marginal effects probit

[bibimiwi]

Good afternoon,

I have already looked at the other topics concerning probit analysis but none could answer my questions.

I want to calculate the marginal effects of a probit model with one binary dependent variabel and one explanatory variable for the start.

I did the following:

1) click on "Proc", Forecast, "Index - where Prob=1-F(-Index)", Ok
2) "Proc", "Make Residual Series", "Name for Resid series", resid01

Then I get 77 different values. I do know that I have to multiply my coefficient with @dnorm(-resid01), but I do not know how to do this as I have only 77 different values for resid 01 but 98 different values for my explanatory variable. Or do I have to multiply my one and only coefficient by all those 77 values? How is this supposed to be calculated?
Do I do this calculations somewhere in EViews, or in a seperate Excel sheet?

As I only began to work with EViews, I would be very greatful for answers as simple as possible!

Thank you very much

[bibimiwi]

Can nobody help me with my problem?

[EViews Glenn]

To evaluate at the mean values of the data:

1. click on "Proc", Forecast, "Index - where Prob=1-F(-Index)", Ok, say you save this to xb
2. multiply the coefficient on the variable of interest by by @dnorm(-@mean(xb))

To evaluate one for each data value:

1. p. 261 of EV8 User's Guide II

To evaluate somewhere else:

Do 2. above, replacing @mean(xb) with your favorite value

[bibimiwi]

Thank you very much for your reply!
But I still do have some difficulties..

To evaluate at the mean values of the data I did exactly as you said but as output I only get a graph. And when I delete the cross in "Output" "Forecast graph" I only get my original estimation of the equation. What am I doing wrong?

You told me to multiply the coefficient of the variable of interest by @dnorm(-@mean(xb)). WHERE and how exactly do I do this calculation? Do I have to estimate a completely new equation (Import the data, Quick --> Estimate Equation etc.)?

[EViews Glenn]

In the forecast dialog, you'll save the index results into a series, call it, say "XBF".
Then the command

Code: Select all

vector meff = @dnorm(-@mean(xbf))*eq1.@coefs

will compute the set of marginal effects, one for each coefficient.

[larrydavid]

The calculated marginal effects, are they accurate for binary variables, i.e. if it takes the value 1 then the marginal effect is as reported?

[EViews Glenn]

In that case they are the continuous approximations. To get the discrete ones you'll need a different form of the computation.

Evaluate the mean XB with your dummy set to 0 and then the XB with your dummy set to 1 (an easy way to do this is to take the mean XB you've already computed and subtract the dummy coefficient multiplied the mean of the dummy (that will give you the mean of the XB with the dummy set to 0). Call this meanxb0. Evaluate the probability using the @cnorm(meanxb0).

Then take the mean of the XB you've already computed an add the coefficient multiplied by 1-the mean of the dummy (that will give you the mean of the XB with the dummy set to 1). Call this meanxb1. Evaluate the probability using @cnorm(meanxb1).

The difference between the two computed probabilities gives the discrete marginal effect at means, but with the dummies evaluated at 0 and 1.

You'll need to do this for each of your dummy variables.

Something like,

Code: Select all

equation eq1.fit(i) xb scalar meanxb = @mean(xb) scalar meandum1 = @mean(dummy1) scalar meanxb0 = meanxb - eq1.c(2)*meandum1 scalar meanxb1 = meanxb + eq1.c(2)*(1 - meandum1) scalar meffect1 = @cnorm(meanxb1) - @cnorm(meanxb0)

where the second variable in the equation is your dummy variable (hence the use of C(2) --- modify as necessary).