Converting Variables to a dummy variable

[Salma Zayed]

Hello, Actually I have three variables in percentage (Formal credit - semi-formal credit - Informal credit) between 0 -100%

sum Formal_credit informal_credit Semi_formal_credit

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
Formal_credit |     81,342    15.10992    25.61789          0        100
informal_credit |     82,974    62.47527    34.42988          0        100
Semi_formal_credit |     78,439    1.368165    7.610706          0        100

Now, I am trying to create a categorical variable (e.g. Formal credit = 1, informal credit = 2, semi-formal credit = 3), but the problem, a firm can use formal and informal credit in the same time. my question How I avoid this problem?

Thanks a lot

[Maarten Buis]

The categorical variable you want to create assumes that these are mutually exclusive categories, and that does not seem to be the case. So that is that: you cannot create the variable you want. You can create three indicator variables: One for formal credit or not, one for informal credit or not, and one for semi-formal credit or not, and firms can score 1s on multiple variables. You could create a categorical variable with more categories: only formal, formal and informal, formal and semi-formal, etc.

[Salma Zayed]

Thanks for your replying, so what I did;

generate formal_dummy = .
replace formal_dummy = 0 if Formal_credit == 0
replace formal_dummy = 1 if Formal_credit > 0 

generate informal_dummy = .
replace informal_dummy = 0 if informal_credit == 0
replace informal_dummy = 1 if informal_credit > 0

generate Semi_formal_dummy = .
replace Semi_formal_dummy = 1 if Semi_formal_credit  > 0
replace Semi_formal_dummy = 0 if Semi_formal_credit  == 0

So, I got These results:

sum formal_dummy informal_dummy Semi_formal_dummy

  Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
formal_dummy |     81,342    .3695016    .4826728          0          1
informal_d~y |     82,974    .9377998    .2415203          0          1
Semi_forma~y |     78,439    .0534173    .2248656          0          1

But for that case, I have the percentage of formal = 36% , informal = 93%, semi formal = 5% That higher to 100%.
Is that right?

[Nick Cox]

You need to look carefully at missings too. With your definitions, each indicator (you say dummy) will be

1 if the value is positive or missing

0 if the value is zero

missing is the value is negative.

Is that what you want?

[Maarten Buis]

The problem Nick pointed out is that a missing value in Stata is the highest possible number. So if Formal_credit is missing on an observation, then Formal_credit > 0 will evaluate to true for that observation, which is probably not what you want. One way of dealing with Nick's comment is:

generate formal_dummy = .
replace formal_dummy = 0 if Formal_credit == 0 r
eplace formal_dummy = 1 if Formal_credit > 0 & !missing(Formal_credit)

generate informal_dummy = .
replace informal_dummy = 0 if informal_credit == 0
replace informal_dummy = 1 if informal_credit > 0 & !missing(informal_credit)  

generate Semi_formal_dummy = .
replace Semi_formal_dummy = 1 if Semi_formal_credit  > 0 & !missing(Semi_formal_credit)
replace Semi_formal_dummy = 0 if Semi_formal_credit  == 0

The missing() function returns a 1 (true) when its argument is a missing value and a 0 (false) otherwise. The ! negates, so true (1) becomes false (0) and false (0) becomes true (1). So in my mind I read !missing(Formal_credit) as "not missing on Formal_credit"

After you have created the indicator variables correctly, the proportions will in all likelihood still not add up to 1. They would have added up to 1 if they were mutually exclusive, be we established in #1 that that is not the case. That is not a problem, this is why you created these variables this way.