You are on page 1of 8

Running Head: INTRODUCTORY ECONOMETRICS

ECON2206: Introductory Econometrics

Student’s Name

Institutional Affiliation

Date
INTRODUCTORY ECONOMETRICS 2

INTRODUCTORY ECONOMETRICS

The file beauty.dta collects information on wages, years of experience and gender for 1,260
individuals. We also have information on a subjective index of beauty (looks) measured from 1 to 5.

lwage = β0 + β1exper + β2exper2 + β3looks + β4female + u (1)

Q1. Estimate the parameter of the econometric model in equation (1) using STATA. What is the
estimated effect of experience on wages?
Answer: The effect of experience on wage is given by β1+2β2exper, which is estimated as
0.0462−2∗0.0008∗exper. For instance one additional year of experience for an individual with
5 years of experience is associated with an increase of 100∗(0.046−2∗0.0008∗5) = 3.8% in
wages.

Q2. What can we say about the effect of an increase of one point in the looks scale on wages? Is this
larger or smaller than the effect of an additional year of experience for an individual with 5
years of experience? Is the difference statistically significant?
Answer: The effect of an increase of one point in the looks scale on wages is β3, estimated at
0.088. This implies a 8.8% increase in wages. As noted above, the effect of an additional year
of experience for an individual with 5 years of experience is β1 +β2∗5, estimated as 3.8%,
smaller than the effect of one point in looks. In order to test whether the difference is significant
we need to specify the null and alternative hypothesis H0 : β3 = β1 + 2β25; H1 : β3 ≠ β1 + 2β25.
This is a test on a linear combination of parameters, as we could equivalently write H0 : θ = 0;
H1 : θ = 0̸ , where θ = β3 − β1 − 2β25. A form test can be easily obtained by reparameterizing the
model to include θ. To this end write β3 = θ + β1 + 2β25 and substitute in the original model:

lwage = β0 + β1exper + β2exper2 + (θ + β1 + 10β2)looks + β4female + u,


= β0 + β1(exper + looks) + β2(exper2 + 10looks) + θlooks + β4female + u.

Now we can test for significance of θ by running the regression of lwage on exper+looks, exper2
+ 10looks and looks.
. gen exper_looks = exper + looks

. gen expersq_10looks = expersq + 10*looks

1
. reg lwage exper_looks expersq_10looks looks female
Source | SS df MS Number of obs = 1,260
-------------+---------------------------------- F(4, 1255) = 120.89
Model | 123.767068 4 30.9417669 Prob > F = 0.0000
Residual | 321.212905 1,255 .255946538 R-squared = 0.2781
-------------+---------------------------------- Adj R-squared = 0.2758
Total | 444.979972 1,259 .353439215 Root MSE = .50591
INTRODUCTORY ECONOMETRICS 3

--------------------------------------------------------------------------------lwage | Coef. Std.


Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
exper_looks | .0462051 .0046515 9.93 0.000 .0370796 .0553307
expersq_10looks | -.0008054 .0001034 -7.79 0.000 -.0010082 -.0006025
looks | .0505996 .0211726 2.39 0.017 .009062 .0921372
female | -.4664837 .030967 -15.06 0.000 -.5272365 -.405731
_cons | 1.078386 .0842657 12.80 0.000 .9130683 1.243703
---------------------------------------------------------------------------------

The parameter associated with looks in this regression is θ. The p-value of the test on
significance of θ is 0.017, therefore we reject the null hypothesis that the effects are equivalent
at the 5% significance level.

Q3. Is there any evidence of functional form misspecification for equation (1)?
Answer: In order to test for functional form mispecification we could consider the RESET test.
We first need to obtain fitted values of the regression model (1) using predict fitted. We then
generate the squared and cubic fitted values and add them to the list of regressors in the original
specification:
. reg lwage exper expersq looks female

. predict fitted
(option xb assumed; fitted values)

. gen fitted2 = fitted^2

. gen fitted3 = fitted^3

. reg lwage exper expersq looks female fitted2 fitted3


Source | SS df MS Number of obs = 1,260
-------------+---------------------------------- F(6, 1253) = 81.21
Model | 124.585707 6 20.7642845 Prob > F = 0.0000
Residual | 320.394266 1,253 .255701728 R-squared = 0.2800
-------------+---------------------------------- Adj R-squared = 0.2765
Total | 444.979972 1,259 .353439215 Root MSE = .50567
-----------------------------------------------------------------------------lwage | Coef. Std. Err. t
P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | -.1887952 .1694564 -1.11 0.265 -.5212447 .1436543
expersq | .0032838 .0029513 1.11 0.266 -.0025062 .0090738
looks | -.3612019 .3228477 -1.12 0.263 -.9945835 .2721797
female | 1.94128 1.738547 1.12 0.264 -1.469505 5.352065
fitted2 | 3.242035 2.482928 1.31 0.192 -1.62912 8.11319
fitted3 | -.6647855 .5414562 -1.23 0.220 -1.727046 .3974752
_cons | -1.874179 2.252184 -0.83 0.405 -6.292646 2.544288
------------------------------------------------------------------------------

An F-test of joint significance of the parameters associated to fitted2 and fitted3 would provide
evidence of functional form misspecification. H0 : β5 = β6 = 0, H1 : otherwise:
INTRODUCTORY ECONOMETRICS 4

,
to be compared to an F distribution with 2 and 1,253 degrees of freedom. The critical value at
the 5% is 3, so we cannot reject the null hypothesis that the functional form is not misspecified.

Q4. How would you test the hypothesis that beauty has the same effect on wages both for males and
females? Provide a p-value for such test. Including an interaction term between looks and
female in the original specification would allow us to estimate the differential effect of looks
on wages between males and females:
. gen looks_female = looks*female

. reg lwage exper expersq looks female looks_female


Source | SS df MS Number of obs = 1,260
-------------+---------------------------------- F(5, 1254) = 96.99
Model | 124.09724 5 24.819448 Prob > F = 0.0000
Residual | 320.882732 1,254 .255887346 R-squared = 0.2789
-------------+---------------------------------- Adj R-squared = 0.2760
Total | 444.979972 1,259 .353439215 Root MSE = .50585
-----------------------------------------------------------------------------lwage | Coef. Std. Err. t
P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0466609 .0046682 10.00 0.000 .0375025 .0558193
expersq | -.0008166 .0001039 -7.86 0.000 -.0010204 -.0006128
looks | .0700875 .0267244 2.62 0.009 .017658 .122517
female | -.622831 .1410797 -4.41 0.000 -.8996093 -.3460528
looks_female | .0489631 .0431046 1.14 0.256 -.0356019 .1335282
_cons | 1.134749 .0977813 11.60 0.000 .9429163 1.326582
------------------------------------------------------------------------------

The difference is estimated as βˆ5 = 0.0489, but it is not statistically significant. In fact the p-
value of the associated significance test is 0.256, providing little evidence to the hypothesis that
beauty has a different effect on wages between males and females.

Q5. Is there any evidence of heteroskedasticity in equation (1)? Provide an heteroskedasticity robust
estimate of the parameters in equation (1).
Answer: We discussed a number of test for heterosckedasticity. An example is the Breusch-
Pagan test which considers the auxiliary regression of the squared residuals from
(1) on the explanatory variables of interest. In this case we have:
. reg lwage exper expersq looks female

. predict res, residuals

. gen res2 = res^2

. reg res2 exper expersq looks female


Source | SS df MS Number of obs = 1,260
-------------+---------------------------------- F(4, 1255) = 1.81
Model | 1.63478954 4 .408697386 Prob > F = 0.1242
Residual | 283.134458 1,255 .225605146 R-squared = 0.0057
-------------+---------------------------------- Adj R-squared = 0.0026
INTRODUCTORY ECONOMETRICS 5

Total | 284.769248 1,259 .226186853 Root MSE = .47498


-----------------------------------------------------------------------------res2 | Coef. Std. Err.
t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0113268 .0043671 2.59 0.010 .0027592 .0198944
expersq | -.000225 .0000971 -2.32 0.021 -.0004155 -.0000345
looks | .0088438 .0197905 0.45 0.655 -.0299823 .0476699
female | .0159496 .0290736 0.55 0.583 -.0410886 .0729878
_cons | .1217781 .0791135 1.54 0.124 -.0334313 .2769875
------------------------------------------------------------------------------

An F-test of joint significance of all the coefficients in this regression provides evidence on the
potential heterosckedasticity of the errors. In this csase the STATA output provides the value
of the F-statistic 1.81 and the associated p-value 0.1242, pointing to very limited evidence
towards heterosckedasticity, i.e. we would not reject the null hypothesis of homosckedastic
errors even at the 10% significance level. Still, an heterosckedasticity robust estimator is
obtained as
. reg lwage exper expersq looks female, robust
Linear regression Number of obs = 1,260
F(4, 1255) = 120.28
Prob > F = 0.0000
R-squared = 0.2781
Root MSE = .50591
------------------------------------------------------------------------------
| Robust
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exper | .0462051 .0044223 10.45 0.000 .0375293 .054881
expersq | -.0008054 .0000967 -8.33 0.000 -.0009952 -.0006156
looks | .0887511 .0216065 4.11 0.000 .0463623 .13114
female | -.4664837 .0310153 -15.04 0.000 -.5273313 -.4056362
_cons | 1.078386 .0855464 12.61 0.000 .9105557 1.246215
------------------------------------------------------------------------------

Problem Set 2 (13 Points Total)

The file approval.dta contains information on 78 months of data during the presidency of George W.
Bush spanning the period between February 2001 and July 2007. In particular we have information
on the approval rate for George W. Bush (approve), the logarithm of consumer price index for food
(lcpifood), the logarithm of the gas price (lrgasprice) as well as several event dummies for the 9/11
terrorist attack (sep11 - equal to one for September 2001 and zero otherwise), Iraq invasion
(iraqinvade - equal to one from March 2003 to June 2003 and zero otherwise) and hurricane Katrina
(katrina - equal to one for September 2005 and zero otherwise).
Consider the following econometric model:

approvet = α0 + α1lcpifoodt + α2lrgaspricet + ut (2)


INTRODUCTORY ECONOMETRICS 6

Q6. Using STATA obtain the OLS estimates for α1 and α2 and interpret the estimated coefficients.
Are they statistically significant? Do they have the sign you expected?
Answer: α1 and α2 are estimated as −111.66 and −34.91 respectively, implying that an increase
of one percent in the food price index and gas prices would decrease the approval rate for
George Bush by 1.11% and 0.48% percentage points respectively. These effects are statistically
significant, with both p-value of the significance tests almost zero. The sign is somewhat
expected with increasing prices being detrimental for government approval rates.

Q7. Explain why it would be safe to include a trend in the analysis. Consider both a linear and a
quadratic trend: which one provides a better fit to the data? (hint: the variable t takes value 1
for the first observation and 78 for the last one)
Answer: Considering a trend is desirable as the estimated effects might be the result of spurious
correlation if the considered time series are characterized by similar trends. We can consider
both a linear and quadratic trend by adding t and t2 to the original
specification
. reg approve lcpifood lrgasprice t
Source | SS df MS Number of obs = 78
-------------+---------------------------------- F(3, 74) = 110.80
Model | 13803.16 3 4601.05333 Prob > F = 0.0000
Residual | 3072.9882 74 41.5268675 R-squared = 0.8179
-------------+---------------------------------- Adj R-squared = 0.8105
Total | 16876.1482 77 219.170756 Root MSE = 6.4441
-----------------------------------------------------------------------------approve | Coef. Std. Err. t
P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------lcpifood | 46.2275
148.4313 0.31 0.756 -249.5284 341.9834 lrgasprice | -34.73899 6.364566 -5.46
0.000 -47.42066 -22.05731 t | -.3410848 .3133622 -1.09 0.280 -.9654726 .283303
_cons | -16.13092 759.0111 -0.02 0.983 -1528.493 1496.232
------------------------------------------------------------------------------

. gen t2 = t^2

. reg approve lcpifood lrgasprice t t2


Source | SS df MS Number of obs = 78
-------------+---------------------------------- F(4, 73) = 82.31
Model | 13813.4107 4 3453.35268 Prob > F = 0.0000
Residual | 3062.73746 73 41.9553077 R-squared = 0.8185
-------------+---------------------------------- Adj R-squared = 0.8086
Total | 16876.1482 77 219.170756 Root MSE = 6.4773
-----------------------------------------------------------------------------approve | Coef. Std. Err. t
P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lcpifood | 92.03591 175.6353 0.52 0.602 -258.0048 442.0766
lrgasprice | -33.97001 6.583757 -5.16 0.000 -47.09142 -20.8486
t | -.3680589 .319667 -1.15 0.253 -1.005154 .2690365
t2 | -.0009817 .001986 -0.49 0.623 -.0049399 .0029765
_cons | -255.8154 903.9765 -0.28 0.778 -2057.438 1545.807
------------------------------------------------------------------------------
INTRODUCTORY ECONOMETRICS 7

Both specifications do not point to significant effects due to either linear or quadratic trends,
as the trend coefficients are not statistically significant, with p-values always above any
sensible choice of significance level.

Q8. Is there any evidence of the seasonality? We can test for the presence of seasonality by including
dummies for month in the original specification (with or without trend).
. reg approve lcpifood lrgasprice m1 m2 m3 m4 m5 m6 m7 m8 m9 m10 m11
Source | SS df MS Number of obs = 78
-------------+---------------------------------- F(13, 64) = 25.41
Model | 14137.4045 13 1087.49265 Prob > F = 0.0000
Residual | 2738.74369 64 42.7928701 R-squared = 0.8377
-------------+---------------------------------- Adj R-squared = 0.8048
Total | 16876.1482 77 219.170756 Root MSE = 6.5416
-----------------------------------------------------------------------------approve | Coef. Std. Err. t
P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lcpifood | -69.80675 37.98874 -1.84 0.071 -145.698 6.084488
lrgasprice | -44.40966 7.962119 -5.58 0.000 -60.31582 -28.5035
m1 | -1.575608 3.777489 -0.42 0.678 -9.12201 5.970794
m2 | -2.516389 3.668624 -0.69 0.495 -9.845308 4.812529
m3 | -.580534 3.735828 -0.16 0.877 -8.043708 6.88264
m4 | 4.148441 3.894985 1.07 0.291 -3.632686 11.92957
m5 | 3.227651 3.965506 0.81 0.419 -4.694358 11.14966
m6 | 2.106702 3.888648 0.54 0.590 -5.661764 9.875169
m7 | 1.362126 3.83089 0.36 0.723 -6.290957 9.015208
m8 | .8485692 4.001053 0.21 0.833 -7.144453 8.841592
m9 | 6.639915 4.036373 1.65 0.105 -1.423667 14.7035
m10 | 2.599934 3.877502 0.67 0.505 -5.146267 10.34613
m11 | -.5131445 3.800689 -0.14 0.893 -8.105894 7.079605
_cons | 619.241 166.7822 3.71 0.000 286.0553 952.4268
------------------------------------------------------------------------------

An F-test of joint significance of the coefficients associated to the month dummies would
provide evidence of seasonality. H0 : α3 = α4 = ... = α13 = 0; H1 : otherwise.

,
to be compared to an F distribution with 11 and 64 degrees of freedom. The critical value at the
5% significance level is 1.94, hence we do not reject the null hypothesis that there is no
seasonality at the 5% significance level.

Q9. Do your conclusions in question 6 change when we control for major events (i.e. including
sep11, iraqinvade, katrina)? What was the impact of the 9/11 terrorist attack on Bush’s approval
rate according to this model?
Answer: Controlling for major events we get
. reg approve lcpifood lrgasprice sep11 iraqinvade katrina
Source | SS df MS Number of obs = 78
-------------+---------------------------------- F(5, 72) = 113.52
Model | 14976.4657 5 2995.29314 Prob > F = 0.0000
INTRODUCTORY ECONOMETRICS 8

Residual | 1899.68247 72 26.3844788 R-squared = 0.8874


-------------+---------------------------------- Adj R-squared = 0.8796
Total | 16876.1482 77 219.170756 Root MSE = 5.1366
------------------------------------------------------------------------------
approve | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lcpifood | -77.14759 26.09259 -2.96 0.004 -129.1622 -25.13296
lrgasprice | -37.97949 5.298071 -7.17 0.000 -48.54101 -27.41798
sep11 | 20.12637 3.127197 6.44 0.000 13.89242 26.36032
iraqinvade | 5.972716 3.075878 1.94 0.056 -.1589338 12.10437
katrina | 9.499789 5.403378 1.76 0.083 -1.271648 20.27123
_cons | 628.4726 115.8209 5.43 0.000 397.588 859.3573
------------------------------------------------------------------------------

The effects of the price index and gas price on the approval rate is negative and statistically
significant. The magnitude of the effects is now smaller. Part of the effect is now explained by
these major events. In particular, the effect of the 9/11 terrorist attack increased the approval
rate by α3, which is estimated as 20.126 and strongly significant. This implies that, other things
being equal, the approval rate rose by roughly 20 percentage points around that event.

You might also like