Forum: help

Monitor Forum | Start New Thread

Nested Flat Threaded Ultimate

Show 25 Show 50 Show 75 Show 100

RE: Marginal effects for the joint probability [ Reply ] By: Arne Henningsen on 2015-05-16 05:50

[forum:42246]

Hi H. You are right: your command "margEff( EQ1, addMean = TRUE, cond=TRUE )" returns the marginal effects of the X's on the conditional probabilities, i.e. d Pr( y1 = 1 | y(-1)i, xi ) / d xi, d Pr( y2 = 1 | y(-2)i, xi ) / d xi, ..., d Pr( yn = 1 | y(-n)i, xi ) / d xi for i = 1, ..., N, where n is the number of dependent variables, N is the number of observations in the data set specified by argument "data", xi is the vector of (observed) explanatory variables at observation i in the data set specified by argument "data", and y(-j)i is the vector of (observed) dependent variables at observation i in the data set specified by argument "data" with the j's dependent variable removed. The last point basically answers your second question: when calculating the marginal effects on the conditional probabilities, the values of the other dependent variables are specified by argument "data". By default, argument "data" is the data set that you used for the estimation. Alternatively, you can create a data set with any values of the explanatory variables and any values of the dependent variables, at which you want to calculate the marginal effects. Best regards, Arne

RE: Marginal effects for the joint probability [ Reply ] By: H Jackman on 2015-05-15 12:28

[forum:42245]

Hi Arne Thanks for the response. You understood me correctly. I wanted the marginal effect of a variable X on the joint probability. Also I am not that techy to even begin to figure out how to implement the calculation of joint marginal effects - I can barely use R! A follow up question: I estimated a multivariate probit model with four binary dependent variables on a set of independent variables (X) and call it EQ1 I then use the command COND <- margEff( EQ1, addMean = TRUE, cond=TRUE ) after estimation to get the (marginal) impact of the X's on the conditional probabilities. Is this correct? Or is more needed? Also, does the first set of estimates give the impact of the X's on the conditional effects (y1=1|y2=1, y3=1, y3=1), the second set (y2=1|y1=1, y3=1, y4=1) etc? Or is the default, 0's? Thanks!

RE: Marginal effects for the joint probability [ Reply ] By: Arne Henningsen on 2015-05-13 21:37

[forum:42241]

Dear H Do I understand correctly that you want to calculate the marginal effect of an explanatory variable, say xi, on the joint probability pr( y1=1, ... , yn=1 ), i.e. d pr( y1=1, ... , yn=1 ) / d xi ? Unfortunately, this is not (yet) implemented in the mvProbit package. Sorry! In case of two explanatory variables (i.e. the bivariate probit model), one can decompose the joint probability pr( y1=1, y2=1 ) to a product of a conditional probability and an unconditional probability: pr( y1=1, y2=1 ) = pr( y1=1 | y2 = 1 ) * pr( y2=1 ) or pr( y1=1, y2=1 ) = pr( y2=1 | y1 = 1 ) * pr( y1=1 ) In this case, the marginal effect can be decomposed to a combination of marginal effects on unconditional probabilities, marginal effects on conditional probabilities, unconditional probabilities, and conditional probabilities: d pr( y1=1, y2=1 ) / d xi = pr( y1=1 | y2 = 1 ) * ( d pr( y2=1 ) / d xi ) + ( d pr( y1=1 | y2 = 1 ) / d xi ) * pr( y2=1 ) or d pr( y1=1, y2=1 ) / d xi = pr( y2=1 | y1 = 1 ) * ( d pr( y1=1 ) / d xi ) + ( d pr( y2=1 | y1 = 1 ) / d xi ) * pr( y1=1 ) You could calculate these marginal effects based on the probabilities and marginal effects obtained by the mvProbit package. In case of more than two dependent variables, it seems to me that this decomposition is either very cumbersome or not even possible. If you would like to implement the calculation on joint marginal effects in the mvProbit package, I could assist you. Best regards, Arne

Marginal effects for the joint probability [ Reply ] By: H Jackman on 2015-03-25 19:36

[forum:42097]

Hi I have a quick question about the mvProbit package After estimating my model labelled as EQ1, I was able to get the unconditional marginal effects using ME1<-margEff(EQ1) I know that one can the conditional probabilities like Pr(y1=1|y2=1, y3=1, y3=4) using the package. Is there a way to calculate the average marginal effects for the joint probability that all dependent variables = 1 that is pr(y1=1, y2=1, y3=1, y4=1). I know that the biprobit in Stata can provide marginal effects for the bivariate probabilities . However, I have a model with 4 dependent variables and was just wondering if this was a possibility with your package.