version 9.2

log using C:\Teaching\IGP\log\12.diagnostics.log, replace

* Chapter 12 Goodness of fit and regression diagnostics
* modified on Feb. 11, 2007 by Leena Choi

***************************************************************************************
* Section 12.2
***************************************************************************************
*** hemoglobin data in Table 3.2
use "C:\Teaching\IGP\data\haemoglobin.dta", clear

* histogram with normal curve
hist hemo, normal

* inverse normal plot
qnorm hemo

* check skewness and kurtosis
summarize hemo, detail

* Shapiro-Wilk test
swilk hemo

***************************************************************************************
* Section 12.3 
***************************************************************************************
use "C:\Teaching\IGP\data\EMS\perulung.dta", clear
regress fev1 age height male

* get residuals from the previous regression
predict r, residual

* histogram of residuals with normal curve
hist r, normal
* inverse normal plot of residuals
qnorm r
* Shapiro-Wilk test for residuals
swilk r

* Scatter plot of standardized residuals against fitted values
predict yhat, xb
scatter r yhat, yline(0)

* You can draw the same plot without calculating residuals and fitted values using rvfplot command
rvfplot, yline(0)

* if you want to get stadardized residuals, use:
predict rs, rstandard

* Fig. 12.4, Table 12.2
use "C:\Teaching\IGP\data\cookD.dta", clear
regress y x
scatter y x || lfit y x

predict r, residual

* influence statistics: calculate Cook's distance (D)
predict cook, cooksd

* how big is big: Cook's D > 4/n
list id y x cook r

* sensitivity analysis
regress y x
regress y x if id!=10

log close