Prediction and residual values analysis

Below it is shown how to extract prediction and residual values for the various types of regression analyses. The use of histogram and hexbin for analysis of the results is demonstrated. In particular, histogram is a very useful command that can be used to visually study the degree to which the residuals are normally distributed. But you can in principle make use of all available and relevant commands for further analyses.

require no.ssb.fdb:23 as db

create-dataset regresjonsdata
import db/INNTEKT_WLONN 2019-12-31 as lønn
import db/INNTEKT_BER_BRFORM 2019-12-31 as formue
import db/BEFOLKNING_FOEDSELS_AAR_MND as faarmnd
import db/BEFOLKNING_KJOENN as kjønn
import db/BEFOLKNING_STATUSKODE 2020-01-01 as bosattstatus

keep if bosattstatus == '1'

generate alder = 2019 - int(faarmnd/100)

generate mann = 0
replace mann = 1 if kjønn == '1'

//regress
regress lønn alder mann formue
regress-predict lønn alder mann formue
histogram predicted
hexbin predicted lønn
regress-predict lønn alder mann formue, residuals(res) predicted(pred) cooksd(cook)
regress-predict lønn alder mann, residuals(res2) predicted(pred2) cooksd(cook2)
histogram pred
histogram res
histogram cook
histogram res2

//ivregress
ivregress lønn mann (formue = alder)
ivregress-predict lønn mann (formue = alder), residuals(res3) predicted(pred3)
histogram pred3
histogram res3

//logit
summarize lønn formue
histogram lønn
histogram formue
generate høylønn = 0
replace høylønn = 1 if lønn > 800000
generate høyformue = 0
replace høyformue = 1 if formue > 4000000

logit høylønn alder mann høyformue
logit-predict høylønn alder mann høyformue, residuals(res4) predicted(pred4) probabilities(prob4)
histogram pred4
histogram res4
histogram prob4


//probit
probit høylønn alder mann høyformue
probit-predict høylønn alder mann høyformue, predicted(pred5) probabilities(prob5)
histogram pred5
histogram prob5

//mlogit
generate lønnkat = 0
replace lønnkat = 1 if lønn > 0
replace lønnkat = 2 if lønn > 800000

mlogit lønnkat alder mann høyformue
mlogit-predict lønnkat alder mann høyformue, predicted(pred6) probabilities(prob6)
summarize pred6_1
histogram pred6_2
histogram prob6_1
histogram prob6_2


//regress-panel
sample 0.05 54321

clone-units regresjonsdata paneldata
use paneldata
import-panel db/INNTEKT_WLONN db/BEFOLKNING_FOEDSELS_AAR_MND db/BEFOLKNING_KJOENN db/INNTEKT_BER_BRFORM 2017-12-31 2018-12-31 2019-12-31
rename INNTEKT_WLONN lønn
rename INNTEKT_BER_BRFORM formue
generate alder = 2019 - int(BEFOLKNING_FOEDSELS_AAR_MND/100)
generate mann = 0
replace mann = 1 if BEFOLKNING_KJOENN == '1'

regress-panel lønn mann alder formue
regress-panel lønn mann alder formue, re
regress-panel-predict lønn mann alder formue, predicted(ppred1) residuals(pres1) effects(peff1)
regress-panel-predict lønn mann alder formue, re predicted(ppred2) residuals(pres2) effects(peff2)
histogram ppred1
histogram pres1
histogram peff1
histogram ppred2
histogram pres2
histogram peff2
hausman lønn mann alder formue