Chapter 7

Problem Question 1

Part (i)

The coefficient for the gender dummy variable "male" is 87.75. This indicates that, all else being equal
(i.e., when other variables are held constant), men sleep 87.75 minutes more per week than women.

To determine how strong the evidence is, we can look at the standard error associated with the
coefficient. The standard error for the "male" coefficient is 34.33. Typically, a coefficient is considered
statistically significant if it's more than about 2 times its standard error. In this case, the coefficient 87.75
is more than 2 times 34.33, which suggests that there is statistically significant evidence to support the
claim that men sleep more than women, all other factors being equal.

Part (ii)

The coefficient for the variable "totwrk" (total weekly minutes spent working) is -0.163. This means that
for every additional minute worked in a week, the total sleep decreases by 0.163 minutes or roughly
0.163/60 = 0.00272 hours.

To determine if this tradeoff is statistically significant, we check the standard error associated with the
coefficient of "totwrk", which is 0.018.

A rule of thumb is that a coefficient is considered statistically significant if it's more than about 2 times its
standard error. In this case:

|0.163| > 2(0.018)

Thus, there is a statistically significant tradeoff between working and sleeping. Specifically, for every
additional minute worked per week, sleep is reduced by approximately 0.163 minutes.

Part (iii)

To obtain , the R-squared from the restricted regression, we need to estimate the model without age
and agesq. When age and age 2 are both in the model, age has no effect only if the parameters on
both terms are zero.

To test the null hypothesisH0:β3=β4=0, you would estimate the following restricted regression:

sleep=β0+β1totwrk+β2educ+β5male+ϵ

After running both regressions (original and restricted), you would perform an F-test to compare the sum
of squared residuals (SSR) from the restricted model to the SSR from the original model. If the F-test
indicates a significant difference in the fit of the two models, you would reject the null hypothesis and
conclude that age does have a significant effect on sleeping. If not, you would fail to reject the null
hypothesis, suggesting that age does not have a significant effect on sleeping when other factors are
held constant.

Problem Question 2

Part (i)
This means that for each additional cigarette smoked per day, the logarithm of the birth weight
(log(bwght)) decreases by 0.0044 units, on average, holding all other factors constant.

To determine the effect on birth weight from smoking 10 more cigarettes per day: Effect = Coefficient ×
Change in cigs Effect = -0.0044 × 10 = -0.044

Thus, smoking 10 more cigarettes per day is associated with a decrease of 0.044 units in the logarithm of
the birth weight, holding all other factors constant.

If Δcigs = 10 then = −.0044(10) = −.044, which means about a 4.4% lower birth weight.

Part (ii)

In the first equation, the coefficient for the variable white is 0.055.

This means that, holding all other factors constant, the logarithm of the birth weight (log(bwght)) for a
white child is predicted to be 0.055 units higher than for a nonwhite child.

For the statistical significance, we look at the standard error associated with the coefficient, which is
(0.013). Generally, a coefficient is considered statistically significant if it's more than about 2 times its
standard error (using the rough rule for a 95% confidence interval).

Coefficient/Standard Error = 0.055/0.013 ≈ 4.23

Since 4.23 is greater than 2, the difference is statistically significant at the 5% level. This means the
difference in birth weight between white and nonwhite children, as predicted by this model, is
statistically significant.

A white child is estimated to weigh about 5.5% more, other factors in the first equation fixed. Further,
twhite ≈ 4.23, which is well above any commonly used critical value. Thus, the difference between
white and nonwhite babies is also statistically significant.

Part (iii)

The variable motheduc does not appear in the first equation you provided; it is present in the second
equation. In that equation, the coefficient for motheduc is -0.0030, with a standard error of (0.0030).

The estimated effect of -0.0030 indicates that, holding all other factors constant, a one-unit increase in
motheduc is associated with a decrease in the logarithm of the birth weight (log(bwght)) by 0.0030 units.

For the statistical significance, we again compare the coefficient to its standard error:

Coefficient/Standard Error = -0.0030/0.0030 = -1

Considering the rough rule for a 95% confidence interval, where a coefficient is statistically significant if
it's absolute value is more than about 2 times its standard error, the value of -1 is less than 2 in absolute
terms. Thus, based on this rule, the coefficient for motheduc is not statistically significant at the 5% level.
This means that, according to this model, the mother's education (motheduc) does not have a
statistically significant effect on birth weight when considering the other variables in the model.

Part (iv)
From the given information, we are unable to compute the F statistic for joint significance of motheduc
and fathedrec because we lack the necessary summary statistics from the regression output. Specifically,
we need:

1. The residual sum of squares (RSS) from a restricted model (a model without motheduc and
fathedrec).

2. The RSS from the unrestricted model (the model with motheduc and fathedrec).

3. The number of restrictions (which is 2, since we are testing two variables).

4. The total number of observations.

To compute the �F statistic, the formula is:

Where:

 RSS restricted is the residual sum of squares for the model without the variables in question.

 RSS unrestricted is the residual sum of squares for the model with the variables.

 q is the number of restrictions (2 in this case for motheduc and fathedrec).

 n is the number of observations.

 k is the number of parameters (including the intercept) in the unrestricted model.

So, to compute the F statistic for joint significance of motheduc and fathedrec, we would need the RSS
for both the restricted and unrestricted models and the total number of observations.

Problem Question 3

Part (i)

1. The coefficient for hsize2 is -2.19 and the standard error is 0.53. To determine if there is strong
evidence for the inclusion of hsize2 in the model, we can use the t-statistic:

This yields a t-value of approximately -4.13. Typically, in hypothesis testing, a t-value with an absolute

much larger than 2, which indicates that there's strong evidence that ℎ����2hsize2 is statistically
value greater than 2 is considered statistically significant at the 5% level. Here, the absolute t-value is

significant and should be included in the model.

2. Optimal high school size: To determine the optimal high school size, we need to find the value of
hsize for which the first derivative of the equation with respect to hsize is zero (i.e., the
maximum or minimum point). The equation provided is a quadratic function with respect to
hsize, and its first derivative will give us the slope.

Given:
sat=1,028.10+19.30hsize−2.19hsize2

The first derivative with respect to hsize is:

Part (ii)

To determine the estimated difference in SAT scores between nonblack females and nonblack males,
holding hsize fixed, we need to consider the coefficients of the relevant variables.

The estimated difference between nonblack females and nonblack males, when holding hsize fixed, is:
−45.09.

This indicates that, according to the model, nonblack females are predicted to score 45.09 points lower
than nonblack males on the SAT, all else being equal.

To determine how statistically significant this difference is, we examine the coefficient's standard error,
which is 4.29 for the 'female' variable.

This yields a t-value of approximately -10.51.

Given the magnitude of this t-value, it's far beyond the typical thresholds (like 1.96 for the 5%
significance level in a two-tailed test), indicating that the difference is highly statistically significant. Thus,
there's strong evidence that nonblack females and nonblack males have significantly different SAT scores
based on this model.

Part (iii)

To determine the estimated difference in SAT scores between nonblack males and black males, holding
hsize fixed, we need to consider the coefficients of the relevant variables.
The estimated difference between black males and nonblack males, when holding hsize fixed, is:

−169.81

This indicates that, according to the model, black males are predicted to score 169.81 points lower than
nonblack males on the SAT, all else being equal.

Now, to test the null hypothesis:

To determine how statistically significant this difference is, we examine the coefficient's standard error,
which is 12.71 for the 'black' variable.

This yields a t-value of approximately -13.35.

Given the magnitude of this t-value, it's far beyond the typical thresholds (like 1.96 for the 5%
significance level in a two-tailed test), indicating that the difference is highly statistically significant. Thus,
there's strong evidence to reject the null hypothesis in favor of the alternative that there is a significant
difference in SAT scores between black males and nonblack males based on this model.

Part (iv)

To determine the estimated difference in SAT scores between black females and nonblack females,
holding hsize fixed, we need to consider the coefficients of the relevant variables.

The difference between the two, when considering the effect of being black and female is:
−169.81+62.31=−107.5
This indicates that, according to the model, black females are predicted to score 107.5 points lower than
nonblack females on the SAT, all else being equal.

Problem Question 4

Part (i)

The coefficient for the utility industry is -0.283. Since transportation is the omitted category, its
coefficient is essentially 0.

The difference in the log of the estimated salary between the utility and transportation industries is:
Difference = -0.283 - 0 = -0.283

2. Convert Log Difference to Percentage Difference

For small changes, the percentage difference for a variable y with respect to a small change in x is
approximately 100 times the change in the log of y.

Percentage Difference = 100 * (-0.283) = -28.3%

This implies that, holding sales and roe fixed, CEOs in the utility industry earn approximately 28.3% less
than those in the transportation industry.

3. Test for Statistical Significance

The standard error for the utility coefficient is 0.099.

To determine if the difference is statistically significant at the 1% level, we'll calculate the t-statistic: t-
statistic = coefficient / standard error t-statistic = -0.283 / 0.099 = -2.86

You would compare this t-statistic to the critical value from a t-distribution with a significance level of 1%
(for a two-tailed test). For a large sample size like n=209, the critical t-value for a two-tailed test at the
1% level is approximately 2.58.

Given that 2.86 > 2.58 (in terms of absolute value), the difference is statistically significant at the 1%
level.

Conclusion: Holding sales and roe fixed, the salary for CEOs in the utility industry is approximately 28.3%
less than those in the transportation industry, and this difference is statistically significant at the 1%
level.

Part (ii)

Given equation 7.10

Percentage difference=100×[exp(β1)−1]

Where: β1 is the coefficient of the industry variable we're interested in.

From the previous information: β1 for the utility industry is −0.283.

Plug this value into equation 7.10: Percentage difference= 100×[exp(−0.283)−1]

Now, let's compute it.

This means that, based on the equation provided, CEOs in the utility industry earn approximately 24.66%
less than those in the transportation industry, when holding sales and roe fixed.

Part (iii)

Given the values for the coefficients: βf=0.158 (for finance)

βcp=0.181 (for consumer products)

You can compute the approximate percentage difference in estimated salary between the consumer
products and finance industries using the formula provided:

Percentage difference=100×[exp(βcp−βf)−1]

Plugging in the values:

Percentage difference=100×[exp(0.181−0.158)−1]

Percentage difference=100×[exp(0.023)−1]

Percentage difference≈2.32

So, the estimated salary in the consumer products industry is approximately 2.32% higher than in the
finance industry, based on the given coefficients.