[Gretl-devel] order of string-value codes
This is somewhat related to recent discussion of "Binary Logit with
string series".
As you may know, when gretl reads string-valued series from (e.g.) a
CSV file, consecutive 1-based numeric codes are attached to the
strings in their order of occurrence. In some cases that's fine, but
if the string values have a "natural" ordering it may not be so
fine.
Consider the following CSV (I'll call it strorder.csv):
income,working
"high","yes"
"low","yes"
"middle","no"
"middle","no"
"low","yes"
"high","yes"
"middle","no"
On importing these data "high" will get code 1, "low" code 2, and so
on. One may want to recode income as low=1, middle=2, high=3, and
also recode working as no=1, yes=2. That's not super difficult in
hansl, but I wonder if it would be worthwhile to provide a built-in
function to do the job: this could be a new function, or perhaps
could be enabled as a variant of strvsort(), via an optional second
argument.
Anyway, here's a hansl prototype for such a function. The idea is
that you pass in the original series along with an array in which
the original string values are ordered as you want them coded, and
you get back a new string-valued series.
<hansl>
set verbose off
function series strorder (series x, strings S)
Sx = strvals(x)
ns = nelem(Sx)
if ns != nelem(S)
funcerr "Replacement strings array is of wrong length"
endif
matrix v = zeros(ns, 1)
matrix chk
loop i=1..ns
chk = instrings(S, Sx[i])
if rows(chk) == 0
funcerr "Bad replacement string"
endif
v[i] = chk[1]
endloop
series ret = replace(x, seq(1,ns), v)
stringify(ret, S)
return ret
end function
open strorder.csv -q
print "string values of income:"
eval strvals(income)
income2 = strorder(income, defarray("low", "middle",
"high"))
print "string values of income2:"
eval strvals(income2)
print income income2 -o # should be identical
print "string values of working:"
eval strvals(working)
working2 = strorder(working, defarray("no", "yes"))
print "string values of working2:"
eval strvals(working2)
print working working2 -o # should be identical
</hansl>
Allin Cottrell