ENDOGENEITY - SIMULTANEITY

1
ENDOGENEITY - SIMULTANEITY

• Development
• Workshop

2
Basic logic behind this analysis

• People may be in the shadow
• because they want to evade taxation
• gt net unregistered wages gt net registered wages
• because they have no other option
• gt net unregistered wages gt net registered wages
• Nice identification strategy can test it on
  data
• BUT people in gray concentrate in sectors and
  occupations (not that many gray teachers or
  lawyers or top managers)
• Thus need to control somehow for the
  nonrandomness of unregistered employment

3
Propensity Score Matching
Confounding Influence
Treatment
Treatment
Outcome
4
Data and method

• Data comes from a dedicated survey by MLSA
• Original sample size huge
• Pros nice source of data, because gray people
  report too
• Cons low quality of data, many missing and
  incomplete (how about representativeness?)
• Method
• Can use parametric approach (disregard
  nonrandomness of unregistered employment) gt run
  estimates of discrimination
• Can use nonparametric approach (accounting for
  nonrandomness) gt get differences in wages

5
Parametric

• Get coefficients of wage regressions
• Decompose wage differentials into
• part attributable to differences in endowments
• part attributable to differences in returns to
  endowments
• Can do it at
• mean (Oaxaca-Blinder decomposition)
• median (Juhn-Murphy-Pierce decomposition)
• whole distribution (rather complex quintile
  estimators, we dont do this )

6
(No Transcript)
7
The obtained propensity score is irrelevant (as
long as consistent)

• NEAREST NEIGHBOR (NN)
• Pros gt tzw. 11
• Cons gt if 11 does not exist, completely
  senseless

8
The obtained propensity score is irrelevant (as
long as consistent)

• CALIPER/RADIUS MATCHING(NN)
• Pros gt more elastic than NN
• Cons gt who specifies the radius/caliper?

9
The obtained propensity score is irrelevant (as
long as consistent)

• Stratification and Interval
• Pros gt eliminates discretion in radius/caliper
  choice
• Cons gt within strata/interval, units dont have
  to be similar
• (some people say 10 strata is ql)

10
The obtained propensity score is irrelevant (as
long as consistent)

• KERNEL MATCHING (KM)
• Pros gt uses always all observations
• Cons gt need to remember about common support

Treatment Control