Basically this is a question if I got the general idea behind the Type II Error right. I’ll set up a little reproducible example:

Given $p_0=0.6$ with an $\alpha$-level of $\alpha=0.1$. As Binomial tests are conservative with respect to $\alpha$ we first calculate (using R) the actual $\alpha$ used:

Calculating the actual $\alpha$-level:

$P(X< c_1)= 0.0123$

for(i in 1:10){ # lower bound
    if(sum(dbinom(0:10, 10, 0.6)[1:(i+1)])<0.05){
        c <- sum(dbinom(0:10, 10, 0.6)[1:(i+1)])
    else break


for(i in 10:1){ # upper bound
    if(sum(dbinom(0:10, 10, 0.6)[11:i])<0.05){
        d <- sum(dbinom(0:10, 10, 0.6)[11:i])
    else break

Hence we have an actual $\alpha$ of


Calculating the critical values $c_1$ and $c_2$:

The critical values $c_1$ and $c_2$ are:


qbinom(0.0123, 10, 0.5)


qbinom(1-0.0464, 10, 0.5)

Calculating Type II Error for $p_1=0.5$ and $p_2=0.7$:


pbinom(8, 10, 0.5) - pbinom(3, 10, 0.5)


pbinom(8, 10, 0.7) - pbinom(3, 10, 0.7)

Logic behind this:

The Type II Error $\beta$ with respect to a specific value of a parameter $\vartheta\in H_1$ is defined as $\beta_\vartheta = P(c_1\le X\le c_2|\vartheta\in H_1)$. The critical constants $c_1$ and $c_2$ are calculated under $H_0$ but the probability of $X$ being greater than or equal $c_1$ and less than or equal to $c_2$ is calculated under $H_1$. Is this correct?


1 Answer 1


Yes, the solutions are correct. Simply put, the probability of not rejecting the null $p_0$ is when $c_1<X<c_2$. Now Type II error happens when we don't reject the null and the alternative is true. For alternative $p_i$, this is the evaluation of $P(c_1 < X < c_2 | p=p_i)$, which is what you have done.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.