Chapter 2: Simple Regression – Computer exercise
6.
(i) Each additional dollar spent is unlikely to have the same effect on the pass rate. A
diminishing effect is more reasonable because extra spending is more effective at low levels
than at high levels. Using log(expend) captures this diminishing return.
(ii) In the model, a 10% increase in spending implies Δlog (expend )≈ 0.10.
Therefore, β 1 /10is the percentage point change in math10 from a 10% increase in spending.
(iii) Estimate Std. Error t value Pr(>|t|)
(Intercept) -69.341 26.530 -2.614 0.009290 **
log(expend) 11.164 3.169 3.523 0.000475 ***
---
Multiple R-squared: 0.02966, Adjusted R-squared: 0.02727
F-statistic: 12.41 on 1 and 406 DF, p-value: 0.0004752
Estimate equation: ^
math=−69.34+11.16 log (expend ), n = 408, R2 = 0.03
(iv) The coefficient on expand is 11.16. A 10% increase in spending increases the math pass
rate by about 11.16/10 = 1.116 percentage points.
(v) Range of fitted value: from 21.21697 to 30.15376
This is not a serious concern because, within the observed range of spending, the fitted values
of math10 are well below 100.
Chapter 3: Multiple Regression – Computer exercise
1.
(i) β 1 is expected to be positive because higher family income allows better prenatal care and
nutrition, which increase birth weight.
(ii) cigs and faminc are likely to be negatively correlated because women from higher-income
families tend to smoke less during pregnancy. However, the correlation could be weak or
positive if smoking habits are driven by other social factors.
(iii) Estimate Std. Error t value Pr(>|t|)
(Intercept) 119.77190 0.57234 209.267 < 2e-16 ***
cigs -0.51377 0.09049 -5.678 1.66e-08 ***
---
Multiple R-squared: 0.02273, Adjusted R-squared: 0.02202
F-statistic: 32.24 on 1 and 1386 DF, p-value: 1.662e-08
Estimate equation: ^
bwght =119.77−0.51 cigs ,n = 1388, R2 = 0.023
Estimate Std. Error t value Pr(>|t|)
(Intercept) 116.97413 1.04898 111.512 < 2e-16 ***
cigs -0.46341 0.09158 -5.060 4.75e-07 ***
faminc 0.09276 0.02919 3.178 0.00151 **
---
Multiple R-squared: 0.0298, Adjusted R-squared: 0.0284
F-statistic: 21.27 on 2 and 1385 DF, p-value: 7.942e-10
Estimate equation: ^
bwght =116.97−0.46 cigs+ 0.09 faminc ,
n = 1388, R2 = 0.0298
Discussion: Adding faminc slightly reduces the magnitude of the coefficient on cigs, but the
effect remains negative and sizable. This suggests smoking has a robust negative effect on
birth weight, even after controlling for income. When faminc is added to the model, the
adjusted R2increases (0.0298 > 0.023), indicating that family income helps explain additional
variation in birth weight. This suggests that including faminc is appropriate and improves the
model.
2.
(i) Estimate Std. Error t value Pr(>|t|)
(Intercept) -19.31500 31.04662 -0.622 0.536
sqrft 0.12844 0.01382 9.291 1.39e-14 ***
bdrms 15.19819 9.48352 1.603 0.113
---
Multiple R-squared: 0.6319, Adjusted R-squared: 0.6233
F-statistic: 72.96 on 2 and 85 DF, p-value: < 2.2e-16
The OLS estimate: ^
price=−19.315+0.128 sqrft +15.20 bdrms
2
n=88 , R =0.632 .
(ii) The coefficient on bdrms is 15.2. Holding square footage constant, one more bedroom
increases price by $15,200.
(iii) The estimated increase in price for a house with an additional bedroom that is
140 square feet in size is 15.2 + 0.128 × 140 = 33.12 (thousand dollars) = $33,120.
This is much larger than in part (ii).
(iv) R2=0.632 means that about 63.2% of the variation in house prices is explained by square
footage and number of bedrooms.
(v) The predicted selling price for this house from the OLS regression line when sqrft = 2,438
and bdrms = 4 is $354,605
(vi) The actual selling price of the first house in the sample was $300,000. Therefore,the
residual is $300,000 - $354,605 = -$54,605, meaning the house sold for less than predicted.
This suggests the buyer underpaid for the house.
3.
(i) Estimate Std. Error t value Pr(>|t|)
(Intercept) 4.62092 0.25441 18.163 < 2e-16 ***
log(sales) 0.16213 0.03967 4.087 6.67e-05 ***
log(mktval) 0.10671 0.05012 2.129 0.0347 *
---
Multiple R-squared: 0.2991, Adjusted R-squared: 0.2911
F-statistic: 37.13 on 2 and 174 DF, p-value: 3.727e-14
Model: log (salary )=β 0 + β 1 log (sales)+ β2 log ( mktval)+u
Estimated equation: ^
log ( salary)=4.62+0.162 log (sales)+0.102 log (mktval)
2
n=177 , R =0.299 .
(ii) Estimate Std. Error t value Pr(>|t|)
(Intercept) 4.687e+00 3.797e-01 12.343 < 2e-16 ***
log(sales) 1.614e-01 3.991e-02 4.043 7.92e-05 ***
log(mktval) 9.753e-02 6.369e-02 1.531 0.128
profits 3.566e-05 1.520e-04 0.235 0.815
---
Multiple R-squared: 0.2993, Adjusted R-squared: 0.2872
F-statistic: 24.64 on 3 and 173 DF, p-value: 2.53e-13
Estimated equation:
^
log ( salary)=4.60+0.161 log (sales)+ 0.098 log (mktval)+0.00005 profits
2
n=177 , R ≈ 0.2993 .
Profits can be zero or negative, so taking logs is not possible. Firm performance variables
explain some, but not most, of the variation in CEO salaries, as the R2is around 0.30.
(iii) Estimate Std. Error t value Pr(>|t|)
(Intercept) 4.558e+00 3.803e-01 11.986 < 2e-16 ***
log(sales) 1.622e-01 3.948e-02 4.109 6.14e-05 ***
log(mktval) 1.018e-01 6.303e-02 1.614 0.1083
profits 2.905e-05 1.503e-04 0.193 0.8470
ceoten 1.168e-02 5.342e-03 2.187 0.0301 *
---
Multiple R-squared: 0.3183, Adjusted R-squared: 0.3024
F-statistic: 20.08 on 4 and 172 DF, p-value: 1.387e-13
The coefficient on ceoten is 0.017 means that another year of CEO tenure increases salary by
about 1.7%, holding other factors fixed.
(iv) The sample correlation coefficient between the variables log(mktval) and profits is 0.78.
The variables are highly correlated, which suggests potential multicollinearity. This may
inflate standard errors but does not bias OLS coefficients.