running.test() 仅适用于偶数 n1 和 n2

Question

I have a list of vectors (dados_obs) that contains "0" and "1". Head and Tails.

I have to test it's randomness. So I made a huge simulation to get the p-value and compare with my list of vectors of 0 and 1 to test if they were made up or actually random. It worked fine

I discovered it runs.test does that for me, but i have a problem. It only works when my number of 0's and 1's are evenly split with n1 = n2 = 50, with n = 100.

runs.test(dados_obs[[16]])

Gives me

Runs Test

data:  dados_obs[[16]]
statistic = 0.60305, runs = 54, n1 = 50, n2 = 50, n = 100, p-value = 0.5465
alternative hypothesis: nonrandomness

But

runs.test(dados_obs[[17]])

Gives me

    Runs Test

data:  dados_obs[[17]]
statistic = NaN, runs = 1, n1 = 0, n2 = 49, n = 49, p-value = NA
alternative hypothesis: nonrandomness

Is there a way to overcome this limitation? When n1 differs from n2 (Sum of Head differs from Sum of Tails)?

Answer 1

The runs.test from randtests looks like it has not been updated since 2014. Maybe try the one in snpar ? (Also need magrittr for pipes.)

library(snpar)
library(magrittr)

For example:

> sample(c(0,1),20,replace=TRUE) %>% snpar::runs.test()

    Approximate runs rest

data:  .
Runs = 13, p-value = 0.3581
alternative hypothesis: two.sided

> sample(c(0,1),100,replace=TRUE) %>% snpar::runs.test()

    Approximate runs rest

data:  .
Runs = 43, p-value = 0.1146
alternative hypothesis: two.sided

Answer 2

Actually, I had the same problem as you. Then I added the cut point that i used, in the syntax (with randtests package). Example with your code :

runs.test(dados_obs[[17]], threshold = mean(dados_obs[[17]]))

The cut points can be mean, mode, median, etc. We used it to specified the n1 and n2.