Data Analysis MT

Element Sampling

Sample individual units

Cluster Sampling

sample groups, all units within the groups sampled

Non-coverage

systematically excludes populations

Unit non-response

data missing for all variables of individual

Item non-response

Data missing for some variables of individual

Censored variables

systematically excludes fraction of population in way s that depend on value of variable of interest (ex. capping income at 500k)

Interpret b in Y = a + bln(X)

1% increase in X results in b/100 change in Y

Interpret b in ln(Y) = a + bX

1 unit increase in x is b*100% increase in YI

If outcome variable is binary, interpret b in Y = a + bX

additional unit X changes % chance of Y by b percentage points

Balanced vs unbalanced panel data

Balanced observes every unit in every period of time, unbalanced has missing values

In panel data there is likely serial correlation between regression errors of same individual over time, so use what

Clustered standard errors (use “vce(cluster id)” in stata)

For DID coefficient to be causal, what assumptions must be true

Parallel trends, treatment and control would change in same way without treatment, and no simultaneous confounding changes that would affect groups differently

What is causal effect formula

Yi(1) - Yi(0)

ATE formula with random assignment

E(Yi | Di = 1) - E(Yi | Di = 0)

ATT Formula

E(Yi(1) - Yi(0) | Di = 1)

Threats to internal validity

failure to randomize, noncompliance, internal attrition

Threats to external validity

Non representative sample, non representative program, general equilibrium effects (study changes overall population behavior)

Stata command to identify panel data

xtset id year (id is individual, year is time variable)