L10 - (PL) Introduction to Panel Data Models

Question 1

What types of data do we come across more in regression analysis?

Answer 1

• •Time series is a set of observations on the values that a variable takes at different times (daily, weekly, monthly, quarterly, annually, etc.).

• Cross-sections are data on one or more variables collected at the same point in time.

•Pooled data combine both time series and cross-sectional data. (but its different cross-sectional units over time (different firms looked at each different year))

•Panel or Longitudinal data are a special type of pooled data in which the same cross-sectional unit (a family or a firm) is surveyed over time.

* If time series dimension is different for different cross-sectional units we say we have a unbalanced panel

Question 2

What are the different types of Panel Data we gather?

Answer 2

• Micro panels ((British Household Panel, German Socio-economic Panel);
• British Household Panel –> cover things such as demographic, education, income, marital status
  
  • Collected for a large number of individuals (100,000s) over a short period of time (Min. 2years - Max 10-20years)
    
    • Long cross-sectional dimension but short time series one
  • Usually, firms/households are randomly sampled so are unlikely to be correlated
• Macro panels (IMF International Financial Statistics, world bank data ).
  
  • IMF International Financial Statistics –> approx. 30,000-time series data coving more than 200 countries starting from 1948 this includes exchange account and the main global country economic indicators
    
    • Number of countries (e.g. 20 OECD countries (not exceeding 100-200 countries) over a period of time (annual over 20-60 years)
    • Can also be very high-frequency data like daily observations of a stock index changing over time
    • Econometrics techniques problems for macro panels –>
      
      • deal with non-stationarity/unit roots
      • cointegration
      • cross-country dependence –> there is a likelihood country can be correlated

Question 3

What are the benefits of using Panel data?

Answer 3

• •Controlling for individual heterogeneity.
• Give more informative data, more variability (within states and between them - not just aggregated on a country level), less collinearity among the variables, more degrees of freedom and more efficiency.
• Dynamics of adjustment (change, duration of economic stance, speed of adjustment to economic policy changes, intertemporal relations).
  
  • observes changes
• Identify and measure effects that are simply not detectible in pure cross-sectional and pure time-series data.
  
  • Women have a 50% chance of participating in the workforce
    
    • Does this mean there is a huge turnover or there are 50% of women who work full-time and some not at all –> only panel data can discriminate between these cases
• Construct and test more complicated behavioural models than purely cross-section and time-series data
• .Biases resulting from aggregation over firms or individuals may be reduced or eliminated (Micro panels).

Question 4

Controlling for individual heterogeneity: Benefits of using Panel data?

Answer 4

Panel Data suggests that individuals, firms, states and countries of heterogenous –> not controlling for this leads us to get biased results

• Baltagi and Levin (1992)- cigarette demand across 46 American states for the years 1963-1988
• •Function of price and income;
• What else could influence this (potentially hard to measure/observe)
  
  • •State-invariant (cross-sectional unit) variables, i.e. advertising on national TV and radio;
  • •time-invariant variables i.e. religion and education.
    
    • It May be hard to get a figure for how many Mormons there are in each state (but will this change a lot over time)
• omission of these variables could lead to biased estimators
  
  • panel data can control for these variables given the fact they are observed or not while time series/cross-sectional results cannot
  • UTAH has low smoking rates but that isn’t because of income and price but because it is a Mormon state as is prohibited by their doctrine

Question 5

Limitations of Panel Data?

Answer 5

• •Design and data collection problem (problems of coverage, nonresponse (not giving a proper or any answer), recall (respondent does not remember things correctly), etc.).
• •Distortions in measurement error.
  
  • Error in test, memory error, deliberate distortion in responses ( individuals don’t want to admit they take drugs)
  • Also, inappropriate informants, recording errors and interviewer effects
• •Selectivity problem:
  
  • –Self-selectivity. –> people choose not to work cause their reservation wages > that actual wage –> observe the characteristics of these individuals but not their wage
    
    • As it’s only the wage missing we call the sample censored
    • However if we do not observe all the data on these people we call this a truncated sample
  • –Non-response (partial and complete).
    
    • Refusal to answer or no one at home etc.
    • Partial –> one or more questions are unanswered or fail to provide a useful response
    • Both cause efficiency loss and misidentification problems in the population parameters
  • –Attrition.
    
    • respondents may die, move away or find the cost of responding too high
    • biasing attrition —> those leaving the sample found to have lower earnings, low education levels –> introduces biases into the estimation parameters
• •Short time-series dimension for micro panels.
  
  • Increases the time span of a panel can lead to increased chances of attrition
    
    • Increases the computational difficulty for limited dependent variable panel models
• •Cross-section dependence for macro panels.
  
  • long time series on countries could lead to cross-country dependency issues