Scribd
Upload
127 views4 pages

Omitted Variable Tests

1. A regression of hourly pay on work experience, tenure, and education found a significant relationship but omitted variable tests detected missing variables. 2. Adding higher order terms of the predictors (squared terms) improved the model fit and removed evidence of omitted variables. 3. Further improving the model by adding a dummy variable for urban location, which proved significant, correctly specifying the wage equation. 4. The RESET test only checks for nonlinear relationships with original predictors, not additional missing variables, so could not detect the influence of urban location variable.

Uploaded by

Andrew Tandoh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
127 views4 pages

Omitted Variable Tests

1. A regression of hourly pay on work experience, tenure, and education found a significant relationship but omitted variable tests detected missing variables. 2. Adding higher order terms of the predictors (squared terms) improved the model fit and removed evidence of omitted variables. 3. Further improving the model by adding a dummy variable for urban location, which proved significant, correctly specifying the wage equation. 4. The RESET test only checks for nonlinear relationships with original predictors, not additional missing variables, so could not detect the influence of urban location variable.

Uploaded by

Andrew Tandoh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Omitted Variable Tests

The data set nlsy.dta contains information on 252 women in work in the United States

A regression of the log of hourly pay on years of work experience (exper), years at the
current firm (tenure) and years of education (educ) gives the following
. reg lwage exper tenure educ if female==1

Source | SS df MS Number of obs = 252


-------------+------------------------------ F( 3, 248) = 22.23
Model | 10.4983092 3 3.49943639 Prob > F = 0.0000
Residual | 39.0353026 248 .157400414 R-squared = 0.2119
-------------+------------------------------ Adj R-squared = 0.2024
Total | 49.5336118 251 .197345067 Root MSE = .39674
------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0022675 .0021046 1.08 0.282 -.0018776 .0064127
tenure | .0102745 .0052232 1.97 0.050 -.000013 .020562
educ | .0800361 .0104624 7.65 0.000 .0594296 .1006426
_cons | .3561163 .1410633 2.52 0.012 .0782814 .6339512

Stata’s version of the Ramsey RESET test gives


. ovtest

Ramsey RESET test using powers of the fitted values of lwage


Ho: model has no omitted variables
F(3, 245) = 6.51
Prob > F = 0.0003

The equivalent manual version with 3 powers of the predicted variable


predict yhat
g yhat2=yhat^2
g yhat3=yhat^3
g yhat4=yhat^4

reg lwage exper tenure educ yhat2 yhat3 yhat4 if female==1


Source | SS df MS Number of obs = 252
-------------+------------------------------ F( 6, 245) = 15.11
Model | 13.3791313 6 2.22985522 Prob > F = 0.0000
Residual | 36.1544805 245 .147569308 R-squared = 0.2701
-------------+------------------------------ Adj R-squared = 0.2522
Total | 49.5336118 251 .197345067 Root MSE = .38415
------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0076626 .0274456 0.28 0.780 -.0463968 .0617221
tenure | .0326196 .1236523 0.26 0.792 -.2109375 .2761768
educ | .2807438 .9637558 0.29 0.771 -1.61756 2.179048
yhat2 | -6.316156 16.1629 -0.39 0.696 -38.15213 25.51982
yhat3 | 4.110669 9.099808 0.45 0.652 -13.81317 22.03451
yhat4 | -.8021463 1.838969 -0.44 0.663 -4.424352 2.820059
_cons | 1.877042 1.30147 1.44 0.151 -.6864544 4.440539

. test yhat2=yhat3=yhat4=0

( 1) yhat2 - yhat3 = 0
( 2) yhat2 - yhat4 = 0
( 3) yhat2 = 0

F( 3, 245) = 6.51
Prob > F = 0.0003

Similarly the equivalent LM Test


reg uhat exper tenure
predict uhat if e(sample), resid

. reg uhat exper tenure educ yhat2 yhat3 yhat4

Source | SS df MS Number of obs = 252


-------------+------------------------------ F( 6, 245) = 6.70
Model | 5.49871214 6 .916452023 Prob > F = 0.0000
Residual | 33.5365905 245 .136884043 R-squared = 0.1409
-------------+------------------------------ Adj R-squared = 0.1198
Total | 39.0353027 251 .155519134 Root MSE = .36998
------------------------------------------------------------------------------
uhat | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | -.0022505 .0020113 -1.12 0.264 -.0062121 .001711
tenure | -.0092757 .0051921 -1.79 0.075 -.0195026 .0009511
educ | -.078739 .0163717 -4.81 0.000 -.1109862 -.0464918
yhat2 | 2.175833 1.807114 1.20 0.230 -1.383628 5.735294
yhat3 | -1.588593 1.517461 -1.05 0.296 -4.577527 1.400341
yhat4 | .3727902 .3515099 1.06 0.290 -.3195767 1.065157
_cons | -.3174041 .7014818 -0.45 0.651 -1.699109 1.0643
------------------------------------------------------------------------------

a
and NR 2 ~ χ (2No.ofrestrictions ) = 252*0.1409 = 35.5 ~χ2(3) > 7.85 (95% critical value)
So decide to add higher order powers of the right hand side variables (in this case the
squares of experience tenure & education) [in truth the significance of the test is partly the
result of the large standard errors on the yhat variables partly the result of multicolinearity]
. reg lwage exper tenure educ exper2 ten2 ed2 if female==1

Source | SS df MS Number of obs = 252


-------------+------------------------------ F( 6, 245) = 18.84
Model | 15.6378395 6 2.60630658 Prob > F = 0.0000
Residual | 33.8957723 245 .138350091 R-squared = 0.3157
-------------+------------------------------ Adj R-squared = 0.2989
Total | 49.5336118 251 .197345067 Root MSE = .37195
------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0158065 .0064825 2.44 0.015 .003038 .0285751
tenure | .0368875 .0112897 3.27 0.001 .0146503 .0591248
educ | -.0897278 .0375597 -2.39 0.018 -.163709 -.0157466
exper2 | -.000335 .0001412 -2.37 0.018 -.0006132 -.0000569
ten2 | -.0012862 .0004836 -2.66 0.008 -.0022388 -.0003336
ed2 | .0072476 .001617 4.48 0.000 .0040626 .0104327
_cons | 1.191018 .2359601 5.05 0.000 .7262488 1.655787

This time the higher order terms are significant and the RESET test suggests no evidence
of functional form misspecification

. ovtest

Ramsey RESET test using powers of the fitted values of lwage


Ho: model has no omitted variables
F(3, 242) = 1.32
Prob > F = 0.2683
However if we add a dummy variable to indicate whether the individual works in an urban
area, the urban dummy variable is positive and significant (there is a wage premium to
working in an urban area)

. reg lwage exper tenure educ exper2 ten2 ed2 urban if female==1

Source | SS df MS Number of obs = 252


-------------+------------------------------ F( 7, 244) = 17.69
Model | 16.6735777 7 2.38193967 Prob > F = 0.0000
Residual | 32.8600341 244 .134672271 R-squared = 0.3366
-------------+------------------------------ Adj R-squared = 0.3176
Total | 49.5336118 251 .197345067 Root MSE = .36698

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0177786 .0064352 2.76 0.006 .005103 .0304541
tenure | .0363432 .0111403 3.26 0.001 .0143997 .0582867
educ | -.101155 .0372855 -2.71 0.007 -.1745975 -.0277124
exper2 | -.0003656 .0001398 -2.62 0.009 -.0006409 -.0000903
ten2 | -.0012701 .0004772 -2.66 0.008 -.00221 -.0003302
ed2 | .0075117 .0015982 4.70 0.000 .0043636 .0106598
urban | .150235 .0541733 2.77 0.006 .043528 .256942
_cons | 1.16322 .2330184 4.99 0.000 .7042357 1.622204
------------------------------------------------------------------------------

Moral: the RESET test is only a test of whether the model is linear in the original
variables. It cannot pick up the influence of other variables

You might also like