
σ

· Unknowns: µa , σ, n, β
· Knowns: µ0 , α
· Specify any 3 of the unknowns and you can solve for the remainder

8/12
Notes
· The calculation for Ha : µ < µ0 is similar
· For Ha : µ ≠ µ0 calculate the one sided power using α/2 (this is only approximately right, it
excludes the probability of getting a large TS in the opposite direction of the truth)

· Power goes up as α gets larger


· Power of a one sided test is greater than the power of the associated two sided test
· Power goes up as µ1 gets further away from µ0
· Power goes up as n goes up
√n(µa −µ0 )
· Power doesn't need µa , σ and n, instead only
σ
µa −µ0
- The quantity σ is called the effect size, the difference in the means in standard deviation
units.
- Being unit free, it has some hope of interpretability across settings

9/12
T-test power
· Consider calculating power for a Gossett's T test for our example
· The power is

ˉ−µ
> t1−α,n−1 | µ = µa )
X 0
P(
S/√n

· Calcuting this requires the non-central t distribution.


· power.t.test does this very well

- Omit one of the arguments and it solves for it

10/12
Example
power.t.test(n = 16, delta = 2/4, sd = 1, type = "one.sample", alt = "one.sided")$power

## [1] 0.604

power.t.test(n = 16, delta = 2, sd = 4, type = "one.sample", alt = "one.sided")$power

## [1] 0.604

power.t.test(n = 16, delta = 100, sd = 200, type = "one.sample", alt = "one.sided")$power

## [1] 0.604

11/12
Example
power.t.test(power = 0.8, delta = 2/4, sd = 1, type = "one.sample", alt = "one.sided")$n

## [1] 26.14

power.t.test(power = 0.8, delta = 2, sd = 4, type = "one.sample", alt = "one.sided")$n

## [1] 26.14

power.t.test(power = 0.8, delta = 100, sd = 200, type = "one.sample", alt = "one.sided")$n

## [1] 26.14

12/12

You might also like