Marginal effects in Tobit Model

[Girl From Ipanema]

Dear Glenn,
I have used the code you've indicated, following the steps and omitting the corrections for the expersq (as I don't have non-linear terms). I assume that I can use the results I have obtained, only for the continuous regressors, in the following way: R12=0.28 means that for each additional unit of variable, my dependent increases for 28%.
Next, I would like to get the marginal effects for dummies, and theoretically I would set all of the variables at their sample average and run two regressions: one for D=0 and other D=1, and see the difference in dependent. I have tried your suggestions regarding saving the eq as model object, but it doesn't work. Actually I know how I should get this in theory, but the practice is lagging behind Can you help?

[Girl From Ipanema]

Is it too pragmatical if I just deduce the first outcome from the second one:
FIRST
equation eq1.tobit...
eq1.fit(i) xb
xb=xb-dummy_of_interest*eq1.c(...)
scalar meanxb=@mean(xb)+@dummy_of_interest

SECOND
equation eq1.tobit...
eq1.fit(i) xb
xb=xb-dummy_of_interest*eq1.c(...)
scalar meanxb=@mean(xb)

[Girl From Ipanema]

[quote="EViews Glenn"]

If I were you, I would save my equation into a model object.

[Girl From Ipanema]

Hi Glenn,

since no news from you I am trying to copy my equations into the Model Object in order to get the marginal effects via scenarios, but when copying the equation, model returns an error: @IDENTITY is not a Genr or series expression function in EQ1. How to deal with this?

[EViews Glenn]

Is your EViews up-to-date?

As to the dummy variables, it really depends on how you want to handle them. Some people are happy with doing mean adjustments, others want to evaluate at values of 0 and values of 1 and then compare...

[Girl From Ipanema]

Hi Glenn, working with model object involves some hussle, so I have tried to recalculate my equations with dummy variables once equal to zero and once equal to one (one at a time, others held @Actual).Due to near singular matrix, I will have to work with model scenarios after all. If I name Scenario 1 and override one of my dummy variables, where exactly do I command Eviews to alter this dummy to be equal to one? It is just overridden for the time of being.

[EViews Glenn]

You should look at the documentation on the model object for more on this.

Change the values of the override series itself. What the override allows you to do is to change the scenario specific exogenous without touching the original data. So what you could do is to create two scenarios. One, say "_1" where you override the dummy, say "DUM_1" to be zero and another, say "_2" where you override the dummy, say "DUM_2" to be one.

[Girl From Ipanema]

Hi Glenn, I have been studying the manual on this subject for a few days now, and the scenarios just do not turn out right. It returns NA for the whole sample for my endogenous and the message is 'Solve terminated - Unable to compute due to missing data in ...equation'. What I did is I have created in the wf the dummy1_1 containing values equal to 1 and dummy1_2 containing values equal to 0. I have scenario 1 where dummy1 is overridden and Scenario 2 where dummy1 is overridden. Additionally, I am confused with the fact that the Scenario returns the series of the endogenous as large as the sample (in my case NA unfortunately), whereas I only want to know what is the expected value of my endogenous (1 number) in the Scenarios.
Thanks!

[EViews Glenn]

The missings probably means that there are missings in the exogenous variables for your sample. If you uncheck the Stop on missings checkbox on the solver tab of the dialog you'll see which ones it can solve for and which ones it can't.

What you are getting for solutions are values for each of the observations in the workfile using the data for that observation. If you want a single value, you can override all of your RHS variables, replacing them with means or whatever else you desire. Then solve for a single period (or you can just solve for all and look at the common value).

[Girl From Ipanema]

Dear Glenn, I am afraid that unchecking the Stop on missings got us nowhere. It returns NA for the whole sample for the exogenous. Moreover, you say that if one single value for exogenous is needed, I can play with endogenous, get common value for exogenous etc. However, the common value of NA is NA, so no use of playing. A very concrete question: how can I get marginal effects of my dummy variables in EViews?

I want my other exogenous to be actual, and the dummy of interest should be once 1, once 0. What I did so far is:
1) created dummy1_1 and dummy1_2 in the workfile, because nowhere in the model object you can find a possibility to claim with which values the overridden dummy should be filled in
2) filling them with values 0 and 1 respectively,
3) created a model object, filling it with the equation I am observing,
4) created two scenarios in which dummy1 is overridden. I assume that Sc1 should then use the dummy1_1 from the workfile and Sc2 should use the dummy1_2 from the workfile.

It just does not function the way it has been explained so far.
Thank you!

[EViews Glenn]

I really can't say what's going on with your model, but you may want to take a look at the following model where everything works as described...

[Girl From Ipanema]

Thanks, I will try to figure out what is wrong with my settings