clear
capture log close
log using final_example, text replace

use final_1

* begin analysis


/*
checklist:
1. sampling probability: create weights
2. influential cases?
3. heteroskedasticity?
4. multicollinearity?
5. missing data?
6. estimate and interpret cofficients
(note: real data may be more complex...(explain)
7. iv?
*/

* your analysis will differ depending on the specific characteristics
* of your data.
* remember to explain why you do things and how you interpret the output
* and the potential consequences of choices you have made and problems you encounter

notes
des 
sum


gen wgt=1/p


* influential cases
reg math effort tv gender
predict d, cooksd

sum d
list if d>.3 & d~=.
drop if d>.3 & d~=.

* heteroskedasticity
reg math effort tv gender [w=wgt]
estat hettest, rhs
estat hettest tv
predict ehat, resid
gen ehat2=ehat^2
reg ehat2 effort tv gender [w=wgt]

gen wgt2=wgt/tv
reg math effort tv gender [w=wgt2]
estat hettest, rhs
capture drop ehat ehat2
predict ehat, resid
gen ehat2=ehat^2
reg ehat2 effort tv gender [w=wgt2]


reg math effort tv gender [w=wgt2], robust

* multicollinearity
* substitute current #1 model here
reg math effort tv gender [w=wgt2]
estat vif



ice math effort tv gender using impute [w=wgt2], m(5) replace
drop _all
use impute
* substitute current #1 model here
micombine reg math effort tv gender [w=wgt2]

* iv?  (remember untestable assumptions must be argued theoretically)
reg effort rocky
predict efforthat
micombine reg math efforthat tv gender [w=wgt2]

ivreg math (effort=rocky) tv gender [w=wgt]


log close