In: Statistics and Probability
A claim has been made that only 5% of men in the U.S. play golf. If I want to be 90% confident, and have 90% statistical power, what sample size would I need to disprove this claim if the true percentage of men playing golf is 8%?
Null hypothesis: Proportion of men play goal, P_0 = 5% = 0.05
Alternate hypothesis: Proportion of men play goal,P_1 = 8% = 0.08
Criteria : reject H0 if p < p_critical
90% confidence means ( 100 - 90 )% = 10% alpha level.
Since, the value of null hypothesis is less than alternate hypothesis, the test is one tailed.
power of test = 90% = 0.9
alpha = 10% = 0.1
Under H0
Variance = V_0 = P_0*( 1 - P_0 )/n = 0.05*(1-0.05)/n = 0.05*0.95/n = 0.0475/n
standard deviation, S_0 = sqrt(V_0) = 0.2179/sqrt(n)
Under H1
Variance = V_1 = P_1*( 1 - P_1 )/n = 0.08*(1-0.08)/n = 0.08*0.92/n = 0.0736/n
standard deviation, S_1 = sqrt(V_1) = 0.2713/sqrt(n)
alpha = P[ Reject H0 when it is true ] = P[ p < p_critical | H0 ] = P[ ( p - 0.05 ) / S_0 < ( p_critical - 0.05 ) / S_0 ] = P[ Z < ( p_critical - 0.05 ) / S_0 ] = 0.1
P[ Z < Z_alpha ] = 0.1
therefore, Z_alpha = 1.28 ( 10% alpha for one tailed )
Comparing
( p_critical - 0.05 ) / S_0 = 1.28
p_critical - 0.05 = 1.28*S_0 ........................................(1)
Power of test = P[ Reject H0 when it is false ] = P[ p < p_critical | H1 ] = P[ ( p - 0.08 ) / S_1 < ( p_critical - 0.08 ) / S_1 ] = P[ Z < ( p_critical - 0.08 ) / S_1 ] = 0.9
P[ Z < Z_beta ] = 0.9
therefore, Z_beta = -1.28 ( 90% power for one tailed )
Comparing
( p_critical - 0.08 ) / S_1 = -1.28
p_critical - 0.08 = S_1*(-1.28) ..........................................(2)
from (1)
p_critical = 1.28*S_0 +0.05
substitute in (2)
1.28*S_0 +0.05 - 0.08 = S_1*(-1.28)
1.28*S_0 + 1.28*S_1 = 0.03
1.28*( 0.2179/sqrt(n) + 0.2713/sqrt(n) ) = 0.03
1.28*( 0.4892/sqrt(n) ) = 0.03
0.6262 = 0.03*sqrt(n)
sqrt(n) = 0.6262/0.03 = 20.8725
n = 20.8725^2 = 435.6 = 436 ( approx )