scholar_vit

На пальцах теория изложена тут: http://www.stat.yale.edu/Courses/1997-98/101/chisq.htm. Для практических вычислений удобна R (http://www.r-project.org), где этот тест делает функция prop.test:

> control <- 48177*c(.13,1-.13)
> small.win <- (5675-354)*c(.14,1-.14)
> large.win <- 354*c(.18,1-.18)


> prop.test(matrix(c(control,small.win), byrow=TRUE,nrow=2))

2-sample test for equality of proportions with continuity correction

data: matrix(c(control, small.win), byrow = TRUE, nrow = 2)
X-squared = 4.122, df = 1, p-value = 0.04233
alternative hypothesis: two.sided
95 percent confidence interval:
-0.0198992486 -0.0001007514
sample estimates:
prop 1 prop 2
0.13 0.14

> prop.test(matrix(c(control,large.win), byrow=TRUE,nrow=2))

2-sample test for equality of proportions with continuity correction

data: matrix(c(control, large.win), byrow = TRUE, nrow = 2)
X-squared = 7.3146, df = 1, p-value = 0.00684
alternative hypothesis: two.sided
95 percent confidence interval:
-0.091556474 -0.008443526
sample estimates:
prop 1 prop 2
0.13 0.18

> prop.test(matrix(c(small.win,large.win), byrow=TRUE,nrow=2))

2-sample test for equality of proportions with continuity correction

data: matrix(c(small.win, large.win), byrow = TRUE, nrow = 2)
X-squared = 4.025, df = 1, p-value = 0.04483
alternative hypothesis: two.sided
95 percent confidence interval:
-0.082599157 0.002599157
sample estimates:
prop 1 prop 2
0.14 0.18