Urgent help on Ordered Probit Model

[michelleli]

Hi,
I have estimated an ordered probit model in Eviews 6 and trying the calculate the marginal effect. My procedue to calculate the marginal effect is as follows:
1.Make model from the Proc function
2.press "solve"
The output page indicates "solve terminated-unable to compute due to missing data in ....."

I have successfully estimated the ordered probit model and have restricted the "solution sample" to workfile sample used in the the orderd probit model. How come Eviews unable to compute due to missing data?
Workfile attached and I would like to estimate marginal effect for ``panel probit" .
Please help!

Cheers
Michelle

[michelleli]

Urgent help needed!

[EViews Glenn]

There are missing values in your estimation sample. EViews automatically excludes them in estimation...

For model solution, missings are not automatically excluded. You'll need to go to the Solver tab of the model solution dialog, and uncheck the "Stop solve on missing data" option...

[michelleli]

There are missing values in your estimation sample. EViews automatically excludes them in estimation...

For model solution, missings are not automatically excluded. You'll need to go to the Solver tab of the model solution dialog, and uncheck the "Stop solve on missing data" option...

[michelleli]

Problem Solved ! Thanks a lot!

[michelleli]

Thanks for your reply, Glenn. I solved the the problem using the tip you provided ablve. However, another problem occured. After I "solved"the model, Eviews 6 only produced the probability of oberserved valued in the dependent variable for each catergory with respect to the latent variable, and it didn't "solve" the estimated limite points for my sample. Without the latter, I couldn't calculate the marginal effect. Can you please help? Much appreciated! Eviews file attached!

There are missing values in your estimation sample. EViews automatically excludes them in estimation...

For model solution, missings are not automatically excluded. You'll need to go to the Solver tab of the model solution dialog, and uncheck the "Stop solve on missing data" option...

[EViews Glenn]

I'm not certain that I understand your question. The limit points are coefficients.

[michelleli]

As explained by trubador in http://forums.eviews.com/viewtopic.php?f=4&t=337 OR
(Since your dependent variable has three distinct categories, you might have estimated two threshold/limit values. The latent variable (or index function), however, can only be constructed after the estimation of coefficients of your explanatory variables (i.e. the linear combination). You can either do this by yourself or choose Make Model option under the Proc menu in the equation dialog box and solve it. If you do the latter, you will see that probability of observed values in your dependent variable for each category is computed with respect to this latent variable as well as the estimated limit points. Therefore, change in any one of the variables will lead to a change in related probabilities. Marginal effects are simply the derivative of these probabilities with respect to explanatory variables and can easily be calculated via chain rule. Suppose that the name of your calculated latent variable and estimated limit values are latent, limit_0 and limit_1, respectively. ), the "solve" function should produce the probability of oberserved value for each eategory with respect to the latent variable as well as the estimated limited points. However, when I use Eviews 6 to "solve", it only produces the probability with respect to the latent variable, Not estimated limit points. I wonder why!

I'm not certain that I understand your question. The limit points are coefficients.

[EViews Glenn]

I think I understand what you are asking.

The ordered probit (or any of the other ordered specifications) evaluates a cdf at a set of limits for each realization of X. If, for example, you have three categories, you'll have two limit points which are coefficients, say C(1) and C(2). Then letting XB be the evaluated linear index in your variables (not including the limit point coefficients) you'll have the evaluation points:

c(1) + xb
c(2) + xb

in the three category model. The ordered probit evaluates the category probabilities as

@cnorm(c(1) - xb)
@cnorm(c(2) - xb) - @cnorm(c(1) - xb)
1 - @cnorm(c(2) - xb)

When you make a model, EViews makes variables of the following (pseudo) form:

i_var = xb
p1 = @cnorm(c(1) - i_var)
p2 = @cnorm(c(2) - i_var) + @cnorm(c(1) - i_var)
p3 = 1 - @cnorm(c(2) - i_var)

where "var" is the name of the dependent variable. You can look at the equations in the model to verify this. Note that I_VAR is just the XB part of the specification.

Having written this out, I now think I understand your question: you are asking why EViews doesn't produce model solutions to:

c(1) - i_var
c(2) - i_var
c(3) - i_var

which are the the actual index points used in evaluation (instead of the I_VAR). The reason is simply that the EViews model representation uses the minimum number of variables required to compute the probabilities in the categories; since I_VAR is common to all categories the limit series only differ by a constant value. Thus if you had 30 categories, and solved the model you would have 29 series in the workfile that differ only by a constant value. EViews only has the one I_VAR, and the probabilities are computed using this series with the appropriate coefficient offset.

If you want the actual indices used in the probability evaluation, you simply have to offset the I_VAR by the coefficients (in this case c(1) and c(2)). Ideally, we'd provide an option to Make Model which allowed you to make a model with just the limit points (with the constants baked-in). This would be analogous to what we do in the single equation forecast.

No promises, but I'll make a note of this.