This preview shows pages 1–6. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Panel Data Methods PAM 3100 Professor Michael Lovenheim Fall 2010 PAM 3100 Panel Data Methods Panel Data Introduction Thus far, we have dealt with cross sectional regressions, where each data point is a unit of observation. Often, we have data on repeated observations of individuals, states, etc.. Such data are called panel data and can be used to overcome may of the problems associated with selection on observables. Examples: Panel Study of Income Dynamics: repeated observations of people over time. Statelevel repeated cross sections: repeated observations of some state outcome over time (like income, number of car crashes, cigarette sales, etc.). National Educational Longitudinal Study of 1988: Study a representative cohort of 8 th graders in 1988 and followed until 2000. PAM 3100 Panel Data Methods Panel Data Introduction All of these data sets are characterized by having a time component, t , and a crosssectional component, i . Index each observation y it , where i = 1 . . . N and t = 1 . . . T . For example, if we have repeated crosssections of states for ten years, T=10 and N=50. We have 50 states over 10 years for a total of 500 observations. For policy purposes, having repeated observations of a given unit will allow us to examine the change over time within unit of the effect of the policy. PAM 3100 Panel Data Methods Example: Minimum Legal Drinking Age Prior to 1986, states had different drinking ages. But, from the mid1970s to the mid1980s many states increased their MLDA. Lets say we want to examine the effect of changing the drinking age to 21 (from below 21) on teen involvement in fatal accidents. We have statelevel number of fatal crashes with 1820 year old drivers from 19771990. We want to find the average effect of the change within each state  we will call this the average treatment effect. Why do we not just take a cross section in 1980, for example, and estimate a crosssectional regression of the number of accidents on statelevel observables? The problem is selection on unobservables: why do states have different drinking ages? Is it due to unobserved factors that also influence teen drunk driving? If so, our estimate of the effect of the change in drinking age will be biased. PAM 3100 Panel Data Methods Example: Minimum Legal Drinking Age Also, what if the timing of MLDA changes is not random and is correlated with shocks or secular trends in teen drunk driving? Then, we will confound such trends with a program effect. In this case, there was a big national antidrunk driving campaign sponsored by MADD that likely had an effect on drunk driving. This is also the time period of the MLDA changes, so we want to disentangle trends due to MADD from the the effect of MLDA changes....
View Full
Document
 '08
 ABDUS,S.

Click to edit the document details