Creating dummy variables

[Brian Conway]

I'm working with data on countries, running separate country models. I have one variable that is coded as 0, 1, 2 and 3 and wish to create dummy variables from it. As one country has values of 0, 1 and 2 (i.e., there is no 3) on this variable, I created three dummy variables with one as the reference category. For a different country which had values of 0 and 1 on this variable, I created two dummy variables with one as the reference category. I was wondering if this is correct? Thank you.

[Andrew Musau]

You likely don't need to create the dummies as most estimation commands nowadays support factor variables. See

help fvvarlist

You can specify the same base across regressions using

ib#.catvar

where you replace # with the specific level and "catvar" with the name of the categorical variable.

[Brian Conway]

Thank you Andrew. Even so, does the way I created the dummy variables make sense? Many thanks.

[Nick Cox]

At most you need one indicator variable (dummy in your terminology) fewer than the number of distinct values. So, one if there are two distinct values, two if there are three, so forth.

But Andrew Musau is right. It's most unlikely that you need to generate any extra variables at all. Also, if you are running separate models for each country, you should still use the same syntax as far as possible.

[NANA YAW POKU]

gen dummy_med_tna_sum=0
replace dummy_med_tna_sum=1 if tna_sum>=dummy_med_tna_sum

please is the above command correct in generating a variable

[Andrew Musau]

gen dummy_med_tna_sum=0
replace dummy_med_tna_sum=1 if tna_sum>=dummy_med_tna_sum

please is the above command correct in generating a variable

One needs to know the context to say whether or not the code does what it is supposed to do. You can simplify the code to:

gen dummy_med_tna_sum = tna_sum>0 & !missing(tna_sum)

This will assign a value of 1 to the generated variable if and only if "tna_sum" is positive, 0 otherwise.

[Brian Conway]

I created the dummies using the commands below, which I think does the same thing. These are the commands I'm used to using for dummy variables (though slightly less neater than your ones).

generate var=.
*I then copied the values of this variable (coded as 0 or 1) into the data editor
generate var1=0
replace var1 = 1 if var == 0
generate var2 = 0
replace var2 = 1 if var == 1

tab var1
tab var2

I ran the frequencies on the dummies just to check and everything looked fine.

Thank you very much.