566 Panel Data Methods
are injured in the same crash, allowing for different severities of accident (accident
fixed effects). Also the sample of passengers is used to abstract from differences
in the quality of driving between the treated and control groups. Robustness is
assessed by doing separate analyses by diagnoses and by medical procedures. In the
light of these results, the lower levels of treatment for the uninsured are attributed
to decisions made by providers in response to insurance status rather than dif-
ferences in background characteristics of the patients. Similar issues are faced by
Levitt and Porter (2001), who address the problem of selection bias in the US
Fatality Analysis Reporting System (FARS), which they use for an analysis of the
effectiveness of seat belts and air bags. The problem arises because data is only
included for fatal crashes. The use of safety devices influences the probability of
survival and hence of inclusion in the sample. The identification strategy adopted
to get around this problem is to use a sample based on crashes where someone in
a different car dies. The aim is to make the sample selection independent of the
observation’s own treatment status and outcomes. In doing so they find that seat
belts are more effective and air bags are less effective than previous evidence had
suggested.
12.2.2.2 Economic shocks
Evans and Lien (2005) make use of the 1992 Port Authority Transit (PAT) strike in
Allegheny County, Pennsylvania, as a source of independent variation in access
to prenatal care. Prenatal visits were affected most for black women and city res-
idents (in Pittsburgh) and the results show that, for these groups, missing visits
early in pregnancy had a detrimental effect, but missing those later in the preg-
nancy did not. The main source of information is observational data from the
1990–94 US Natality Detail Files, which contain a census of births in each given
year, taken from birth records. This is augmented by survey data that is used to
assess the impact of the strike on access to prenatal care. A control group of coun-
ties that were not affected by the strike are selected on the basis of regression
analyses. The use of prenatal care by women who were pregnant at the time of
the strike is included in regression equations for birth weight, gestation, mater-
nal weight gain and maternal smoking. Models are estimated by ordinary least
squares (OLS) and two-stage least squares (2SLS), the latter using the strike as an
instrument. These produce similar results, suggesting that selection bias is not a
problem. The clearest effect of prenatal care is on maternal smoking. The robust-
ness of the findings is tested by checking for a general decline in earnings or
employment coincident with the strike and for evidence of increases in abortions or
“unwanted” births.
In Frijterset al.(2005) the reunification of Germany in 1990 provides a natural
experiment to assess the causal effect of income on self-reported health satisfac-
tion. A positive and statistically significant effect is found, but the effect is small.
The increase in incomes for those in East Germany is used as a source of inde-
pendent variation in income that is not contaminated by reverse causality from
health. The suitability of this setting as a natural experiment is justified by the fact
that the changes in income associated with the fall of the Berlin Wall are assumed