Can anyone recommend a standardized effect size similar to OR for the effect of an IV in probit regression. The OR conversion (multiply coefficient by 1.6 or 1.8 and exponentiate) seems to be discouraged.
Thank you for the thought, Adrian. Unfortunately for this model we have to use WLSMV estimation which gives probit coefficients.
When using the ordered probit model (or any other model, for that matter), marginal effects are used to infer the effect of the variable on the response. Marginal effects are a great tool, as they provide an absolute change in outcome probability due to a one-unit increase in explanatory variable.
I have a couple sources that show this and believe it would help a lot. I definitely recommend and encourage the use of marginal effects, especially for probit and logit models.
-Jason
Hi Jason. Can you recommend a couple of citations for marginal effects. This looks like what I need! From what I'm reading, marginal effects go beyond inference about change in z-scores and allow inference about change in Y?
Hi Jason,
Another question -- is there a program that can compute marginal effect from a probit regression coefficient (I assume the ME is constant for all values of X and other X values in the model?)
Hello Anne,
I would recommend reading the attached article, it explains the usage of marginal effects quite well. Also, I would recommend William Greene's textbook: Econometric Analysis.
These are used in terms of logit, but the same concept applies to probit modeling.
Hope this helps.
Best,
Jason
Thank you, Jason! To share with others, I've learned you can compute marginal effects in Mplus using the Model Constraint command.
Hi Anne,
I have the same situation in Mplus, using WLSMV and want to get marginal effects for the probit coefficients. Are you willing to share example code or other sources on using Model Constraint to compute the marginal effects? I could not find much on statmodel.com
Best,
Lisa
Hi Lisa,
I'm sorry for the delay in responding. I just got this from Bengt Muthen via the online Mplus discussion. I hope it's helpful!
For the probit, the marginal effect is
phi(a + b*x)*b
where phi is the standard normal distribution
phi = exp(-0.5(y-a-b*x)^2))/sqrt(2*pi).
The expression phi(a + b*x)*b can therefore easily be computed in Model Constraint by writing it out as shown here.
I am sharing more information generously provided by Bengt Muthen about the above formula (from the Mplus discussion board, Marginal Effects topic: http://www.statmodel.com/discussion/messages/23/828.html?1510616749)
The "xb" expression in Long's book refers to a vector of x's multiplied by a vector of b slopes (regression coefficients), that is, he writes the regression with multiple x's in matrix algebra terms. His expression includes the intercept term. So with a single x, his xb is the same as my expression a+b*x. The intercept a is the negative of the Mplus threshold and b is the slope. His b_k = b.
Hi Anne, hi Lisa,
Now it is me who would like to calculate average marginal effects from probit coefficients produced by the WLSMV estimation in mplus.
I am happy that I found your postings, incl. Bengt Muthén's explanation and formulas, but I am still not sure how to tell mplus to use the thresholds for the categorical outcomes needed in the equations [phi(a + b*x)*b] as well as [phi = exp(-0.5(y-a-b*x)^2))/sqrt(2*pi)]. The threshold or intercept is the "a", as I understand, while the "b" is the probit coefficient of one predictor (slopes). That latter I can define by labeling variables in the model command.
If you have time to spare and to help, I would be grateful.
Thanks.
Natalie
Hi Natalie and others,
I am also struggeling with this problem. Do any of you have succeeded in creating a working syntax?
Kind regards, Lisa
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression