How is 'iweight' treated by the tabulate two-way command?

[Rosa Blau]

Hi,

I'm working with the American Community Survey (ACS), a dataset with 3mio + observations. I often have to tabulate two variables I've created, namely naics2017 and skill (both numeric floats). I have a fine-grained version of naics2017, with over 400 unique values, and a broader version, with only 20 unique values. My skill variable always only takes four values. All values are integers, but can be negative.

When using my 20-value naics2017 variable, I can easily use my default ACS svyset and svy commands:

svyset cluster [pweight = perwt], strata(strata)
svy: tab naics2017 skill, row

It takes my computer a full two minutes to process, but that's fine. When I try to do the same thing with my 400+ value naics2017 variable, the calculation never ends and stata eventually just freezes.

I think I've found a decent work-around that enables me to use the same code for both versions of my naics2017 variable. The following code (without the svyset command) works without putting my computer into a coma:

tab naics2017 skill [iweight = perwt], row

But the thing is: I have never used the iweight option before, and I don't quite understand what calculation is behind it if I use it with two-way tabulation in this way. The help command for iweight says that "any command that supports iweights will define exactly how they are treated." However, under help for 'tabulate twoway' I can't find anything about iweights.

Can anybody tell me where to look to learn how tabulate twoway uses iweights? Or can somebody tell me if I am using the iweight option correctly? I am optimistic because using these two commands with my 20-value naics2017 variable gets me the same results as my standard pweight option with the svyset command. But I want to understand the iweight option for the tabulate command, and would be grateful if somebody could point me toward a useful explanation.

The ACS person weight variable btw is perwt:

* Example generated by -dataex-. To install: ssc install dataex
clear
input int perwt
161
128
 71
148
159
 53
131
 85
312
 77
 96
 40
 48
142
  7
250
 59
 73
 89
 58
174
 74
 86
 69
133
 12
 66
 31
 36
 31
227
111
125
226
 46
159
 74
 59
 97
121
 55
198
 17
193
 22
125
 63
147
 94
 54
110
 49
105
122
 45
233
189
103
 44
 50
173
118
 44
 16
174
 79
 60
 18
102
 92
124
195
116
 53
 62
 33
 60
111
113
 39
161
 38
 67
172
133
236
 34
101
137
270
 31
 74
 65
 26
199
 30
 85
 61
163
107
end

[FernandoRios]

Hi Rosa,
a simple rule of thumb that has helped me navigate weights and Svy.
If your task requires just counting (like -tabulate-), or just report means, you do not need using "svy" prefix.
if you are interested in some type of standard errors, (regression analysis for example), then you need to use svy, or pweights.

if what you are interested in is tabulations and shares (not the number of weighted individuals), any weight would do. (except pweights, because those are not allowed)
HTH

[Rosa Blau]

Thanks Fernando, that's a very helpful rule of thumb. By that logic I think I should be fine using iweight in this example.

[Rosa Blau]

FernandoRios, since your last advice was so practical, I am hoping you can help me with another simple question regarding weights. I am trying to figure out how to insert two weights into one simple tabulation. I've posted all the details on the problem in a separate thread here.