Forum: help

RE: Probit selection AND outcome [ Reply ] By: Tuyo Ambali on 2016-08-24 09:42	[forum:43469]
Many thanks for the insight, Arne. Kind regards, Tuyo

RE: Probit selection AND outcome [ Reply ]
By: Arne Henningsen on 2016-08-24 08:57

[forum:43465]

Dear Tuyo

In my example, I used "wage>=5" as dependent variable of the outcome equation just because the comparison operator ">=" creates a binary variable and I wanted to demonstrate in the example that the dependent variable of the outcome equation can be binary. In real life, using the binary variable "wage>=5" instead of the continuous variable "wage" does not make much sense.

BTW: Please don't forget to cite in your publications the R packages that you have used for your analysis (see output of the R command 'citation("<packageName>")').

Best regards,
Arne

RE: Probit selection AND outcome [ Reply ]
By: Tuyo Ambali on 2016-08-17 12:16

[forum:43447]

Dear Arne,

I am a current PhD student at the University of Reading, UK and a new R user. I recently encountered your response to Andy's question where you suggested he should include the inequality sign (e.g wage>=5~edu...). See the link please. https://r-forge.r-project.org/forum/forum.php?thread_id=31866&forum_id=844&group_id=256.

Like Andy, I am currently running two-stage model in R where both my selection and outcome variables are binary (0 or 1). I have run my model (ML) without inequality sign but not sure weather I am on the right path.

However, using inequality (like y>=0 or y<=1) didn't give any result, even though I have mvtnorm package installed.

I have also read the paper written by you and Toomet on sample selection estimation in R in 2008 but found that no such assumption was made.
Does tobit2fit treats the outcome variable as binary or continuous?
I am on the the right path?
What should I do?
Thanks
Tuyo

RE: Probit selection AND outcome [ Reply ]
By: Arne Henningsen on 2016-06-08 20:17

[forum:43265]

Hi Andy

The selection() command of the "sampleSelection" package can estimate sample-selection models with a binary dependent variable of the outcome equation, e.g.:

data(Mroz87)

m <- selection( lfp ~ educ + age + kids5 + kids618 + nwifeinc, wage >= 5 ~ educ, data = Mroz87 )

summary(m)

The above command estimates that model by the maximum likelihood method. You can also estimate this model by the 2-step method instead of the maximum likelihood method:

m2 <- selection( lfp ~ educ + age + kids5 + kids618 + nwifeinc, wage >= 5 ~ educ, data = Mroz87, method = "2step" )

summary(m2)

However, I see no good reason for estimating this model by the 2-step method instead of the maximum likelihood method.

Best regards,
Arne

Probit selection AND outcome [ Reply ] By: Andrew Ballard on 2016-06-06 21:00	[forum:43259]
Hi, I would be interested in implementing a probit model for the outcome process in a 2-stage model. Is this possible with a version of this package in R? My initial glance suggests no, but I'm hoping this is not the case! Thanks, Andy