Let us look at this line by line.
local var year
You defined a local macro var with content "year". This is legal but you never refer to that local macro in this code, so the definition is pointless.
local j = 1996
You defined a local macro j with content "1996". This is legal.
foreach j of var year {
You open a loop and define the loop index to be j. That means that within the loop any reference to local macro j will be interpreted in terms of the list of arguments you provide. (The previous definition of j is irrelevant within the loop, and so has no effect in the rest of your code.)
... of var year
You specify that the loop is over a variable list here. Note that the keyword var here is short for varlist and has absolutely nothing to do the local macro name var you just defined. The variable list consists of the single variable name year.
gen d`j' = 1 if year==`j'
This statement will be interpreted, the one and only time the loop is executed, as
gen dyear = 1 if year==year
as references to the local macro j are replaced with its contents, the variable name year. year==year is true for every observation. The effect is a new variable dyear which is 1 in every observation. That is not an indicator or dummy variable as you want it. If you look at your dataset carefully, you will see that is not a dummy variable for year being 1996.
local ++j
You are trying to increment the local macro j by 1. But you just set local macro j to contain the string "year", which is a variable name. But you can't add 1 to a string, and so the error message will be type mismatch. You don't report that error, which is a surprise. It is a little subtle, as in the previous command the context of generate allows interpretation of the reference to year as an instruction to calculate with the variable year, which is naturally numeric. But local commands are all about string manipulation, which may or may not have numeric interpretation, and your command is equivalent, first of all, to instructing Stata to add
"year" + 1
which triggers a type mismatch error.
Turning away from your code: Consider a loop
forval y = 1996/2012 {
gen d`y' = 1 if year == `y'
}
This is closer to what you want but makes clearer another bug in your code. This would create variables d1996 to d2012 but each will be 1 in the year specified but missing otherwise, which is not what you want.
You could fix that by adding a further line in the loop
replace d`y' = 0 if year != `y'
but a much cleaner way to do it is the single line
gen d`y' = year == `y'
The expression
year == `y'
is evaluated as 1 when true and 0 when false, which is what you want.
All this is standard technique documented in [U] or [P].
As @Roberto Ferrer pointed out, however, experienced Stata users would not define dummies this way, as tabulate offers an option to do it without a loop.
A tutorial that brings together comments on local macros, foreach and forvalues loops is within http://www.stata-journal.com/sjpdf.html?articlenum=pr0005
search foreach
within Stata would have pointed to that as one of various pieces you can read.
