Scribd
Upload
915 views3 pages

Mostly Harmless Econometrics Notes Part 1

This document discusses key concepts in causal inference research: 1) It introduces four frequently asked questions (FAQs) about causal research design, including identifying the causal relationship of interest and the ideal experimental method. 2) Randomized experiments are described as the best research design but are often infeasible, so natural sources of random variation are exploited instead. 3) Regression analysis is discussed as a tool for causal inference, but cannot prove causality without random assignment to control for selection bias. 4) The conditional expectation function is presented as a way to decompose a random variable and understand the predictive power of regression even when it cannot prove causation.

Uploaded by

Liam Donovan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
915 views3 pages

Mostly Harmless Econometrics Notes Part 1

This document discusses key concepts in causal inference research: 1) It introduces four frequently asked questions (FAQs) about causal research design, including identifying the causal relationship of interest and the ideal experimental method. 2) Randomized experiments are described as the best research design but are often infeasible, so natural sources of random variation are exploited instead. 3) Regression analysis is discussed as a tool for causal inference, but cannot prove causality without random assignment to control for selection bias. 4) The conditional expectation function is presented as a way to decompose a random variable and understand the predictive power of regression even when it cannot prove causation.

Uploaded by

Liam Donovan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Chapter1:QuestionsaboutQuestions

4FAQs:

Whatisthecausalrelationshipofinterest?
Descriptiveresearchislessinterestingb/citdoesntanswerquestions
ofbusinessorpolicy
Whatistheidealexperiment?
Oftenimpossibletoactuallycarryout
guidesthestructure/goalsoftheactualexperiment
Helpdecideonfruitfulresearchtopicsifyoucantsatisfactorilyresolve
thequestionwithanidealexperiment,itsawasteoftime
AvoidFUQdquestions(FundamentallyUnidentified
Questions)
exampleistheeffectofschoolentrydateoneducational
outcomesolderkidsdobetterbecausetheyreolder,evenif
theyendupworseoffatage20
Whatisyouridentificationstrategy?
Usedwhendataisnotgeneratedbyarandomizedtrialtoapproximatea
realexperiment
Whatisyourmodeofstatisticalinference?
Datararelycoversthewholepopulationhowdoyouextrapolate?

Chapter2:TheIdealExperiment
Randomassignmentisthebestresearchdesign,buttheyreveryexpensive
Correlationdoesntimplycausationdohospitalsmakepeoplesicker?
,y
isthemeasureofinterest,D
isthetreatmenteffect(foran
i
i
individual)
y
isthebaseoutcomeforthepatient(withoutthetreatment
0i
effect)
y
y
isthecausaleffectofthetreatmentfortheindividual
1i
0i

inrealdata,selectionbiasskewsresults
sickpeoplegotohospitals,healthypeopledont

randomassignmentgetsridofselectionbias,lettingususetheobserveddifferencein
outcomesasareliablemeasureofcausality

However,randomizedtrialsaredifficultandexpensive,somostresearchexploitsnatural
sourcesofrandomvariation
Regressionisuseful:
iftreatmenteffectissameforeverybody:

=treatmenteffect,=baseoutcome,
=randomvariationfromE(y
)
0i

selectionbias=correlationbetweentheregressionerror(
i

)&theregressor(D
)
i
tofindthetreatmenteffectinarandomexperiment,regre
ssY
onD
i
i
Controllingforvariablesreduces
theresidualvariance,makingtheestimatemoreaccurate

MakingRegressionMakeSense
Withoutrandomizedexperiments,regressionjustmakespredictions,cantspeaktocausality
Predictivepowersummarizedby
ConditionalExpectationFunction(CEF)
CEFisthepopulationaverageofY
whenmultiplecovariatesX
areheldfixed(E[Y
|X
])
i
ki
i
i

f
istheconditionaldensitydistribution
y
thatis,themeanofy
acrossthepopulationistheunconditional
i
expectationoftheCEF(
lawofiteratedexpectations)

importanceisthatitbreaksarandomvariableintotwopieces:
theCEFDecompositionProperty

ismeanindependentofX
(
E[
|X
])
i
i
i
i

isthusuncorrelatedwithX
i
i

(usetheCEFDecompositionProperty,thenbreakapartthetermsoftheexpectationfunction
thesecondtermwillreduceintoE[Y
|X
]bythelawofiteratedexpectations)
i
i

TheCEFPredictionProperty
saysthattheCEFisthebestpredictorofY
givenX
b/cit
i
i
solvesaminimummeansquarederrorpredictionproblem
formalstatement:

argmin(f(m(x
))=thevalueofm(x
)forwhichthefunctionf(m(x
)isminimized
i
i
i

Proof:varianceofY
isvarianceofCEF+varianceofresidual
i
2
varianceoftheresidual=E[
i]asvarianceistheexpectationofthedifferencefromthemean
(whichis0fortheresidual).

linearregressionisdonebyfindingtheCEFthatfitsthefollowingfunction

(X
b=m(x)fromthegeneralizedCEFequation)
i
WhenweassumethattheexpectationoftheerroracrossX
is0:
i

uselessmaterialoutputs
MSW=refusefrommunicipalities(households,smallbusinesses,institutions)

You might also like