all weeks Flashcards
what is correlation fallacy?
The logical mistake of believing that because two events occurred together , there is a cause-and-effect relationship.
What is a treatment effect?
The effect of a treatment on the treated
What is selection bias?
Selection bias arises when individuals are selected (or self-selected) for treatment based on (typically unobserved) characteristics that may also affect their outcomes.
This makes it difficult to disentangle the impact of the treatment from the factors that drove the selection.
What does endogeneity refer to? What conditions must be met broken for endogeneity?
Endogeneity refers to a situation in which Cor(X,Є) not equal 0 and can be due to a measurement error, Omitted variable bias and/or Reverse Causality.
If even one of these is true, then there is endogeneity
What are the two conditions that must be met for an instrument variable (IV) to be valid?
a) it should be correlated with the endogenous regressor (X) (First stage)
b) It should not be correlated with the error term (Є) (Exclusion restriction)
if both 1) and 2) hold IV estimates capture
the causal effect of endogenous X on Y
How do we conduct 2SLS?
✓ First stage: 𝑋 = 𝜃 + 𝜌𝑍 + 𝜇 , where Z is now a vector of two or more instruments.
✓ Collect predicted values 𝑋 from stage 1.
✓ Second stage: 𝑌 = 𝛼 + 𝛽 𝑋 + 𝜀
✓ Important: if next to endogenous regressor 𝑋, you have several exogenous regressors,
you need to include these exogenous regressors in both the first and second stage of the
2SLS procedure
What is the most commonly use method in papers to control for selection bias?
Selection bias can be completely removed when individuals are randomly assigned to treatment and control groups.
What are some alternative methods of randomization?
1) Classical
2) Oversubscription method
3) Within-group randomization
4) Encouragement designs
What do we describe as the statistical power of a test?
The power of a test is the probability that the test correctly rejects the null hypothesis (Ho) when it is false or alternatively rejects H1 when the null hypothesis is true.
What are the other factors that influence the statistical power of a test?
i) the significance level used in the test
ii) the magnitude of the effect - the greater the effect size the higher the power of the test
iii) the sample size - determines the amount of sampling error inherent in a test result
iv) the precision with which data are measured - the lower the precision, the lower the power of the test.
What does an RCT data timeline look like?
1) Listing - short census of the population
2) Baseline - typically a 60-90 minute survey (not necessary)
3) Intervention
4) midline - short-term (6-12 months) measure of outcome variables (not necessary)
5) Endline - measure of outcome variables after 12-24 months
long term line - measure of outcomes after 4 or longer years
What is stratified (block) randomization?
Randomization is performed separately within each stratum. Example of a stratum: median age.
We can stratify by more variables (e.g, race, initial ability, teachers, gender, etc.)
Which variables should we choose to stratify?
Variables we believe are strongly correlated with the outcome of interest, or may interact with the treatment effect.
How many variables should be stratify on?
In principle as many as you want, but there is a limit; usually 1 - 5 depending on the N
How do we use strata variable in the analysis of the treatment effect?
The strata variables should be added as covariates in the regression analysis -> increased efficiency and power of the hypothesis test.
What is attrition?
Attrition refers to a decrease in the number or size of the sample.
How does attrition act as a threat to the identification of the Average treatment effect (ATE)?
Two separate problems:
i) as N decreases -> Statistical power decreases -> decrease in the precision of estimates of ATE
ii) if there is differential survey attrition between treatment and control or differential compliance -> higher chances that ATE is biased.
If those who are benefiting least from a program tend to drop out of the sample, ignoring this
fact will lead us to overestimate a program’s effect
How do we distinguish between the two types of treatment effects?
1) Effects on those who were assigned to treatment (ITT - Intention To Treat)
2) Effects on those who adopted the treatment (ToT - Treatment On the Treated or LATE)
What is known as ‘Exclusion restriction’ and what are the possible threats in poses to treatment assignment?
Exclusion restriction - Treatment assignment affects outcomes only through treatment take-up.
Potential threats:
i) Placebo effects
ii) Demoralization effects (by participants in the control group)
iii) Experimenter demand effects (Hawthorne effects)
What is known as ‘Stable Unit Treatment Value Assumption’ (SUTVA) and what are the possible threats posed against it?
SUTVA - the instrument and treatment associated with person i do not influence the instrument and treatment associated with person j.
potential threats:
1) Spillover effects (from treatment to control and vice cersa)
2) General equillibrium effects
What is a spillover effect?
Treatment effects on some individuals may influence treatment effects on other individuals.
What is the relationship between Local Average Treatment Effect (LATE) and ToT?
If nobody in the control group gets treated (e.g., due to
spillovers from T to C) then E(Received T| C) = 0 and LATE = ToT
What is the ANCOVA specification?
Adds the baseline value of the outcome to the regression specification to reduce the variance of the treatment estimator.
This does not affect the value of the estimator of b0 (only its variance)
Regression specification
Y2(i) = a + b0Treat(i) + b1Y1(i) + e(i)
What is external validity?
The extent to which we can apply the conclusions of the study outside the context of the study.Regression specification
Y2(i) = a + b0Treat(i) + b1Y1(i) + e(i)inte
What is internal validity?
How well an experiment is done, especially whether it avoid confounding.
The less chance for confounding in a study, the higher its internal validity.
Provide a checklist for RCTs.
Randomisation (type, balance, etc.)
* Attrition and non-compliance (symmetric vs. differential)
* Treatment effect estimates (ITT, TOT, HTEs)
Threats to exclusion restriction?
Theory for differential effects for specific sub-groups?
Theory for externalities, spill-overs, general-equilibrium effects?
* Estimation precision (ANCOVA estimation, covariates, etc.)
* Overall: Internal/external validity
What is the general principle for identifying a poverty trap?
There is a poverty trap whenever the scope of growing income or wealth at a fast rate is limited for those who have too little to invest, but expands dramatically for those who can invest a bit more.
There is no poverty trap if the potential for fast growth is high among the poor, and then gradually stops as one gets richer.
What does a poverty trap look like on a graph where we plot income in the future against income today?
An S- shaped curve on that graph indicates a poverty trap.
Will aid help the poor escape from poverty?
if no poverty trap:
- Aid will make the person start from Y1 instead of Y0
- At best aid can just help a poor person to move up a little bit faster
if poverty trap:
- once in a lifetime income/help/nudge can have a huge effect to a person’s life
- Aid (policies/ interventions) can (although not necessarily) boost a poor person’s income permanently.
What are the two systems if thinking that consumers use when trying to maximize utility?
Automatic system
- Considers automatically what comes to mind
- Effortless
- Associative
- Intuitive
Deliberative system
- Considers a broad set of relevant factors
- Effortful
- Based on reasoning
- Reflective
What is a health-based poverty trap?
A health-based poverty trap occurs when individuals in poverty have reduced productive capacity due to the health complications they face due to the environments in which they live.
