I am trying to analyze cognitive disorders (in 3 categories) according to total number of stressful life events occurred during the entire life (continuous variable) with Stata. I have interaction with a continuous variable and I would like to graphically represent this interaction. How can I extract Logit with post-estimation commands?
I wonder how many people know what a logit is! I would use Stata's -margins- command to output the predicted probabilities at different levels of your predictor variables (for example, at each quartile of stressful events) and graph these. People understand predicted probabilities, but would be baffled by a logit scale.
Have you ever seen this done? I'd like to see an example of what you have in mind.
Thank you Ronan for your answer! I'm not very expert in statistics and it is difficult for me to explain what I really want. I attached a graph I would like to represent with my data. As I work with logistic model, I cannot represent the probability since I will have only 1 or 0 so the idea is to represent the Logit... I'm not sure that is clearer :-(
Hi Sophie,
One option is to graph using "graph twoway function."
Imagine you have a logit model that has obtained an intercept of -4 and a coefficient of .1. Imagine also your predictor (x) ranges from 1 to 100.
This could be graphed using the following syntax:
graph twoway function y = invlogit(-4 +.1*x), range(1 100)
You will need to simply replace the "range()" with your observed range and the coefficients with the ones you observe from your logit model. The y-axis can be interpreted as the predicted probability and needs no complicated log-odds interpretation.
- joe
Thank you so much Joseph. I will try and I'll let you know ;-)
Sophie.
Try this
. sysuse auto
. logistic foreign mpg
. predict f_hat
. tw line f_hat mpg, sort
Thank you Ronàn for your reply to my last question! Very sorry for the late reply!
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Software