A short introduction to applied econometrics Part C: Introduction to STATA

1
A short introduction to applied econometricsPart
C Introduction to STATA

• presented by
• Dipl. Volkswirt Gerhard Kling

2
How to transfer data to STATA?

• Stattransfer special software package
• Very convenient!
• Excel, SPSS, Eviews, STATA, Gauss etc.
• Use import function in STATA
• Similar to the Excel import procedure
• Sometimes mistakes (dot comma etc.)
• Advice Store your data in Excel
• Easy for calculations
• Still the best figures

3
The windows of STATA

• Output window shows estimation results
• Command line To type your instructions
• Review box You can go back to former commands
  and correct mistakes
• Variables List of current variables, can be
  included into commands
• Data editor similar to SPSS shows data set
• Do-file editor To run sequence of commands or
  self-designed programs

4
Help function

• Search for commands or online sources
• New programs are downloadable!
• Access to example files
• Files are written in STATA syntax
• Syntax of STATA commands
• command varlist if in, options required
  specifications
• Example
• regress gdp pop sav if gdplt2, robust nocons

5
The syntax of STATA

• Help file tells you the appropriate syntax
• Some commands have funny abbreviations like tsset
• Note write if gdp2 not if gdp2
• But write gen gdp2
• Some commands require to specify an option
  command (like lag(2) after newey)
• After a while, you will learn the syntax
  automatically!

6
Extended Harrod Domar model

• Problems with omitted variables
• Hence, we include additional information
• Urban of population in urban areas
• Trade of GDP (ImportExport)
• Debt of GDP (only external debt of state)
• Inflation CPI
• Real real interest rate
• FDI of GDP
• AID of GDP
• Over real overvaluation of the currency

7
How should we start?

• Get an overview of data structure
• list, data editor
• Use descriptive statistics
• sum variable, detail
• tabstat varlist, statistics(mean sd...)
• Missing values overview
• codebook varlist

8
How are variables distributed?

• Histograms histogram variable
• Kernel density kdensity variable
• Outliners are considerable
• Debt, inflation
• Extreme distribution of the aid variable
• Solution Transform aid variable
• Construct dummies for categories
• Less 10, less 25, less 50, extreme cases

9
Generate dummy variables

• We now try to build up categories of foreign aid
• gen d101 if aidlt10
• replace d100 if d10.
• gen d251 if aidlt25 aidgt10
• replace d250 if d25.
• gen d501 if aidlt50 aidgt25
• replace d500 if d50.
• gen dextreme1 if aidgt50
• replace dextreme0 if dextreme.

10
Correlation matrix

• To identify multicollinearity
• Later investigate relations precisely
• Command correlate varlist
• Correlation coefficients seem to be small
• But Many variables auxiliary regressions?

11
Use scatterplots

• Growth rate of GDP should appear on the vertical
  axis several explanatory variables
• Command scatter two variables
• Scatter with fitted curve
• Command twoway (scatter gdp sav) (lfit gdp sav)

12
First impression

• No relation urban, trade, fdi, over
• Positive sav
• Negative pop, real (few observations)
• Outliners and few observations
• Debt
• Inflation
• Preliminary results can help to specify
  regression!

13
How to specify a regression model?

• Different methodologies I prefer from general to
  specific
• Advantage Important effects included
• Disadvantage Too many variables
• Risk overspecification!
• Run regression with all variables
• Command reg depvar varlist
• Be careful when inserting dummies!

14
Regression with all variables

• Too few observations only 25!
• All coefficients are not significant
• Specification problem different sample size
• F-test indicates low explanatory power
• Adjusted R2 even negative
• What should we do now?
• Solution Use information provided by scatter
  plots drop variables with outliners and few
  observations

15
Drop variables with too few observations

• Drop debt and inflation few observations
• Drop urban, trade, fdi, over no relation
• Drop aid variables
• Specification tests like log likelihood are in
  this case not really reliable
• Because Change in sample size!

16
Log likelihood ratio test

• Skip variables with too few observations
• Then regress with sav, pop, urban, trade, fdi,
  over, aid dummies
• Save regression est store A
• Reduce model and test if reduction is allowed
• Command lrtest A
• Important You have to fix the sample size using
  if option!
• We end up with our HD model!

17
Latin American crisis

• Thus far, our approach was not a success
• We know from session two that Latin American
  countries behave differently!
• By inserting an intercept dummy, we control for
  this shock
• Now, re-run the specification procedure!
• Model C cannot be rejected!

18
Test whether CLR assumptions are fulfilled

• Model with latin, sav, pop, urban, real
• Heteroscedasticity hettest, whitetst
• Correlation matrix correlate varlist
• Ramsey RESET ovtest
• Results
• Homoscedastic disturbances
• No multicollinearity
• OV or non-linearity

19
How to deal with non-linearity

• Generate squared explanatory variables
• Command gen poppoppop etc.
• Include these squared variables
• Test the model down using LR tests
• We end up with latin, sav, pop, pop2, urban,
  urban2, real2

20
Conclusion

• Stick to a specific model selection methodology
• Sometimes theory offers good specifications
• Try to interpret the model
• Make illustrations of the non-linear impacts
  use Excel!
• Improvement Panel data analysis, causality,
  dynamics etc.

21
Impact of population growth
22
Impact of urbanization