Homework 8 Answers

Chapter 26


1. (a) True.
   (b) False.  The null hypothesis says the difference is due to
chance; the alternative says it's "real". (See pp. 478-479)


2. (a) null: avg of box equals 18/38
       alt:  avg of box more than 18/38

   (b) z = [1890-1800]/{[(18/38)(20/38)]^0.5(3800)^0.5} is about 2.92,
so the p-value is about 0.002.

   (c) It is very likely that it's biased in favor of reds.


4. (55-63)/[20/(30^0.5)] is about -2.19.

   So, we would expect that a TA section would do this badly (or
worse) due to chance only about (100-97.22)/2 = 1.39%.


6. (a) [102-175]/[(1/2)(350^0.5)] is about -7.80, so the p-value is
almost 0.

   (b) 102/(102+248) is about 0.29.  This is the fraction of females
in the group.

       (0.29)(100) = 29 So we expect 29 of 100 to be female.

       SE = (100^0.5) [(0.29)(0.71)]^0.5 is about 4.54

       correction factor: [(350-100)/(350-1)]^0.5 is about 0.846

       SE for number (drawing without repl): 0.846(4.54) is about 3.8

       (9-29)/(3.8) is about -5.2, so p-value is almost 0.  This is
the chance of drawing 9 women or fewer in a random sample of the
group.

   (c) Since the likelihood of these events is so small, it seems hard
to believe the judge could have chosen like this by chance.

Chapter 27

1.  null hypothesis: tickets 50% positive
    alternative hypothesis: more than 50% positive

exp. number of positives: 500(1/2) = 250
SE number of positives:   (1/2)(500)^0.5 = 5(5)^0.5

    z-score: (276-250)/[5(5)^(0.5)] is about 2.33.  (The closest
number on our normal table chart is 2.35.)

(100-98.12)/2 = 0.94%  --- p-value = 0.0094

This leads us to think that the null hypothesis is probably not correct;
it is very likely that more than 50% of the tickets are positive.


3.  percentage who rated clergyman "very high or high"
1985: 67%   1992: 54%

Results are based on independent simple random sample of 1000 persons
in each year.

(a)  Should make a 2-sample z-test because these results
used a different sample each year (independent of the other year).  We
need to compare the means of the sample.

(b)  Need 2 box models, one for each year.  Each box represents a
different year's state of public opinion.  Tickets show "1" or "0",
with "1" meaning "very high" or "high" opinion of clergymen.  It
doesn't matter how many tickets - just need the right proportion of
"1"'s and "0"'s (67% in 1985 box and 54% in 1992).

null hypothesis:  same proportion rating clergymen "high" or "very high"
in both years - same percentage of "1"'s in both boxes

alternative hypothesis: percentage of "1"'s is lower in the 1992 box

(c)  z-score:  (0-13)/2.17 = -5.99 -- p-value almost 0

SE%(1985) = [(0.67)(0.33)]^0.5/(1000^0.5) x 100% = 1.49%
SE%(1992) = [(0.54)(0.46)]^0.5/(1000^0.5) x 100% = 1.58%

SE for difference of %: [(1.49)(1.49) + (1.58)(1.58)]^0.5 = 2.17%

Since p-value is so low, chance is not a good explanation fo rthis
result.  The test does not prove that scandals were the cause of the
change in opinion, however.


5. 383 students at UBC
200 chosen to got question A, 92 answered "yes"
remaining 183 got question B, 161 answered "yes"

null hypothesis: difference due to chance

alternative hypothesis: difference due to something else (wording of
questions)

SE%(A): [(0.46)(0.54)]^0.5/(200^0.5) x 100% = 3.5%
SE%(B): [(0.88)(0.12)]^0.5/(200^0.5) x 100% = 2.4%

z-score: [(46-88) - 0]/[(3.5)(3.5) + (2.4)(2.4)]^0.5 = -9.90
-- p-value almost 0

Chance isn't a good explanation, so it's probably something about the
wording of the question.


7.  (a) 592 assigned to treatment, 48.3% re-arrested
        154 assigned to control group, 49.4% re-arrested

null hypothesis: treatment group no better than control

alternative hypothesis: treatment group does better (less arrests)

SE%(T): [(0.483)(0.517)]^0.5/(592^0.5) x 100% = 2.05%
SE%(C): [(0.494)(0.506)]^0.5/(154^0.5) x 100% = 4.03%

z-score: [(48.3 - 49.4) - 0]/[(2.05)(2.05) + (4.03)(4.03)]^0.5 = -0.24

p-value = (100-19.74)/2 = 40.13  -- p-value = 0.4

So, there is no significant difference.


    (b)  null hypotheis: no difference in the averages for the 2 groups
	 alternative hypothesis: income support group has lower average


SE for treatment average: 0.7 wks
SE for control average: 1.4 wks

SE for the difference: 1.6 wks
observed difference: -7.5 

z-score: -7.5/1.6 is about -4.7  -- p-value almost 0

So, we reject the null hypothesis in favor of the idea that income
support makes the released prisoners work less.

 

Additional Problem

1.  You should have used JMP-IN to do the calculations.  (See my
posting on the newsgroup if you need the instructions for later
reference.)  You should have gotten a t-score of 1.08 and a p-value
that was greater than 0.2.  Based on the sample, there is no reason to
believe that the average number of disputes is higher or lower on
Mondays than on any other day of the week.