use "http://www.learneconometrics.com/data/stata/insurance.dta", clear
# delimit ;
gen age_sq=age*age;

probit  insured  selfemp;
mfx;
probit  insured  selfemp  male deg_ged deg_hs deg_ba deg_ma 
        deg_phd deg_oth married familysz reg_ne reg_mw 
        reg_so race_bl race_ot age age_sq;
mfx;
estat clas;

/* The coefficient on self employed is significantly negative  
** at the 5% level (one-sided test).
** Also, the estimated marginal effect of self-employment gets 
** substantially bigger once control variables are added.  
** To be safe, use the larger model.  
*/

testparm age age_sq;

/* The joint hypothesis that age doesn't matter is rejected */

/* To test the idea that the self-employment effect differs for 
** older workers requires the creation of an interaction. */

gen age_self = age*selfemp;
probit  insured  selfemp  male deg_ged deg_hs deg_ba deg_ma 
        deg_phd deg_oth married familysz reg_ne reg_mw 
        reg_so race_bl race_ot age age_sq age_self;
mfx;

/* The null hypothesis that there is no interaction is not rejected
** at 5%.  */

/* For the last part we want to know whether employment status 
** has any effect on health status. */

probit healthy selfemp  male deg_ged deg_hs deg_ba deg_ma 
       deg_phd deg_oth married familysz reg_ne reg_mw reg_so 
       race_bl race_ot age age_self;

mfx;
 
/* A one sided test, the self-employment variable is significant 
** at the 5% level.  
** Notice that the marginal effect has a much smaller p-value.....
** Also, age is negative and significant.  Younger workers 
** are probably healthier. 
*/


/* Two-way causality is certainly possible.  
** Moral hazard suggests that those with insurance
** tend to take more health risks and be less healthy on average.  
** In the insurance equations healthier people are more likely to 
** be self employed becuase they percieve less need for 
** insurance (which usually costs more when you pay for it yourself).  */