u panel_lecture . sum

Transcription

1 u panel_lecture sum Variable Obs Mean Std Dev Min Max datastre year total_sa e+07 tot_fixe e+07 emp GENERATES NATURAL LOGS OF THE VARIABLES gen l_out=ln( total_sa) gen l_cap=ln( tot_fixe) gen l_emp=ln(emp) SIMPLE OLS MODEL FOR THE COBB-DOUGLAS PRODUCTION FUNCTION reg l_out l_cap l_emp Source SS df MS Number of obs = F( 2, 636) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = l_out Coef Std Err t P> t [95% Conf Interval] l_cap l_emp _cons COMMAND TO IDENTIFY ENTITIES (THE i PART) AND TIME (THE t PART) sort datastre year iis datastre tis year GENERATING LAGGED VARIABLES BY ENTITY: quietly by datastre: gen l_out_1= l_out[_n-1] sum l_out l_out_1 Variable Obs Mean Std Dev Min Max

2 l_out l_out_ quietly by datastre: gen l_cap_1= l_cap[_n-1] sum l_cap l_cap_1 Variable Obs Mean Std Dev Min Max l_cap l_cap_ quietly by datastre: gen l_emp_1= l_emp[_n-1] sum l_emp l_emp_1 Variable Obs Mean Std Dev Min Max l_emp l_emp_ CREATING THE FIRST-DIFFERENCE VARIABLES: ONE COLUMN SUBTRACTED FROM THE OTHER gen dl_out= l_out- l_out_1 (71 missing values generated) gen dl_cap= l_cap- l_cap_1 (71 missing values generated) gen dl_emp= l_emp- l_emp_1 (71 missing values generated) FIRST-DIFFERENCE REGRESSION (NOTE THE CODE ABOVE TO GENERATE THE FIRST- DIFFERENCED VARIABLES ON LOG OUTPUT, LOG CAPITAL AND LOG EMPLOYMENT reg dl_out dl_cap dl_emp Source SS df MS Number of obs = F( 2, 565) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = dl_out Coef Std Err t P> t [95% Conf Interval] dl_cap dl_emp _cons COMMAND FOR GENERATING TIME DUMMIES tab year, gen(time)

3 INCLUDING TIME DUMMIES IN THE REGRESSION NOTE WE HAVE LOST (TIME1 = 1976 FROM THE FIRST DIFFERENCE reg dl_out dl_cap dl_emp time2- time9 Source SS df MS Number of obs = F( 9, 558) = 3418 Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = dl_out Coef Std Err t P> t [95% Conf Interval] dl_cap dl_emp time time time time time time time time9 (dropped) _cons COMMAND FOR GENERATING FIRM SPECIFIC DUMMY VARIABLES tab datastre, gen(fdum) FIXED EFFECTS REGRESSION (NOTE TO SAVE SPACE, THE FIRM SPECIFIC DUMMIES HAVE BEEN SUPPRESSED - ONLY FIRM2 AND FIRM71 ARE SHOWING OMITTED FIRM: FIRM1) reg l_out l_cap l_emp fdum2- fdum71 Source SS df MS Number of obs = F( 72, 566) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = l_out Coef Std Err t P> t [95% Conf Interval] l_cap l_emp fdum

4 fdum _cons FIXED EFFECTS REGRESSION WITH FIRM AND TIME DUMMIES reg l_out l_cap l_emp fdum2- fdum71 time2- time9 Source SS df MS Number of obs = F( 80, 558) = Model Prob > F = Residual R-squared = Adj R-squared = Total Root MSE = l_out Coef Std Err t P> t [95% Conf Interval] l_cap l_emp fdum fdum time time _cons FIXED EFFECTS (WITHIN GROUPS) REGRESSION xtreg l_out l_cap l_emp, fe Fixed-effects (within) regression sd(u_datastre) = Number of obs = 639 sd(e_datastre_t) = n = 71 sd(e_datastre_t + u_datastre)= T = 9 corr(u_datastre, Xb) = R-sq within = between = overall = F( 2, 566) = Prob > F = l_out Coef Std Err t P> t [95% Conf Interval] l_cap l_emp _cons

5 datastre F(70,566) = (71 categories) RANDOM EFFECTS REGRESSION xtreg l_out l_cap l_emp Random-effects GLS regression sd(u_datastre) = Number of obs = 639 sd(e_datastre_t) = n = 71 sd(e_datastre_t + u_datastre)= T = 9 corr(u_datastre, X) = 0 (assumed) R-sq within = between = overall = chi2( 2) = (theta = 08546) Prob > chi2 = l_out Coef Std Err z P> z [95% Conf Interval] l_cap l_emp _cons HAUSMAN TEST xthaus Hausman specification test ---- Coefficients ---- Fixed Random l_out Effects Effects Difference l_cap l_emp Test: Ho: difference in coefficients not systematic chi2( 2) = (b-b)'[s^(-1)](b-b), S = (S_fe - S_re) = 4581 Prob>chi2 = SAME REGRESSIONS WITH TIME EFFECTS INCLUDED xtreg l_out l_cap l_emp time2- time9, fe Fixed-effects (within) regression sd(u_datastre) = Number of obs = 639 sd(e_datastre_t) = n = 71 sd(e_datastre_t + u_datastre)= T = 9 corr(u_datastre, Xb) = R-sq within = 07879

6 between = overall = F( 10, 558) = Prob > F = l_out Coef Std Err t P> t [95% Conf Interval] l_cap l_emp time time time time time time time time _cons datastre F(70,558) = (71 categories) xtreg l_out l_cap l_emp time2- time9 Random-effects GLS regression sd(u_datastre) = Number of obs = 639 sd(e_datastre_t) = n = 71 sd(e_datastre_t + u_datastre)= T = 9 corr(u_datastre, X) = 0 (assumed) R-sq within = between = overall = chi2( 10) = (theta = 08689) Prob > chi2 = l_out Coef Std Err z P> z [95% Conf Interval] l_cap l_emp time time time time time time time time _cons xthaus

7 Hausman specification test ---- Coefficients ---- Fixed Random l_out Effects Effects Difference l_cap l_emp time time time time time time time time Test: Ho: difference in coefficients not systematic chi2( 10) = (b-b)'[s^(-1)](b-b), S = (S_fe - S_re) = 8559 Prob>chi2 = 00000