Probit Scenario and Graph

[diggetybo]

Hello,

http://www.eviews.com/help/helpintro.ht ... 038.2.html

Under the demonstration section of this link is an example of using a model to evaluate the probability of y given certain scenarios. My question is when it starts talking about using @trend to fill in gpa_1 with values spanning 2 to 4, and since gpa_1 is overriding gpa, does this still retain the dynamics of the original model? Or is this done for graphing purposes only, to show where the original probit probabilities fall on a nice and straight @trend type of series?

Also what is a good way to view the probability of y=1 given a unit increase in x? (other than looking it up on the graph) Assuming one has correctly evaluated the model under scenarios 1 and 2 and subtracts the two (as in Wooldridge 2006 example 17.1) to find an exact corresponding predicted probability for y once we go from 0 to a 1 unit increase in a given x? I still feel rather unclear about this procedure in eviews.

Thanks for reading

[EViews Glenn]

The @trend fill is because we'd like to evaluate GPA at various values for graphing purposes. It's showing the probabilities evaluated at different values for this explanatory variable.

To answer the second question, you have to decide what you are going to do with the other explanatory variables. Evaluating the model with two different scenarios for the X and differencing will give you what you want. The question is what you do with the other regressors. If you simply solve the model with the other regressors left at their existing values, you'll get changes in the probability for each observation that will differ due to nonlinearity. Alternately, you could set the other X values to their means in both scenarios in which case you'll get a single change in probability.

[diggetybo]

Thank you very much :) It's making much more sense now. How was gpa_2 calculated? Also using @trend, or at the mean?

Last question is, for a continuous series, if I wanted to graph as part of the probability response curve, eviews would need me to use @trend right?

In other words, how do we fill in a series for the general case? @trend of the unit range? something like:

Code: Select all

series x_1 = @min(x)*@trend/(@obs(@trend)-1)

For example, I want to plot Wooldridges 2006 graph from 17.1, the series "educ" ranges from 5 to 17.
:?

[EViews Glenn]

Yes, you want to transform for the high and low. Off of the top of my head [edited to fix a couple of things]

Code: Select all

@min(@trend)*@trend/(@obs(@trend-1))

gets you to 0, 1. Then you need to offset by 5 and linearly interpolate to the end

Code: Select all

5 + 12*@min(@trend)*@trend/(@obs*(@trend-1))

I think that will work, but I haven't had coffee yet...

[diggetybo]

Hey,

I just tried it out. and got an error to the effect of "Insufficient arguments for function..."

Is it because of the @min(@trend) part? I tried changing some parenthesis, but then I got syntax errors, so I guess you were closer than me.

Any ideas what I should change to make it work?

Thanks for your continued support

[EViews Glenn]

Just did a quick test

Code: Select all

series x = 5+12*(@trend-@min(@trend))/(@obs(@trend)-1)

works. Couple of minor revisions from above. I didn't do the linearity adjustment correctly...

[diggetybo]

Great! Thanks for all the help.