* SHAZAM code downloaded from UCLA Econ 143 (CAMERON) WebSite: 
* HTML file called e143sh14.htm, and should have 
*  been downloaded as het1.sha

sample 1 52
read(het1.dat) weeks perfect
* now suppress a lot of distracting warnings
set nowarn
set nowarnskip

plot perfect weeks / nopretty

* does performance improve with weeks of experience?
ols perfect weeks
test weeks=0

* given repeated observations at each number of weeks, calculated sample
* estimate of conditional error variances for each number of weeks:

skipif(weeks.ne.1)
stat perfect / var=varp1
delete skip$

skipif(weeks.ne.2)
stat perfect / var=varp2
delete skip$ 

skipif(weeks.ne.3)
stat perfect / var=varp3
delete skip$

skipif(weeks.ne.4)
stat perfect / var=varp4
delete skip$ 

skipif(weeks.ne.5)
stat perfect / var=varp5
delete skip$ 

skipif(weeks.ne.6)
stat perfect / var=varp6
delete skip$ 

skipif(weeks.ne.7)
stat perfect / var=varp7
delete skip$ 

dim varperf 7
sample 1 7
genr wks=time(0)

* why do we need a new variable name for weeks as used here?
genr varperf:1=varp1
genr varperf:2=varp2
genr varperf:3=varp3
genr varperf:4=varp4
genr varperf:5=varp5
genr varperf:6=varp6
genr varperf:7=varp7

plot varperf wks

* now restore the full sample--otherwise only the first six observations
* would be used
sample 1 52

* create regression weights that are exactly equal to 1/(sigma squared) for
* each number of weeks:
if(weeks.eq.1) wsig=1/varp1
if(weeks.eq.2) wsig=1/varp2
if(weeks.eq.3) wsig=1/varp3
if(weeks.eq.4) wsig=1/varp4
if(weeks.eq.5) wsig=1/varp5
if(weeks.eq.6) wsig=1/varp6
if(weeks.eq.7) wsig=1/varp7
stat wsig
* if we use the above weights, what quantity does SHAZAM use to multiply
*  each of the variables (including the intercept) in the transformed 
*  regression?


* acknowledging nonconstant error variances, does performance improve
* with weeks of experience?
ols perfect weeks / weight=wsig
test weeks=0