Z-Tests tests of Two Proportions for Continuous Data? Why can't we do z-tests of two proportions for data from a continuous variable?  For Example, if I have sample sales data (continuous, dollar) from a company and want to see if the proportion of sales of a certain product group has changed from one year to the other, can I roll up the products to product group and just use sales/total sales and compare the two years using z-tests of two proportions?  Here are the variables:
Product  |  Product Group | Sales | Year
 A: After reading Glen_b's comment, I have changed my tune and see major issues with this.
Suppose you want to see how \$3 billion is distributed between two products, say cars (0) and boats (1), in year 0 and year 1. In year 0, boats receive \$2 billion and cars receive \$1 billion. In year two, it is reversed to \$1 billion for boats and \$2 billion for cars.
Since we are dealing with billions of dollars, lets set up the Fisher test in that unit.
m1 <- matrix(c(1, 2, 2, 1), 2, 2)
fisher.test(m1) 

################################################################################

Fisher's Exact Test for Count Data

data:  m1
p-value = 1
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
  0.002852567 14.836128998
sample estimates:
odds ratio 
 0.3219834 

Unsurprisingly, the p-value is quite high.
However, we could have measured in millions of dollars. Let's do the Fisher test in that unit, so that \$1000-million are spent on cars in year 0, etc.
m2 <- matrix(c(1000, 2000, 2000, 1000), 2, 2)
fisher.test(m2) 

################################################################################

    Fisher's Exact Test for Count Data

data:  m2
p-value < 2.2e-16
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
 0.2241996 0.2787648
sample estimates:
odds ratio 
 0.2500625 

Suddenly, the p-value is tiny. No data have changed, just the way we are counting.
That we can rig the p-value to be arbitrarily small by counting in different units makes this problematic. For instance, if the difference between the two groups were less dramatic than the car/boat example, we might have to express in cents. If cents is not enough, maybe we would express the units in fractions of a cent.
