Assessing Studies based on multiple regression

Question 1

Internal validity

Answer 1

the inferences on causal effects are valid for the population

• the estimator of the causal effect should be unbiased and consistent

hypothesis test shoud have the desired significance level

Question 2

external validity

Answer 2

the inferences can be generealized from the population studied to other population of interest or settings

• compare studies on differnte but related pops

Question 3

threats to internal validity

Answer 3

1. Endogeneity problem:
   - omitted variables
   - misspecified functional form
   - erros in variables (measurement error)
   - sample selection
   - simultaneous causality
2. Inconsistent standard errors

Question 4

possible solutions to omitted variables

when OV are observed:

Answer 4

1. what is the coefficient of interest?
2. (before running ols), what is the most important sources of omitted variable bias?
3. test nonzero coefficient of questionable variables
4. provide full disclosure so that readers can see the effect of including questionable variables

Question 5

possible solutions to omitted variables

when OV are NOT observed:

Answer 5

• use a randomized controlled experiment
• If IV is time constant., use panel data
• if an instrument is available, use IV estimation

Question 6

Solutions to measurement error problem:

Answer 6

1. get an accurate measure of X if possible

2.

• use IV ( a variable that is correlated with X but uncorrelated with w)
• mathematically model the measurement error and make a correction

Question 7

solution to simultaneous causality bias:

Answer 7

1. IV if instrument is available

2. randomized controlled experiment if possible

Question 8

Inconsistent standard errors

Answer 8

correlation of the error term across observations ( cross- sectional dependence or serial correlation)

• sampling is based on geographic units (state level data)
• time series or panel data ( repeated observations on the same entity over time)
• cross sectional dependence is possible ( unobserved common shock)
• Test and CI do not work properly ( significant level incorrect)

Question 9

solutions to inconsistent S.E

Answer 9

• heteroskedasticity-robust S.E
• Heteroskedasticity-autocorrelation-consistent S.E
• serial correlation- robust S.E for panel data models