Heteroskedastic ordered probit models - YouTube

hi this is Gabriela Ortiz for the stata YouTube  channel I'd like to introduce you to the new  

hetoprobit command in stata 16 hetoprobit  is an ordinal probit regression model with  

multiplicative heteroskedastic errors this  command generalizes ordinal probit regression  

by incorporating a variance model with its own  set of predictor variables these may include  

variables that are included in the model for the  latent variable mean and variables that are not  

to demonstrate I have loaded an extract from  the American time use survey that eat health  

15 dataset but to keep things simple I will not  take the survey design into account I'd like to  

fill a model for the self rating of physical  health as a function of an individual's age  

body mass index and whether or not they exercise  in the week prior to the survey note that health  

ranges from 1 for poor to 5 for excellent to fit  this model I can go to statistics ordinal outcome  

and heteroskedastic ordered probit regression  here I'll specify the outcome variable which is  

health and the independent variables age BMI and  exercise age and BMI are continuous so go ahead  

and check this box to treat them as continuous  and last I'll add exercise which is an indicator  

variable so I can click OK now I suspect that the  variation in health status is greater in an older  

population as compared with a youthful population  and I believe that health variability may differ  

between those who exercise and those who do not so  I can use header profit to model the variance as a  

function of these variables by simply specifying  them here in the variance model so I'll go ahead  

and add age and exercise and I can click OK once  more to fit this model now I see 3 tables here in  

the output first I see the effect of independent  variables on health status then I see the output  

for the variance model under Ln Sigma and last  I see the cut points for the unobserved latent  

variable and down here in the bottom I also see  a test of homogeneity of the variance next I'd  

like to use the margins command to compute the  expected probability of having excellent health  

for the 3270 age range so I can go to statistics  post estimation and I'll select marginal effects  

marginal means and predictive margins next I'll  click on custom analysis and for the prediction  

I'll specify the fit outcome and click OK next  I'll add the age range so simply specify the  

covariate and specify the age range 30 to 70  in increments of 10 and click OK and now we  

see how the probability of rating one's health  as excellent decreases with each decade and I  

can visualize this relationship between age  and the expected probability of rating one's  

health as excellent by simply typing margins  plot there are a lot of other interesting  

things you can do with margins after fitting a  model with header probit for example I obtained  

population average expected probabilities but  you can also obtain expected probabilities for  

subpopulations or individuals and you might  explore the relationship between covariance  

and the error variance I mentioned before that  this was data from a complex survey and I didn't  

take that into account in the model I fit but  you can use header probit to fit models with  

complex survey data by specifying the svy prefix  you can also use robust cluster robust bootstrap  

and jackknife standard errors and you may even  choose to fit a bayesian heteroskedastic ordered  

probit regression model if you would like to  learn more about the new hetoprobit command in  

Stata 16 you can download the manual from our  website at Stata.com thank you for watching