Question

Question 1

First run the regression:

EARNINGSi = β1 + β2ASVABCi + β3Si + ui

Then run the regression with experience:

EARNINGSi = β1 + β2ASVABCi + β3Si + β4EXPi + ui

Compare the results from these two regressions, do you get an indication that the previous estimate of schooling without EXP was biased? If so, in which direction? And why is that?

Question 2

Add gender dummy variable to the regression (the one running regression of EARNINGS on ASVABC, S, and EXP). There are two gender dummies, FEMALE and MALE. You can use either. Run the regression, show your output, and interpret the coefficient of FEMALE (or MALE).

Question 3

Now, test whether we should run two separate regressions, splitting our sample based on gender. You have two ways to do this (performing a Chow test or doing a group F-test). For the group F-test, you will need to create three additional interaction term variables (remem- ber that you need to include a full set of dummies, including slope dummy variables). Show your output and calculations/conclusions corresponding to BOTH tests. Show that you get the same conclusion from both tests.

*** This was to be done with STATA. I added a photo of my log file for refference.

namet cunnamed> log: X:\ECON 2P90\assignment 3.log log type: text opened on: 25 Nov 2019, 16:23:09 • cd "X:\ECON 2P90" X:\ECON 2P90 use X:\ECON 2P90\LFP2011.dta" • regress EARNINGS ASVABC S Source SS df NS Model Residual 52274.8438 726010.375 226137.4219 3.592 202.118701 Number of obs F(2, 3592) Prob R-squared Adj R-squared Root MSE 3.595 129.32 0.0000 0.0672 0.0666 14.217 Total 1 778285.2193 ,594 216.551257 EARNINGS Coef. Std. Err. t Pt 1956 Conf. Interval) ASVABC 1.852604 .8822497 5.752694 285767 6 .0994388 1.4215144 .48 0.000 8.87 0.000 .050.000 1.292322 .6872876 2.9656398 2.412886 1.077212 .539749 cons • regress EARNINGS ASVABC S EXP Source 1 SS af MS - Model Residual 60664.445 717620.774 320221.4817 3,591 199.838701 ----------------- 3,594 216.551257 Number of obs F(3, 3591) Prob F R-squared Adj R-squared Root MSE 3,595 101.19 0.0000 0.0779 0.0772 14.136 Total 778285.219 EARNINGS Coef. Std. Err. Pt 1956 Cont. Intervall ASVABC 1.62344728634325 .67 1.239804 1132333 10.95 ..61425360 948019 6.48 -3.364927 1.994511 -1.69 0.000 0.000 0.000 0.092 1.062036 1.017796 .4283827 -7.275414 2.184859 1.461812 8001245 5 455594 EXP cons . regress EARNINGS ASVABC S EXP MALE Source ss df - Model Residual 70798.563 707486.656 4 17699.6408 3,590 197.071492 Number of obs P(4, 3590) Prob > F R-squared Adj R-squared Root MSE 3.595 89.81 0.0000 0.0910 0.0900 14.038 Total 778285.219 3,594 216.551257 EARNINGS Coef. Std. Err. P t 1950 Conf. Intervall ASVABC EXP MALE cons 1.51315 1.344501 ..5968517 3.40953 -6.5164522 .2847694 1133904 0 941745 .4754594 .028825 5.31 11.86 6.34 7.17 -3.21 0.000 0.000 0.000 0.000 0.001 .9548241 2.071476 1.1221851 .566817 .4122109 7814925 2.4773334.341728 -10.49422 -2.538687 • log close namet cunnamed log: X:\ECON 2P90\assignment 3.log log type: text closed on 25 Nov 2019, 16:44:54

Answer 1

Q 1)

In both regression equation the estimates are significant but in the previous case the variables were not able to explain the dependent variable very well because R-square value was too less but in the later on including the experience as an independent variable the estimates are better and significant also model is able to explain the dependent variable in a better way as the R-square value is increased.

So we can say that without increasing the experience the model would give a biased estimate in the positive direction because estimates were over-estimated.

Q 2)  

In the regression with gender, we have a binary variable with Male and Female as two categories. In the regression analysis, the coefficient of a dummy variable stands for the change in overall mean that is the change in the intercept rather than a change in slope. So the coefficient 3.40953 indicates the higher-earning for the male with the same level of  ASVABC, S, and EXP with the 3.40953 in comparison to females.

Q 3)

Since we including the Gender affect only the intercept, not the slope that much as we can observe from both tables. So we can have a single equation only difference with two intercepts, therefore, we do not need two-equation for males and females.