ST1232

Tutorial 8
1. It is known that serum‐creatinine level of individuals follows a Normal distribution. This level
was measured in 12 randomly sampled patients, twenty‐four hours after they received a
newly proposed antibiotic.
(a) Use the output below to perform the following one‐sided test instead, at α = 0.05 significance
level. H0 : µ = 1 against H1 : µ < 1

The test statistic, t = 0.176,

The two‐sided P‐value is is 0.864.
For left‐tailed test (alternative hypothesis is ‘<’) here, the corresponding one‐sided P‐value
for this test is the left area of test statistic 0.176, P(t <0.176) which is 1 – (0.864/2) = 0.568.
P‐value = 0.568 > α = 0.05,
we do not reject the null hypothesis (the test is not statistically significant).
We conclude that the mean serum level is not significantly different from 1.

(b) Use the output above to compute a 95% confidence interval for the population mean serum
level of individuals who have taken this antibiotic.
We can obtain the desired two‐sided confidence interval for µ as 1 + (−0.37098, 0.43539) =
(0.63, 1.44)
We can be 95% confident that this interval contains the population mean serum level after
taking this antibiotic.

1
2. In this question, we shall see that in some cases, we can apply a natural logarithm transform
when we wish to compute a confidence interval, but our data do not satisfy the assumptions
that we needed.
A medical researcher took blood samples from n = 28 children who were infected with
malaria, and determined, for each child, the number of malaria parasites in 1 ml of blood. The
following dataset contains the counts in the variable parasites: dataset_tut_08_malaria.sav
We shall use X to denote the parasite counts, and Y = ln X, the natural logarithm of X.
(a) Use SPSS to obtain the sample mean ( ܺത ) and sample standard deviation ( s ) of the parasites
variable.
ܺത = 14247.79; s = 28575.614
(b) Plot a histogram and a QQ‐plot of the parasites variable. Comment on the skewness of the
data.
The histogram clearly looks non‐normal. The data is very skewed to the right.
The QQ Plot shows that the distribution cannot be approximated by a normal distribution.

2
(c) Using the Transform > Compute Variable menu option in SPSS once more, create a new
variable, ln parasites by taking the natural logarithm of parasites. Comment on the skewness
of this new variable.
The histogram is unimodal and symmetric.
The QQ Plot shows that the distribution can be approximated by a normal distribution.

(d) Compute a 95% confidence interval for the mean of ln parasite counts. This is an interval
estimate, but what of?
Let µX be the population mean of X (the population mean of parasite counts).
Let µY be the population mean of ln X (the population mean of ln parasite counts).
µX ≠ µY
ܻത = 8.3304; s = 1.67224; the standard error = 0.31602.
From the t‐distribution with n‐1 = 27 degrees of freedom, t‐score of 2.052 has a right‐tailed
probability of 0.025.
Hence 95% CI for µY = 8.3304 ± (2.052)(0.31602) = (7.68, 8.98)
We are 95% confident that this interval contains the population mean of ln parasite.

3
(e) Take the antilogarithm of the interval in question (d). This provides a 95% confidence interval
for the population median of parasite counts.
As the above interval is on the log‐scale, we exponentiate it to get back an interval on the
actual scale: e(7.68, 8.98) = (2164.62, 7942.63)
We are 95% confident that this interval contains the population median of parasite counts.

3. The following measurements of corneal thickness are obtained for 8 subjects affected with
glaucoma in one eye.

(a) If the data above are dependent samples, use SPSS to check the assumption that the
difference, diff (= Glaucoma – Unaffected) is Normally distributed.
The histogram is not quite useful in this case because there are too few observations.
The QQ plot does not contradict the assumption of Normality, so we shall proceed with the
inference.

(b) Perform a dependent samples t‐test to decide if glaucoma decreases the thickness of the
cornea at α = 0.05 level of significance.
Let µD be the population mean of the diff (Glaucoma –Unaffected)
H0 : µD = 0
H1 : µD < 0
The test statistic, t = –1.917.
The two‐sided P‐value is 0.097.
For left‐tailed test here, the corresponding P‐value P(t <–1.917) = (0.097/2) = 0.0485.
P‐value = 0.0485 < α = 0.05,
we reject the null hypothesis at α = 0.05 (The test is statistically significant).
We conclude that there is evidence showing that Glaucoma reduces the thickness of the
cornea.

4
(c) Suppose that we ignore the fact that the data is dependent. What would our conclusion have
been, at the same significance level? (You may skip the Normality check for this question.)

For two independent t‐test,

H0 : µ1 = µ2
H1 : µ1 < µ2 (mean cornea thickness of Glaucoma < mean cornea thickness Unaffected eyes)

The larger sd is s2 = 31.3141, the smaller sd is s1 = 28.6673, the larger sd is not more than
twice of the smaller  we conclude that the two groups have equal variance, and proceed
with the test.
The test statistic, t = –0.416.
The two‐sided P‐value is 0.683.
For left‐tailed test here, the corresponding P‐value P(t <––0.416) = (0.683/2) = 0.3415
P‐value = 0.3415 > α = 0.05,
Do not reject the null hypothesis at α = 0.05 (The test is NOT statistically significant).
We conclude that there is not enough evidence to conclude Glaucoma reduces the thickness
of the cornea.

5
4. Some common strategies for treating hypertensive patients by non‐pharmacologic methods
include (1) weight reduction and (2) trying to get the patient to relax more via meditational
techniques. Suppose these strategies are evaluated by randomizing hypertensive patients to
four groups who receive the following types of non‐pharmacologic therapy:
Group 1: Patients receive counseling for both weight reduction and meditation.
Group 2: Patients receive counseling for weight reduction but not meditation.
Group 3: Patients receive counseling for meditation but not weight reduction.
Group 4: Patients receive no counseling at all.
Suppose that 80 hypertensive patients are assigned at random to each of the four groups (20
in each), and the change in diastolic blood pressure is noted in these patients after a 1‐month
period. You may assume that change in diastolic blood pressure for all four groups is Normally
distributed.
Each observation in this dataset reflects the mean decrease in diastolic blood pressure for
that individual due to the therapy received. (It is computed as bp.change = before – after.)
Figure below contains a boxplot of the data for the four groups.

The objectives of this study are the following:

• Which groups are significantly different from the control group? This is to be tested at
experiment‐wise error rate α = 0.05.
• Does meditation help when weight reduction is already applied, i.e. is the decrease in b.p
greater for Group 1 when compared to 2? This is to be tested at α = 0.05 level.
(a) Carry out a one‐way ANOVA F‐test at α = 0.05, using the SPSS outputs provided below.
The largest sd is s4 = 6.745, the smallest sd is s2 = 4.705, the largest sd is not more than twice
of the smallest  we conclude that the four groups have equal variance, and proceed with
the test.
Let µ1 , µ2, µ3, µ4, be the respective group means for change in blood pressure.
H0 : µ1 = µ2= µ3 = µ4
H1 : At least two of them are different

The test statistic, F = 6.271

The P‐value is 0.001.
P‐value = 0.001 < α = 0.05,
We reject the null hypothesis at α = 0.05 (The test is statistically significant).
We conclude that there is strong evidence that at least two of the groups have a significantly
different change in blood pressure.

6
7
(b) Carry out the one‐sided test for the following hypothesis at α = 0.05:
H0 : µ1 = µ2
H1 : µ1 > µ2
The required SPSS output is given below.

The test statistic, t = 2.488

The two‐sided P‐value is is 0.015.
For right‐tailed test (alternative hypothesis is ‘>’) here, P(t > 2.488) which is (0.015/2) =
0.0075
P‐value = 0.0075 < α = 0.05,
We reject the null hypothesis at α = 0.05 (The test is statistically significant).
We conclude that there is strong evidence that the mean decrease for Group 1 is larger than
the mean decrease in blood pressure for group 2, at α = 0.05 level.

8
(c) Carry out the appropriate multiple comparison procedure to assess which groups are
significantly different from the control group, at experimentwise error rate 0.05. Use the SPSS
output below.
The appropriate multiple comparison procedure is the Dunnett test because we are comparing
all treatments to the control.
The output shows that the adjusted P‐value for each comparison (3 in total) is:
µ1 – µ4, adjusted P‐value = 0.000 < α = 0.05
µ2 – µ4, adjusted P‐value = 0.179 > α = 0.05
µ3 – µ4, adjusted P‐value = 0.090 > α = 0.05
At experimental error rate 0.05, we conclude that there is very strong evidence that the
treatments in group 1 is significantly different from doing nothing, but the other two
treatments (group 2 and group 3) are not significantly different from the control.