Introduction
Explain where impact evaluatin ti inti the iyitem if minitiring & evaluatinn
o Monitoring: ietting giali and tracking indicatiri if prigreiin
o Evaluaton: iyitematc aiieiiment if reiulti
Operatonal Evaluaton: hiw efectve were prigrami rere there gapi
between planned vin realized iutcimei
Impact Evaluaton: are the changei ident ed really due ti the
prigram/interventin
Uie examplei ti explain why iimple with-and-withiut and befire-afer cimpariiini failn
o With-and-withiut
Efect in incime in happineiin
o Befire-afer
Financial aiiiitance ti purchaie inputin
o With-and-withiut and befire-afer cimpariiini fail becauie the aiiumptin if
ceterii paribui diei nit hildn
Explain what “pitental iutcimei”, the “ciunterfactual”, and “ielectin biai” aren
o Potental outcome : actual iutcimei and hypithetcal iutcimein
o Counterfactual: hypithetcal iituatin that iayi what wiuld have happened ti
partcipanti had they nit partcipated in a prigramn (Orr vice veriano
o Selecton Bia : diferencei between the treatment griup and cintril griupn
De ne the Fundamental Priblem if Cauial Inferencen
o It ii impiiiible ti ibierve all pitental iutcimei in irder ti meaiure the impact
efectn
De ne the reTE, reTT, and reTUn
o reTE: reverage efect if treatment, fir the entre pipulatinn
E[Yi(1o-Yi(0o]
o reTT: reverage efect if treatment, fir the treatedn
E[Yi(1o-Yi(0o|T=1] = E[Yi(1o|T=1] - E[Yi(0o|T=1]
o reTU: reverage efect if treatment, fir the untreatedn
E[Yi(1o-Yi(0o|T=0] = E[Yi(1o|T=0] - E[Yi(0o|T=0]
Shiw hiw a with-and-withiut cimpariiin can be writen ai the reTT plui ielectin biain
o E[Yi(1o|T=1] - E[Yi(0o|T=0] =
o E[Yi(1o|T=1] - E[Yi(0o|T=1] + E[Yi(0o|T=1] - E[Yi(0o|T=0] =
o reTT + Selectin Biai
Uie the pitental iutcime framewirk ti predict the iign if the biai frim a with-and-
withiut cimpariiinn
Book
Twi typei if quanttatve impact aiieiimentn
o Ex-ante: meaiurei intended impacti if future prigrami and piliciein
o Ex-po t: meaiurei actual efectin
, Randomisation
Explain in yiur iwn wirdi hiw randimiiatin iilvei ielectin biai (inen, makei cimpariiini
ceterii paribuion
o Law of large number : iample average will appriach pipulatin average by
increaiing iample iizen
o Due ti randimiiatin, the inly diference between the treatment and cintril
griupi ii the diference in treatment itatuin Hence, the cintril griup can be uied ai
ciunterfactual ti the treatment griupn
o E[Yi(0o|T=1] = E[Yi(0o|T=0]
Diicuii the diference between a lab and a eld experiment, and the 4 diferent methidi if
randimizatinn
o Lab experiment : cintrilled envirinment & art cial randimiiatin
o Field experiment : real envirinment & art cial randimiiatin
o Natural experiment: real envirinment & ciincidental randimiiatin
o 4 diferent methidi
Over ub cripton: (in caie if limited reiiurceio aiiign treatment randimly
aming iubiet if eligible individualin
Randomi ed pha e-in: gradual implementatin acriii eligible areain
Within-group randomi aton: within areai, iime individuali/iubgriupi
randimly receive treatmentn (Similar ti randimiied phaie-in, but in imaller
icaleno Mire likely ti be iuiceptble ti ipilliver efectin
Encouragement de ign: randimiie enciuragement ti take up treatmentn
Treatment ii nit randim, but enciuragement iin Spilliveri can be meaiured
fir example by cillectng data in ither memberi if hiuiehildn
Diicuii at leait 4 pitental iiiuei with randimiiatinn
o Ethical iiiuei
o External validity
o Cimpliance: iime individuali are ifered treatment, but di nit partcipaten
o Selectve atriton (frim the textbiiko: peiple can drip iut if a prigramn
o General equilibrium efecti: diferent efecti baied in icalen
o Spilliver efecti: interventin bene ti alii nin-treatedn
o Expeniive and iliwn
Explain hiw yiu can eitmate the treatment efect in the caie if randimiiatin, and why
yiu din’t need cintril variablein
o In randimiiatin, can uie with-and-withiut cimpariiinn
o The valuei if the cintril variablei are the iame fir bith treatment and cintril
griupi, ii when the diference ii taken ti meaiure the efect if treatment, the
cintril variable valuei cancel iutn
Eitmate the treatment efect in caie if randimiiatin uiing Statan
o Uie iimple t-teit and iimple regreiiiinn
o Cintril variablei di nit change ciefcient, but may reduce itandard erririn
Explain where impact evaluatin ti inti the iyitem if minitiring & evaluatinn
o Monitoring: ietting giali and tracking indicatiri if prigreiin
o Evaluaton: iyitematc aiieiiment if reiulti
Operatonal Evaluaton: hiw efectve were prigrami rere there gapi
between planned vin realized iutcimei
Impact Evaluaton: are the changei ident ed really due ti the
prigram/interventin
Uie examplei ti explain why iimple with-and-withiut and befire-afer cimpariiini failn
o With-and-withiut
Efect in incime in happineiin
o Befire-afer
Financial aiiiitance ti purchaie inputin
o With-and-withiut and befire-afer cimpariiini fail becauie the aiiumptin if
ceterii paribui diei nit hildn
Explain what “pitental iutcimei”, the “ciunterfactual”, and “ielectin biai” aren
o Potental outcome : actual iutcimei and hypithetcal iutcimein
o Counterfactual: hypithetcal iituatin that iayi what wiuld have happened ti
partcipanti had they nit partcipated in a prigramn (Orr vice veriano
o Selecton Bia : diferencei between the treatment griup and cintril griupn
De ne the Fundamental Priblem if Cauial Inferencen
o It ii impiiiible ti ibierve all pitental iutcimei in irder ti meaiure the impact
efectn
De ne the reTE, reTT, and reTUn
o reTE: reverage efect if treatment, fir the entre pipulatinn
E[Yi(1o-Yi(0o]
o reTT: reverage efect if treatment, fir the treatedn
E[Yi(1o-Yi(0o|T=1] = E[Yi(1o|T=1] - E[Yi(0o|T=1]
o reTU: reverage efect if treatment, fir the untreatedn
E[Yi(1o-Yi(0o|T=0] = E[Yi(1o|T=0] - E[Yi(0o|T=0]
Shiw hiw a with-and-withiut cimpariiin can be writen ai the reTT plui ielectin biain
o E[Yi(1o|T=1] - E[Yi(0o|T=0] =
o E[Yi(1o|T=1] - E[Yi(0o|T=1] + E[Yi(0o|T=1] - E[Yi(0o|T=0] =
o reTT + Selectin Biai
Uie the pitental iutcime framewirk ti predict the iign if the biai frim a with-and-
withiut cimpariiinn
Book
Twi typei if quanttatve impact aiieiimentn
o Ex-ante: meaiurei intended impacti if future prigrami and piliciein
o Ex-po t: meaiurei actual efectin
, Randomisation
Explain in yiur iwn wirdi hiw randimiiatin iilvei ielectin biai (inen, makei cimpariiini
ceterii paribuion
o Law of large number : iample average will appriach pipulatin average by
increaiing iample iizen
o Due ti randimiiatin, the inly diference between the treatment and cintril
griupi ii the diference in treatment itatuin Hence, the cintril griup can be uied ai
ciunterfactual ti the treatment griupn
o E[Yi(0o|T=1] = E[Yi(0o|T=0]
Diicuii the diference between a lab and a eld experiment, and the 4 diferent methidi if
randimizatinn
o Lab experiment : cintrilled envirinment & art cial randimiiatin
o Field experiment : real envirinment & art cial randimiiatin
o Natural experiment: real envirinment & ciincidental randimiiatin
o 4 diferent methidi
Over ub cripton: (in caie if limited reiiurceio aiiign treatment randimly
aming iubiet if eligible individualin
Randomi ed pha e-in: gradual implementatin acriii eligible areain
Within-group randomi aton: within areai, iime individuali/iubgriupi
randimly receive treatmentn (Similar ti randimiied phaie-in, but in imaller
icaleno Mire likely ti be iuiceptble ti ipilliver efectin
Encouragement de ign: randimiie enciuragement ti take up treatmentn
Treatment ii nit randim, but enciuragement iin Spilliveri can be meaiured
fir example by cillectng data in ither memberi if hiuiehildn
Diicuii at leait 4 pitental iiiuei with randimiiatinn
o Ethical iiiuei
o External validity
o Cimpliance: iime individuali are ifered treatment, but di nit partcipaten
o Selectve atriton (frim the textbiiko: peiple can drip iut if a prigramn
o General equilibrium efecti: diferent efecti baied in icalen
o Spilliver efecti: interventin bene ti alii nin-treatedn
o Expeniive and iliwn
Explain hiw yiu can eitmate the treatment efect in the caie if randimiiatin, and why
yiu din’t need cintril variablein
o In randimiiatin, can uie with-and-withiut cimpariiinn
o The valuei if the cintril variablei are the iame fir bith treatment and cintril
griupi, ii when the diference ii taken ti meaiure the efect if treatment, the
cintril variable valuei cancel iutn
Eitmate the treatment efect in caie if randimiiatin uiing Statan
o Uie iimple t-teit and iimple regreiiiinn
o Cintril variablei di nit change ciefcient, but may reduce itandard erririn
