Forum: help

RE: Marginal effects for multivariate probit with different regressors [ Reply ] By: Jacquelyn Pless on 2015-02-01 20:21	[forum:41861]
Thank you, Arne. I was able to implement this, however it seems as though the output isn't what I am seeking. I am wondering if there is an implementation for calculating marginal effects (rather than predicted values) as such: http://www.aae.wisc.edu/aae637/articles/greene_marginal_effects_bivariate_probit.pdf. This output provides the marginal effects of the covariates. Best, Jacquelyn

RE: Marginal effects for multivariate probit with different regressors [ Reply ] By: Arne Henningsen on 2015-01-31 22:33	[forum:41860]
Dear Jacquelyn Are you using the latest version of the sampleSelection package (currently 1.0-2)? http://cran.r-project.org/web/packages/sampleSelection/index.html http://www.rdocumentation.org/packages/sampleSelection/functions/predict.selection Best regards, Arne

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Jacquelyn Pless on 2015-01-26 16:33

[forum:41844]

Hi Arne,

Thank you - yes, there are the two different marginal effects for the outcome equation, and I believe they are calculated in a similar way as regular selection models, except it is a bit more complex when the outcome equation is also a probit (rather than OLS in a standard selection model).

As such, when I try using predict() we we normally would, I am getting the following error:

Error in UseMethod("predict") :
no applicable method for 'predict' applied to an object of class "c('selection', 'maxLik', 'maxim', 'list')"

My code is:

A=selection(adoptions~decroi+decsavings+pretirement+padvertisement+pdirectmarketing+hincome+politics,buylease~costperception+easeperception+contractperception+sellperception+othersperception+decroi+decsavings+pretirement+padvertisement+pdirectmarketing+hincome+imphomevalue+X2007+X2008+X2009+X2010+X2011+X2012,data=combinedsolar,method="binaryOutcome")
summary(A)
margeffects=predict(A,type="conditional")

Do you have any thoughts?

Best,
Jacquelyn

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Arne Henningsen on 2015-01-25 20:32

[forum:41839]

Hi Jacquelyn

You can calculate the marginal effects of the selection equation in the same way as you do for a standard probit model. I think that there are two different marginal effects of the outcome equation: (a) the marginal effect on the unconditional expectations and (b) the marginal effects of the conditional expectations (see, e.g. the documentation of predict.selection()).

Best regards,
Arne

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Jacquelyn Pless on 2015-01-25 19:16

[forum:41835]

Arne,

Thank you - and I'm sorry for not being more explicit upfront.

I am currently estimating a model with selection with the sampleSelection package. I am using the selection() function and specifying my method as binaryOutcome since the dependent variable is binary for both my selection and outcome equations.

I came across mvProbit once I was running into trouble figuring out how to calculate the marginal effects from the selection model. I mostly just became curious about mvProbit and its functionalities, but I'm actually using sampleSelection right now.

So - alternatively, my real question might be: is there a mechanism for calculating the marginal effects for models specified with the selection() function?

Best,
Jacquelyn

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Arne Henningsen on 2015-01-25 18:46

[forum:41834]

Dear Jacquelyn

I am not sure which model you want to estimate.

Do you want to estimate a multivariate probit model using function mvProbit() of the "mvProbit" package? This function/package does not account for sample selection.

Or do you want to estimate a probit model with sample selection (truncation; not censoring!) assuming that the error term of the selection equation and the error term of the outcome equation follow a bivariate normal distribution? In this case, you can use function selection() of the "sampleSelection" package, which allows for different explanatory variables in the selection equation and in the outcome equation.

Best wishes,
Arne

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Jacquelyn Pless on 2015-01-25 16:46

[forum:41832]

Arne,

Thank you for your helpful response.

I understand your point. In some cases however, there may be explanatory variables that are only relevant for the outcome equation (or in fact, only observed for those who aren't censored in the selection equation), and in these cases, an exclusion restriction is necessary in order for a bivariate probit to be identified: http://econ.sites.olt.ubc.ca/files/2013/12/pdf_paper_seminar_sukjin_han.pdf.
Admittedly, I am not the best code developer, but I may consider attempting to implement this feature. Are you able to provide me with the original code for mvProbit for me to work from?

Best,
Jacquelyn

RE: Marginal effects for multivariate probit with different regressors [ Reply ]
By: Arne Henningsen on 2015-01-25 13:00

[forum:41830]

Dear Jacquelyn

I do not plan to implement different regressors for each equation of an mvProbit() model in the near future, because in many cases it is reasonable to assume that an explanatory variable that (directly) affects one dependent variable, also (at least indirectly) affects the other dependent variables (see, e.g., http://econpapers.repec.org/RePEc:foi:wpaper:2012_11). However, if a user (e.g. you) wants to implement this feature, I would be happy to support him/her.

Best regards,
Arne

Marginal effects for multivariate probit with different regressors [ Reply ] By: Jacquelyn Pless on 2015-01-21 22:50	[forum:41820]
Hi everyone, I am trying to calculate the marginal effects for a multivariate probit but one in which I use different regressors for each equation. The error from mvProbit currently notes that using different regressors has not been implemented yet - is there a plan to do so, or do you have suggestions on how I might go about doing this? Thank you in advance!