logit model

[anwesha]

hi,
i am using a logit model for estimation. my dependent variable is "history of suffering", where yes=0, no=1.
my independent variables are medical expenditure, averting expenditure, average educational attainment and environmental quality.
after i run the regression i am getting the output where its showing that obs with dep=1 is 113 and obs with dep=0 is 974. but in my data set i have 113 obs with dep=0 and 974 obs with dep=1. so i want to know why is this happening?
and i also want to calculate the marginal effects. how do i do it?
i'm using e-views 3.

regards
anwesha

[hetero]

Hi,

Regarding your first question check carefully your data and try to re-import them.
For the estimation of the marginal effects read the following forum discussion:

http://forums.eviews.com/viewtopic.php? ... ects#p2189

cya

[anwesha]

thanks for your reply. i'll try to do the marginal effect analysis. but as far as the data set is concerned i've imported them correctly. i dont know why they are showing the opposite numbers.

anwesha

[hetero]

Hello,

I have no answer regarding the frequency of your variables. There is a straightforward example in the help function of E-views 5 where it is really easy to estimate the marginal effects. (i do not know if you have access to that version)! Be careful of the probability density function you are going to use ! Use the logistic and not the normal.

cya

[anwesha]

hi
i just read the user manual and i am furnishing the process of computing marginal effect as i understand it.

1. my model is like this
suffering = c + medicalexp + avertingexp+ avgeducation + quality
2. i run logit model and obtain the estimates
3. i go the forecast tab. select index wher prob =(1-F), forecast name sufferingF
4. generate the series sufferingF
5. then multiple this series with the coefficients of independent variables to get the marginal effect.
I want to know if this is the right procedure. And if this is correct then someone please tell me how do i do the 5th step.
The manual says that i have to do somethig like this
@dlogistic(-sufferingF)*(coefficient)
I donot know how to do this.

Thanks
Anwesha :)

[hetero]

Hello,

Usually we estimate marginal effects for a “typical” individual. Therefore, given that all your independent variables are continuous, calculate the mean value for the sufferingF series.

scalar xb= @mean(sufferingF)

then estimate the following scalar

scalar l_xb = @dlogistic(-xb)

And you are done! By multiplying the scalar l_xb to each single coefficient you receive the marginal effect! But now you need to find the associated standard errors since marginal effects are random variables!

Cya

[anwesha]

hi
i just calculated marginal effect using the following steps
1. estimate the logit model suffering= c+ medexp+ avertingexp+ edu, obtain logit coefficients.
2. go to forecats tab, select index, generate series sufferingF
3. generate series scalar xb=@mean(sufferingF)
4. estimate the scalar scalar l_xb=@dlogistic (-xb)
5. multiply the series l_xb with the coefficients of independent variables.

I want to know if this is the right procedure.
i looked into the user guied and it seems that they want to forecast the fitted index (i guess sufferingF series) first and then multiply the auto series @dlogistic(-xb) with the coefficients. i didn't get this point. please healp.

anwesha

[hetero]

hi,

In step 4 you estimate l_xb which is a scalar ( a single value – not a series ).

In stap 5 multiply l_xb (scalar) to each single coefficient to get the marginal effect (l_xb*coefficient)

regards

[yanihm]

hi there hetero,
first of all thanks so much for your advice to the other user. as a new user, i have benefited from them too! just wondering, how do you calculate the standard errors for the marginal effects?
thanks in advance


yanihm

Hello,

Usually we estimate marginal effects for a “typical” individual. Therefore, given that all your independent variables are continuous, calculate the mean value for the sufferingF series.

scalar xb= @mean(sufferingF)

then estimate the following scalar

scalar l_xb = @dlogistic(-xb)

And you are done! By multiplying the scalar l_xb to each single coefficient you receive the marginal effect! But now you need to find the associated standard errors since marginal effects are random variables!

Cya

[hetero]

In a previous post I wrote the following:

For those who might be interested in estimating standard errors for the associated marginal effects take a look at the two following papers!

Anderson, S. and Newell, R. (2003). Simplified marginal effects in discrete choice models. Economics Letters, 81. 321-326.
Carlevaro, F. and Senegas, M. (2006). Simplified marginal effects in discrete choice models: A correction. Economics Letters, 92. 44-46.

Anderson and Newell (2003) have shown that after a simple normalization applied to all the independent variables the computation of the standard errors simplifies dramatically!
A good idea also is to read fist the delta method in Green’s book.
Let me know if you do not have access upon the pre-mentioned papers

Regards