SHAZAM Community
Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

Scaling issue in Probit analysis

A question that I am asking every where. If anyone has an opinion please suggest: Here it is: I have a problem with Probit analysis. I am supposed to calculate and compare the marginal effects across the variables. I know it is not possible to compare nested models because as you introduce more variables the variance is changed while the error variance is not independently identified and is fixed at a given value. For that purpose, I know the solution. But now the problem is that since every variable is in different scale (e.g. Income in 1000s, age in 10s, etc). In this way, the marginal effects reflect different magnitude. If I change the scale of income to 0 to 1 then the marginal probability is changed. In this scenario, how can I compare the marginal effect of variables to determine which variable has the highest and which has the lowest impact? One way would be to change all the variables to 0 to 1. But I cant find any reference to anyone doing such transformation. And then how would I interpret the marginal change based on the transformed variable?

Scaling issue in Probit analysis

A question that I am asking every where. If anyone has an opinion please suggest: Here it is: I have a problem with Probit analysis. I am supposed to calculate and compare the marginal effects across the variables. I know it is not possible to compare nested models because as you introduce more variables the variance is changed while the error variance is not independently identified and is fixed at a given value. For that purpose, I know the solution. But now the problem is that since every variable is in different scale (e.g. Income in 1000s, age in 10s, etc). In this way, the marginal effects reflect different magnitude. If I change the scale of income to 0 to 1 then the marginal probability is changed. In this scenario, how can I compare the marginal effect of variables to determine which variable has the highest and which has the lowest impact? One way would be to change all the variables to 0 to 1. But I cant find any reference to anyone doing such transformation. And then how would I interpret the marginal change based on the transformed variable?