Sociology 229:  Advanced Regression

 

Assignment #6:  Panel Models

 

Due:  Start of class, March 8

 

This assignment requires a dataset on the course website entitled “Assignment 6 Panel.dta” and an accompanying do-file.

 

The unit of analysis is the country, with annual data from approximately 1970 to 2007.  The dataset is unbalanced.

 

The topic is participation in higher education.  The dependent variable is higher education enrollments (“tert”, which stands for “tertiary enrollments”), the percentage of people attending colleges/universities (of all college-aged people).  Key independent variables:

Economic wealth:  GDP per capita (log).  Wealthier countries may have more participation in higher education.

Secondary school enrollments:  The more people that make it through high school, the more can go on to college.

Democracy:  Democratic societies may offer greater educational opportunities to all.

Organizational linkages to the international sphere:  International pressures to expand education may affect countries with strong international links.

 

Key variables:

countryid, the ID variable for each country

year, the variable indicating panel waves

 

  1. Download the dataset in STATA
  2. Create your own “do” file that opens the data
  3. Use the “xtset” command to prepare the data for panel analysis
  4. Use the “xtdes” command to describe the data
  5. Run fixed effects and random effects models, and conduct a Hausman test
  6. Use Allison’s (2009) method of decomposing variables in to means & deviations.  Estimate using a random effects model.
  7. Use the Beck/Katz approach for Time Series Cross Section (TSCS) data, which involves panel corrected standard errors, the lagged dependent variable, and country-level fixed effects
  8. Estimate some other dynamic panel data models, drawing on examples from the “do” file that I provided

 

Questions:

 

1.  Based on results from the “xtdes” command, is the dataset balanced or unbalanced?

 

2.  What is the result of your Hausman test?  Describe the main findings of the preferred model.  Interpret the coefficient for at least two independent variables.

 

3.  Did the Allison (2009) method effectively replicate the FE model?  Did you observe differences between the “within” and “between” coefficients for any variables?  If so, try to provide an interpretation/explanation.

 

4.  Did the Beck/Katz approach differ from the FE model?  If so, how?

 

5.  Did the “System GMM” results differ?  If so, how?

 

6.  Based on your knowledge, which models are most appropriate for this particular analysis?  Explain.

 

 

Turn in the following:

 

1.  Stata output from the basic RE, FE, and Hausman test.

 

2.  Stata output showing results from the Allison method

 

3.  Stata output showing results from the Beck/Katz approach

 

4.  Answers to the questions