05-28-2014 02:31 PM
I'm doing some data analysis for a project and I feel I'm a bit over my head (my experience is limited to descriptive statistics and basic regression analysis). I have been tasked to create figures using predicted probabilities:
1. Mother education probability of child death gradient
2. Mother assimilation probability of child death gradient
My supervisors only use stata, so they're not able to offer me guidance for doing this in SAS.
Here's some of the instructions I've received: "In stata after running the probit, you use the predict command. I think its predict, pr. I am not sure that SAS or stata can generate the figures, that is something that would have to be done in excel. What we want from stata or sas for example is the predicted probability of child death for say someone with no schooling etc......Once you get those then you can plot them (it may be possible to do it directly from SAS but im not sure about that). If you have difficulty using the probits, thats fine just use the linear probability models for child death (ols)."
I think I've been able to generate the predicted probabilities using OLS (syntax below for first mother education group needed for figure 1), but I'm not sure how to do this (if possible) with a probit regression:
*MOTHER EDUCATION PROBABILITY OF CHILD DEATH GRADIENT;
*Use full regression subset since need overseas in there to keep SA in model;
proc reg data=sa.Ipumsi_sa_reg;
model CHDIED=SADC ROA SA;
*This gives one value for CHDIED2 in output data set--predicted probability of child death for mother with edu2=1--correct?*
Is the above correct for generating the predicted probability of child death for someone with EDU2=1 using OLS? Is there a way to do this using probit regression?
Thanks so much for any guidance. Please let me know if I can offer additional information.
05-28-2014 03:49 PM
OLS regression can provide conditional probabilities with a binary dependent variable, however OLS assumtions are violated. I would say use proc logistic with link=probit model option.
05-28-2014 04:23 PM
My suggestion would be to work through the examples under Proc Logistic.
By work through I don't mean read through them, I mean typing them out and checking to see if you understand the results.
You'll also want to look into how SAS defaults referential coding for categorical variables and odds ratio.
My internet is really slow so I can't include any links right now but I'll also suggest the UCLA SAS logistic regression walk through.