lfe (version 2.5-1998)

chainsubset: Chain subset conditions

Description

Chain subset conditions

Usage

chainsubset(..., out.vars)

Arguments

...

Logical conditions to be chained.

out.vars

character. Variables not in data.frame, only needed if you use variables which are not in the frame. If out.vars is not specified, it is assumed to match all variables starting with a dot ('.').

Value

Expression that can be eval'ed to yield a logical subset mask.

Details

A set of logical conditions are chained, not and'ed. That is, each argument to chainsubset is used as a filter to create a smaller dataset. Each subsequent argument filters further. For independent conditions this will be the same as and'ing them. I.e. chainsubset(x < 0 , y < 0) will yield the same subset as (x < 0) & (y < 0). However, for e.g. aggregate filters like chainsubset(x < mean(x), y < mean(y)) we first find all the observations with x < mean(x), then among these we find the ones with y < mean(y). The mean(y) is now conditional on x < mean(x).

Examples

N <- 10000
dat <- data.frame(y=rnorm(N), x=rnorm(N),id=factor(sample(N/100,N,replace=TRUE)))
# It's not the same as and'ing the conditions:
ss <- chainsubset(x < mean(y), y < 3*mean(x))
sum(eval(ss,dat))
sum(evalq(x < mean(y) & y < 3*mean(x), dat))
ss2 <- chainsubset(x < mean(y), y < a*mean(x), out.vars='a')
a <- 3; sum(eval(ss2, dat))
a <- 2; sum(eval(ss2, dat))
# Among observations with x < y, find entire id's with more than
# one fifth of their x's larger than 1/2
ss3 <- chainsubset( x < y, tapply(x,id,function(.xx) {sum(.xx > 1/2) > length(.xx)/5} )[id])
 sum(eval(ss3,dat))