Probit model issue with same variable (blue lines and black lines data editor)

[Mina Sami]

I got a database on: remittances, households, education (Survey).

My problematic question is: Effect of remittances on education decision (1 and 0).

Note: I have 4084 hh that received remittances :

My command is:
probit education remittances.
I want to note that in my data “remittances variable” is in blue line. When I run the command it works well with positive sign.

Otherwise,
I created a dummy
gen remit=1 if remittances==1
replace remit=0 if remittances==.

When I run my command on this dummy: its work but with a negative sign. (stata run on all data the regression 47410

Why do remit and remittances do not have the same effect although they are both the same variable. What is the difference between this data and the the new variable “shouldn't it be the same”?
Here is the data form:

Results with variable in the data itself (number of obs 4084)

Here is my results: (note stata run on 47410 obs)

[Carlo Lazzaro]

Mina:
if I got your query right, you have imposed that missing values for remittance are=0 (i.e., no remittance).
Hence, with -remit- you worked-around the listwise deleteion that Stata made by default when you included -remittance- among predictors.
In summary, the difference in the number of observations derive from your plugging 0 in -remit- instead of considering the existence of missing values (like in -remittance-).

[Mina Sami]

Thank you for your help,
but when i tried now to replace remit with . rather than 0 and re-run the regression, STATA omit the variable directly.
Which is logic as there will no variation in the number within the variable.

So, If stata is working on remittances (4058 variable), he cannot estimate its effect as all the observations will be equals to one and no zero ones (missed obs will not be included in the model).

So how do Stata in remittances calculate its coefficient although it drops other observations (missed).
remittances everywhere is always equals to 1 !
How can we find a coefficient?

[Carlo Lazzaro]

Mina:
if you have no 0s in -remittance- (but only 1 and missing values), Stata omits the variable (probably due to collinearity).
Besides, pretending that what is missing equals "no remittance" (i.e., 0) makes your inference seriously biased and the coefficient you get for -remittance- does not make any (informative) sense.