Вы находитесь на странице: 1из 7

Solutions 215

Part III Review Solutions

III.1. (a) Use a two-sample z for proportions ( p̂1 is the proportion of the gingko group who believed
they were taking gingko, etc.). (b) Use a two-sample t for means ( x1 is the mean score of the
gingko group, etc.).

III.2. (a) Use a t test for means (we want to examine the mean of responses to the attractiveness
question). (b) A matched-pairs test would be appropriate, because we should keep each couple’s
responses together.

III.3. (a) Use a one-sample z for a proportion (with p̂ as the sample proportion of workdays missed
in Spain). (b) Use a two-sample z for proportions ( p̂1 and p̂2 are the proportions of missed
workdays in Spain and Switzerland, respectively).

III.4. (a) Use a two-sample z for proportions. (b) Two-sample t for means. (c) Two-sample z for
proportions.

III.5. (a) Label the subjects from 01 to 44, and choose 22 subjects to eat regular chips first.
From line 101 of Table B, we choose
19, 22, 39, 34, and 05.
(b) Use a matched-pairs test for means (which is, of course, equivalent to a one-sample t test).

III.6. (a) We were given x = 11.4 and s = 26.09 df t* Confidence interval


minutes, and for df = 100, we have t* = 1.984 100 1.984 0.47581 to 8.92419
134 1.97783 0.48890 to 8.91110
(software gives 1.9748 for df = 161).Therefore,
186.017 1.97280 0.49960 to 8.90040
the 95% confidence interval is
either 11.4 ± 4.067 = 7.333 to 5.467 minutes, or 11.4 ± 4.048 = 7.352 to 15.448 minutes. (b) The
standard error of the difference x f − xm is SE  2.12916. The table on the right shows three
possible choices of df and the intervals that arise from them.

III.7. (a) The standard error of the difference df P-value


x f − xm is SE  2.12916 (this was also 100 0.01478
134 0.01449
found in the previous solution). The test 186.017 0.01425
statistic is t = (11.4 − 6.7) / SE  2.21,
which is significant at the 5% level (see the
table on the right for possible P-values). (b) t
procedures are robust with large samples.

( 0.07 )( 0.93) ( 0.14 )( 0.86 )


III.8. SE = 2455
+
1191
 0.01130; the 99% confidence interval is (0.07 – 0.14) ±
(2.576)(0.01130) = –9.91% to –4.09%. If we assume that there were 172 and 167 black child-care
workers in each group, we can use the plus four method (which gives a similar interval):
p1  0.0704 and p 2  0.1408, SE  0.01132, and the interval is –0.0704 ± 0.0292 = –0.0996 to
–0.0413. Because 0 is not in our 99% confidence interval, we would reject H0: p1 = p2 at the 1%
level.
216 Part III Review

III.9. The standard error of x1 − x2 is df t* Confidence interval


1000 2.581 –0.7944 to –0.4056
SE = s12 / 2455 + s22 /1191  0.07533. The table on 1190 2.580 –0.7943 to –0.4057
2456.81 2.578 –0.7942 to –0.4058
the right shows three possible choices of df, their
associated t* values, and the resulting intervals. Because 0 is not in our 99% confidence interval,
we would reject H0: µ1 = µ2 at the 1% level.
pˆ1 (1 − pˆ 2 ) pˆ 2 (1 − pˆ 2 )
III.10. pˆ1 = 840 ˆ 2 = 1077
775  0.9226, p 680  0.6314, and SE =
840
+
1077
= 0.01735; the 99%
confidence interval is (0.9226 – 0.6314) ± 2.576 SE = 0.2465 to 0.3359. Using the plus four
method: p1 = 0.9216 and p 2 = 0.6311, SE = 0.01737, and the interval is 0.2905 ±
0.0447 = 0.2457 to 0.3352. Because 0 is not in our interval, we would reject H0: p1 = p2 at the 1%
level. (In fact, z = 14.79 and P is practically 0.)

III.11. For the hypotheses H0: µ1 = µ2 vs. Ha: µ1 ≠ µ2, we have t = ( x1 − x2 ) / s12 / 840 + s22 /1077
= −0.8658. Regardless of the df used (100, 839, or 1777.52), this gives a P-value close to 0.4, so
we have no reason to doubt H0.

III.12. We find that pˆ = 904 = 0.7687, SE = 0.0123, and the 95% confidence interval is 0.7687 ±
1176
0.0241 = 0.7446 to 0.7928. Using the plus four method: p = 1180
906 =
 0.7678, SE = 0.0123, and the
plus four interval is 0.7678 ± 0.0241 = 0.7437 to 0.7919.

III.13. We have pˆ1 = 0.7687. (as in the previous exercise), pˆ 2 = 0.8526, and pˆ = 904+1024
1176+1201
= 0.8111. Then SE = 0.01606 and z = (0.7687 – 0.8526)/SE = –5.23, for which P < 0.00005.
We have very strong evidence that the proportion favoring mandatory registration was lower in
2001.
III.14. The appropriate critical value is 2.132 (df = 4), so the 90% confidence
interval is 2.9 ± 2.132 ( ) = 2.9 ± 0.4 = 2.5 to 3.3 ng C per liter.
0.4
5
Note: Watch out for students who might try to make something of the
(coincidental) fact that, after rounding, the margin of error is the same as
the standard deviation.
III.15. With the hypotheses H0: µ1 = µ2 vs. Ha: µ1 ≠ µ2, we compute SE
= 0.42 /5 + 0.42 /5 = 0.253 and t = 2.9− 2.8
0.253
= − 0.395 with df = 4. This
gives P > 0.50; the TI-83 output on the right uses the more accurate df = 8,
and gives P = 0.70. There is no evidence of a difference between the top
melt and bottom melt biomasses.
III.16. To test H0: p = 0.5 vs. Ha: p ≠ 0.5, we compute p̂ = 0.56 and
(0.5)(0.5)
z = (0.56 − 0.5)/ 250
= 1.90. The P-value is 0.0578, which is not quite
significant at the 5% level.
Solution 217

III.17. pˆ = 140
250
= 0.56, SE = 0.0314, and the 95% confidence interval is 0.56 ± 0.0615 = 0.4985 to
0.6215. Using the plus four method, p = 142
254
= 0.5591. SE = 0.0312, = and the plus four interval
is 0.5591 ± 0.0611 = 0.4980 to 0.6201.

III.18. (a) The sample sizes (especially for Europeans) are so large that even a
small difference between Americans and Europeans would be significant.
pˆ1 (1 − pˆ1 ) pˆ 2 (1 − pˆ 2 )
(b) With p̂1 = 0.52 and p̂2 = 0.64, SE = 863
+
12, 178
= 0.01755, so
the 99% confidence interval is (0.52 – 0.64) ± (2.576)(0.01755) = –0.1652
to –0.0748. The TI-83 output on the right gives rough confirmation of this, with slight
differences because 0.52 and 0.64 were entered as 449/863 and 7794/12178. If we use those
counts, we can apply the plus four method: p1 = 0.5202 and p 2 = 0.6400, SE = 0.01753, and the
interval is –0.1198 ± 0.0452 = –0.1649 to –0.0746.

III.19. With pˆ = 0.52, we test H0: p = 0.5 vs. Ha: p > 0.5. The test statistic is
(0.5)(0.5)
z = (0.52 − 0.5)/ 863
= 1.175, so P > 0.10. We cannot conclude that
more than half of Americans hold this opinion. The TI-83 output on the
right gives rough confirmation of this, with slight differences because 0.52
was entered as 449/863.

III.20. p̂ = 0.64, and SE = 0.00435, so the 95% confidence interval is 0.64 ±


1.96 SE = 0.6315 to 0.6485. The TI-83 output on the right gives rough
confirmation of this, with slight differences because 0.64 was entered as
7794/12178. Because the sample size is so large, the plus four method gives
7796 = 0.6400, SE = 0.00435, and the
almost exactly the same interval: p = 12182  
plus four interval is 0.6400 ± 0.0085 = 0.6314 to 0.6485.

III.21. To test H0: µ1 = µ2 vs. Ha: µ1 > µ2, we find that the standard error of the difference x1 − x2 is
SE = 0.078716, and the t statistic is t = (5.15 – 4.33)/SE = 10.4. Regardless of whether df = 19 or
32.39, the P-value is tiny; we have strong evidence that the mean remating time is longer for large
spermatophores.

III.22. (a) This is an observational study (one cannot “assign” a baby’s birthweight). (b) The high
school graduation rates are p̂1 = 0.7397 and pˆ 2 = 0.8283. To test H0: p1 = p2 vs. Ha: p1 < p2, we
compute pˆ = 179+193
242+233
= 0.7832, SE = 0.03782, and z = (0.7397 – 0.8283)/SE = –2.34. The one-
sided P-value is P = 0.0096; this is strong evidence that VLBW graduation rates are lower. (c)
With pˆ = 0.7397, we have SE = 0.0282, and the 95% confidence interval is 0.7397 ± 0.0553 =
0.6844 to 0.7950. Using the plus four method, p = 181
246
= 0.7358, SE = 0.0281, and the plus four
interval is 0.7358 ± 0.0551 = 0.6807 to 0.7909.
218 Part III Review

III.23. To test H0: µ1 = µ2 vs. Ha: µ1 < µ2, we find that the standard error of the difference x1 − x2 is
SE = 2.02786, and the t statistic is t = (87.6 – 94.7)/SE = –3.50. Regardless of whether we take
df = 100, 105, or 216.4, the P-value is less than 0.0005; we have strong evidence that VLBW men
have lower IQs.

III.24. For the proportion using drugs, we test H0: (1 = ) p2 vs. Ha: p1 ≠ p2, and compute
37+52 =
pˆ1 = 0.2937, pˆ 2 = 0.4194, and pˆ = 126+124  0.3560. Then SE = 0.06057, and the test statistic is
z = (0.2937 – 0.4194)/SE = –2.08. This has a two-sided P-value of P = 0.0188—a significant
difference.

For comparing IQs, we test H0: µ1 = µ2 vs. Ha: µ1 ≠ µ2, and compute SE = 1.73370. The test
statistic is t = (86.2 – 89.8)/SE = –2.08. For any choice of df (100, 123, or 247.1), this t has a
two-sided P-value of about 0.04—significant at α = 0.05.

III.25. Both stemplots are reasonably symmetrical, though the nitrite group Nitrites Control
may be slightly left-skewed. There are no extreme outliers. The summary 9 4
5
statistics are 5 5 5
22 6 4
n x s 8655 6 56
Nitrites 30 7807.30 1235.20 32 7 01124
Control 30 8072.93 1068.32 97666 7 6888
322110 8 011123
87555 8 5678
To test H0: µ1 = µ2 vs. Ha: µ1 < µ2, we find SE = 298.16 and t = –0.89. 44 9 00013
8 9 9
With either df = 29 or df = 56.8, P > 0.15; there is little evidence that
2 10 3
nitrites decrease amino acid uptake.

III.26. (a) “s. e.” stands for standard error (of Calories Alcohol
the mean). To find s, multiply by n. (b) n x s x s
No: SE = 65.1153 and t = –0.35, so P > 0.50 Drivers 98 2821 435.58 0.24 0.59397
using either df = 82 or df = 173.9. There is Conductors 83 2844 437.30 0.39 1.00215

little evidence that calorie consumption is different. (c) (0.24 – 0.39) ± t*(0.1253) = –0.31 to 0.01
g/day (regardless of the value of t* used; all choices of df give t* close to 1.292).

III.27. (a) Not very significant: SE = 0.1253, t = –1.20, so P > 0.20 using either df = 82 or df =
128.4. (b) 0.39 ± t*(0.11) = 0.207 to 0.573 grams—whether we use t* = 1.664 (df = 80) or t* =
1.6636 (df = 82).

III.28. No: You have information about all California counties (not just a sample).

III.29. We test H0: (1 = ) p2 vs. Ha: p1 ≠ p2, and compute pˆ1 = 54 = 0.8438 and pˆ 2 =
64 
4
32
= 0.1250,
pˆ = 54+4
64+32
= 0.6042, SE = 0.10588, and z = (0.8438 – 0.1250)/SE = 6.79. The P-value is tiny, so
we have strong evidence that walking and resting flies respond differently,
Solution 219

III.30. (a) To test H0: µ1 = µ2 vs. Ha: µ1 > µ2, we find SE = 16.19 mg/dl and t = 1.17, so 0.10 < P <
0.15 using either df = 22 or df = 43.3. This is not enough evidence to reject H0; we cannot
conclude that pet cholesterol levels are higher. (b) The interval is (193 – 174) ± t*SE. Using t* =
2.074 (df = 22), this gives –14.57 to 52.57 mg/dl; using t* = 2.016 (df = 43.3), the interval is –
13.64 to 51.64 mg/dl. (c) With df = 25, the interval is 193 ± (2.060)(68/ 26 ) = 165.53 to 220.47
mg/dl. (d) We are assuming that we have two SRSs from each population, and that underlying
distributions are normal. It is unlikely that we have random samples from either population,
especially among pets.
III.31. The stemplot shows the distribution to be fairly symmetric, with a slightly 48 8
low outlier of 4.88. There is nothing to keep us from using the t procedure. 49
50 7
x = 5.4479 and s = 0.2209; 5.4479 serves as our best estimate of the earth’s
51 0
density, with margin of error t*s/ 29. Some possible answers are given below. 6799
52
53 04469
54 2467
Confidence Level Confidence Interval Margin of error 55 03578
90% 5.3781 to 5.5177 0.0698 56 12358
95% 5.3639 to 5.5320 0.0840 57 59
99% 5.3345 to 5.5613 0.1134 58 5

III.32. For the data in Table I.3. we find n = 86, x = 15.744, and s = 5.909 days. With df = 85, the
95% confidence interval is x ± t * s / n = 14.48 to 17.01 days from April 20, or May 3 through
May 6.
Minitab output
Confidence Intervals
Variable N Mean StDev SE Mean 95.0 % C.I.
Date 86 15.744 5.909 0.637 ( 14.477, 17.011)

III.33. (a) “SEM” stands for “standard error of the mean”; SEM = s / n (b) Two-sample t-tests
were done, because wild-type and aP 2− / − mice are separate (independent) groups. (c) The
observed differences between the two groups of mice were so large that they would be unlikely to
occur by chance if the two groups were not different. Specifically, if the groups were the same,
and we took many samples, the difference in glucose levels would be so large less than 5% of the
time, and the difference in insulin levels would be so large less than 0.5% of the time. We have
stronger evidence of a difference in insulin levels.
III.34. The insulin means are x1 = 5.9 and x2 = 0.75 ng/ml, and the standard deviations are
approximately s1 = 0.9 10 = 2.85 and s2 = 0.2 10 = 0.632 ng/ml. The estimated standard error
of the difference is SE = 0.92 + 0.22 = 0.922, so t = 0.9 − 0.2
0.922
= 5.59. With either df = 9 or df =
9.9, Table C allows us to say that P < 0.001 (for a two-sided alternative)—an even stronger result
than the paper claimed. (And with larger sample sizes, the evidence would be even stronger.)
Note that you do not actually need to compute s1 and s2 to find SE, but you do need them if you
want to find the more accurate estimate for df. Doing so is overkill in this case because we do not
know the correct sample sizes anyway.
220 Part III Review

2
III.35. (a) We test H0: σ1 = σ2; Ha: σ1 > σ2. The test statistic is F = 26.092 = 15.2 with df 161 and
6.69
134. This F value can be compared to Table D values for df 120 and 100; it has a very small P-
value, so we have strong evidence of a difference. (b) No: F procedures depend on Normality,
and large sample sizes do not help.

III.36. We test H0: σ1 = σ2; Ha: σ1 ≠ σ2. The test statistic is F = 682 = 2.388 with df 25 and 22. This
442
F value can be compared to Table D values for df 20 and 20, where we find that 2.12 < F < 2.46,
so 0.025 < P < 0.05 (software gives 0.043). This is significant evidence (at α = 0.05) that the
standard deviations differ.

III.37. (a) With df = 49, we take t* = 2.423 (conservatively) or 2.403 (approximately) from Table C,
or 2.405 (using software). The first and third of these are the best choices, but for a power
computation, the second is acceptable. (b) Reject H0 if t ≥ t*. Because t = x /( s / n ), this means
we reject H0 when x ≥ t * (108 / 50) = 15.2735t *. For the three values of t* given above, this
means x ≥ 37.01, x ≥ 36.70, and x ≥ 36.73, respectively.(c) The power against
µ = 100 is P( x ≥ 37) = P( Z ≥ 37 −100 ) = P( Z ≥ −4.12) > 0.9998. A sample of size 50 will almost
108 / 50
certainly detect a difference of $100, so this should be quite adequate.

III.38. (a) We use either t* = 1.660 (using df = 100 from the table) or 1.649 (df = 349 from

software). (b) Reject H0 if t ≥ t*, so xB − x A > t * 4002 + 4002 ; this means xB − x A must be above
350 350
either c = 50.19 (table) or c = 49.86 (software). (c) The approximate power against µB – µA = 100
is

 
50 − 100
P( xB − x A ≥ 50) = P  Z ≥  = P( Z ≥ −1.65) = 0.05.
 4002 + 4002

 
 350 350 
This test will not often detect a difference of $100.

III.39 We have 30 available observations; the distribution is rightskewed, 2 3


3 4899
with a high outlier of 123 g. The mean is x = 54.07 g and the standard
4 12222455689
deviation is s = 19.41 g; a 95% confidence interval for the mean city 5 017789
particulate level is 54.07 ± 2.045s / 30 = 46.82 to 61.31 g. 6 089
If we discard the outlier, we get x * = 51.69 and s* = 14.65 g, and the 7 12
8 26
interval is 51.69 ± 2.048s * / 29 = 46.12 to 57.26 g. 9
Confidence intervals should be used with caution here: The outlier 10
makes t procedures suspect, and discarding the outlier may cause us to 11
underestimate the actual mean. 12 3
Solution 221

III.40. The distribution of differences (city minus rural) has 26 –0 32


observations. There are two high outliers (15 and 18). If we naïvely –0 11110
 
use the t procedures in spite of the outliers, we get x = 2.192 g, s = 0 01111111
4.691 g, and t = 2.38 (df = 25). The P-value for testing H0: µc = µr 0 2222222
0 5
vs. Ha: µc > µr is 0.01 < P < 0.02 — pretty strong evidence against
0 7
H0. A 95% confidence interval for µc – µr is 2.192 ± 0
2.060 s / 26 = 0.297 to 4.087 g. 1
If we throw out those outliers, the data seem to be more suitable 1
for a t procedure. Using the other 24 observations, we find x * = 1 5
1.000 g, s* = 2.106 g, t = 2.33 (df = 23), and 0.01 < P < 0.02. We 1
can still reject H0, even though we have removed from the data the 1 8
two strongest individual pieces of evidence against H0. A 95% confidence interval for µc – µr is
1.000 ± 2.069s* 24= 0.111 to 1.889 g.

III.41. The plot (right) shows a strong positive linear


relationship; there is only one observation — (51, 69) —
which deviates from the pattern slightly. Because both
variables are measured in grams, the scales should be the
same on both axes.
The regression line ŷ = –2.580 + 1.0935x explains r2 =
95.1% of the variation in the data. (This is the solid line in
the plot.) When x = 88 g, we predict ŷ = 93.65 g.
The point (108, 123) is a potentially influential
observation (although it does not seem to deviate from the
pattern of the other points). Computing the regression line
without this point gives ŷ = 1.963 + 0.9942x and r2 =
92.1%, and ŷ = 89.45 g when x = 88 g. (This is the dashed line in the plot.)

Вам также может понравиться