Exercises for Beck Note: All exercises use STATA. It is not hard to just type commands to get stata to produce results; the trick is to be able to understand and interpret the results. You can make sure you understand by actually doing the interpretation or calculating what is asked for. Please do not get bogged down in the mechanics of stata. If after a reasonable time you can't get stata to do what you want, ask for help! TSCS Use Geoff Garrett's data (from Partisan Politics in the Global Era), garrum6.dta. It should be pretty clear what the vars are, or you can read his section 5.3. (The vars are briefly described in the stata file, type desc in stata.) Note that it is already in xt and ts format. (The Icc are country dummies, the perxxyy are dummies for the period 19xx-yy (one year is omitted so no multicolinearity, the key variables are leftlab (the political power of left parties in the govt), corp (a measure of corporatism or how encompassing the labor movement is and cl_int which is the product of the two - the first two vars are indices which run form 0 to 5 and 0 to 4 respectively. The dep vars are either unem(ployment), infl(ation) or gdp(growth) with lags being indicated by a suffixed "l". oild and oecd_dem are controls for oil dependency (and prices) and the growth rate in all oecd countries. Data are for 14 oecd countries from 1966-90. The data are already in xt form. Use stata xt commands) The country codes are as follows 2 US 20 Canada 200 UK 210 Netherlands 211 Belgium 220 France 260 Germany 305 Austria 325 Italy 375 Finland 380 Sweden 385 Norway 390 Denmark 740 Japan Choose unem, infl OR gdp as your dep var. Use a specification similar to what is in Garrett's book or what was done in class. Estimate fixed and random effects models (xtreg with appropriate subcommands). Use regress to estimate a model with no effects, and then use regress to estimate a model adding fixed effects "by hand." (make sure your model includes the lagged dependent variable) Compare the fixed no effects models. Test for fixed effects. Look for things like regional differences by either examining effects by country or by estimating your model on subset of data. Compare straight OLS to OLS with PCSEs to to full FGLS (never do again!!)\. Command should be xtgls y ivs. panels(correlated) Dynmics Using the model you have estimated, compare what happens when you specify no dynamics (no lagged dep var, no serial correlation) Estimate the model via OLS (do not worry about PCSEs) with and without the ldv and without but correcting for AR! errors. How do the results differ? Now estimate the full ADL model (lagged y, current x's, lagged x's). What does this tell you?