You are on page 1of 21

1

Introduction
On the pages below are a series of tables and figures that you might
consider using in your report. You should feel free to edit these if you
want. There are also a few notes in red font to give you some idea of my
thought processes during the analysis of the data.
This is the third time we have done this practical exercise, so I have
included data from 2014, 2015 and 2016, giving a total of 958 sets of
observations (although some are incomplete). I took the following
approach to the analysis of the data-set:

I first wanted to know something about the characteristics of the


sample. This is important because, in any analysis, you must be able to
consider the extent to which your observations and conclusions are
generalizable to a population. These data are presented in Tables 1 and
2. Table 1 presents the categorical variables and Table 2 presents the
continuous variables. How generalizable do you think this sample is to
the wider population?

For continuous variables I chose to use the standard deviation as my


measure of dispersion because, unlike the standard error of the mean,
this is largely independent of the sample size. Consequently, the
standard deviations can be compared between the various categories,
even though some of the sample sizes within each factor vary widely.

The next step was to partition the data according to gender (Table 3;
Figure 1), diet (Table 4; Figure 2) and level of physical activity (Table 5;
Figure 3). This enabled me to test the hypothesis that the continuous
variables differed according to these categorical factors. Note that this
analysis does not tell us anything about causality. You should also
consider whether you find the tables or the figures more useful for
presentation of the data. You should include one or the other in your
report, not both. One of the rules of scientific presentation is that you
should not present the same data in tables and figures. You should
make a decision about which of these modes of delivery will be most
effective in your report. In general, I use figures if there are patterns in
the data that can best be illustrated graphically. I use tables if the
absolute levels of a variable matter. Tables can also be useful
sometimes if you have a large set of data that would otherwise take up
too much space in the form of figures.

I then moved to considering the relationship between body mass index


and cardiovascular function (Figure 4). Note, again, that these data

provide information about association but not causation. My first step


for this was to generate scatterplots. I then performed regression
analysis to ascertain whether arterial pressure and heart rate vary with
body mass index. I calculated the line of best fit both by ordinary leastsquares (OLS) regression analysis and ordinary least-products (OLP)
regression analysis. Note that the two methods give different lines of
best fit. Why do you think this is the case? Which do you think is
better? Why? I used two variables to determine the strength of the
relationship, r2 and P (Table 6). The P value tests the null hypothesis
that the slope of the relationship is zero (i.e. that there is no linear
relationship between the two variables). The Pearson product moment
correlation coefficient (r2) quantifies the proportion of the variance in
the dependent variable (i.e. the one on the y-axis) that can be
explained by the (OLP) linear relationship with the independent
variable (body mass index). Do you think arterial pressure and heart
rate differ according to body mass index? How strong do you think the
relationship is?

I also present the intercepts and slopes of these relationships in Table


6. What do the slopes tell us about the relationships between body
mass index and haemodynamic function?

The next step was to consider whether the three categorical variables
(gender, diet and level of physical activity) influence the relationships
between body mass index and arterial pressure and heart rate. For this,
I used analysis of covariance (Table 7). You will see three sets of P
values in each case.
o PBMI tests the hypothesis that, independent of gender or level of
physical activity, there is a linear relationship between body
mass index and the four haemodynamic variables. Note the
consistency between the PBMI values here and the P values in
Table 6. That is, they seem to support the idea that systolic
pressure, diastolic pressure and mean arterial pressure vary with
body mass index, but maybe heart rate does not, at least when
the entire sample is considered in one hit.
o PGender tests the hypothesis that the haemodynamic variables
differ by gender, independently of body mass index. PActivity tests
the hypothesis that the haemodynamic variables differ by level of
physical activity, independently of body mass index. PDiet tests
the hypothesis that the haemodynamic variables differ in
vegetarians compared with non-vegetarians, independently of
body mass index. The data seem to indicate some relationship
between gender and heart rate, and between usual level of

physical activity and heart rate. Contrast these P values with


those presented in Tables 3, 4 and 5, which test slightly different
hypotheses. Can you explain this slight difference?
o PGender*BMI and PActivity*BMI and PDiet*BMI are what we call interaction
terms. They test the hypotheses, respectively, that the slope of
the relationships between body mass index and the
haemodynamic variables differ according to gender, level of
physical activity, or diet. Note that PGender*BMI is 0.05, so it would
only be reasonable for us to reject the null hypothesis in that
case. That is, it seems likely that gender influences the way in
which BMI is associated with heart rate.

Finally, I present scatterplots of the data with the two genders (Figure
5) and the four levels of physical activity (Figure 6) shown in different
colours. I have not bothered to generate separate plots according to
diet, because there was no evidence that diet altered any of the
relationships between body mass index and the haemodynamic
variables (Table 7). In cases where the analysis of covariance (Table 7)
provided evidence that the relationships between body mass index and
the haemodynamic variables are dependent on the categorical
variable, I present the individual regression lines. It seems that the
relationship between heart rate and body mass index is dependent, to
some extent, on both gender and usual level of physical activity.
Hopefully, these plots will get you thinking about the nature of these
apparent effects. I wonder why the relationship between body mass
index in males has a positive slope whereas it appears to have a
negative slope for females? Similarly, the relationship seems to have
negative slope in all but the individuals who do the most exercise. To
dig deeper you might like to try to interpret the output of the statistical
analysis (see Systat output for Table 7 and Figures 5 and 6.pdf).

Table 1: Characteristics of the sample: categorical variables


____________________________________
Variable
n
% total
____________________________________
Gender
Male
360
37.6
Female
597
64.6
____________________________________
Diet
Vegetarian
61
6.4
Non-vegetarian
897
93.6
____________________________________
Level of Physical Activity
None
155
16.2
1-2 Times Per
347
36.3
Week
3-5 Times Per

317

33.2

Week
>5 times per

137

14.3

week
____________________________________
Data are compiled from a total of 958 individuals.
Note to students: Large differences in sample size can distort the
outcomes of parametric statistical tests (e.g. t-tests, analysis of variance
or analysis of covariance). Thus, you should apply a bit of suspicion to
analyses of the effects of diet, since there were so few vegetarians (61)
compared to non-vegetarians (897).

Table 2: Characteristics of the sample: continuous variables


___________________________________________________________________________
Variable

Mea
n

Standa Standa
rd

rd

Deviati

Error

on

of the

Medi

Range

an

Mean
___________________________________________________________________________
Age (years)
Height (cm)

954
957

20.8
168.

1.9
9.6

0.1
0.3

20
168.0

18-39
146.0-

Weight (kg)

949

7
66.0

13.5

0.4

64.1

200.0
40.0-

117.8
BMI (kg/m2)
948 23.0
3.5
0.1
22.5 15.6-38.9
___________________________________________________________________________
Systolic Pressure

955

114.

12.6

0.4

114.0

81.0-

(mmHg)
Diastolic Pressure

955

2
68.8

7.8

0.3

68.0

183.5
43.5-96.0

(mmHg)
Mean Arterial

955

83.9

8.2

0.3

83.2

59.0-

75.5

116.5
40.0-

Pressure (mmHg)
Heart Rate

950

76.2

11.6

0.4

(beats/min)

120.0

Data are compiled from a total of 958 individuals. Mean arterial pressure
was calculated as diastolic pressure plus one third of pulse pressure
(systolic-diastolic pressure).

Table 3: Characteristics of participants according to gender.


________________________________________________________________________
Variable

Male
Mean

Female
Mean

SD
SD
________________________________________________________________________
Age (years)
Height (cm)

358
359

21.0 1.9
177.0

596
598

20.6 1.8
163.7

0.01
<0.0

Weight (kg)

359

7.8
76.6

590

6.8
60.1 9.5

01
<0.0

358

13.8
24.1 3.7

22.4 3.3

01
<0.0

BMI (kg/m )

590

01
Systolic Pressure

360

121.9

595

109.5

<0.0

(mmHg)
Diastolic Pressure

360

11.5
69.3 7.9

595

10.8
68.5 7.8

01
0.09

(mmHg)
Mean Arterial

360

86.8 7.8

595

82.2 7.9

<0.0

Pressure (mmHg)
Heart Rate

356

74.9

594

77.1

01
0.007

(beats/min)

12.1

11.1

Data are from a total of 958 participants. P values are the outcomes of
Students unpaired t-test and are shown in bold if 0.05. SD = standard
deviation, BMI = body mass index.

Table 4: Characteristics of participants according to diet.


________________________________________________________________________
Vegetarian
Variable

Mean

NonVegetarian
n
Mean

SD
SD
________________________________________________________________________
Age (years)
Height (cm)

61
61

20.8 1.5
167.5

893
896

20.7 1.9
168.7

0.80
0.78

60

7.3
64.9

889

9.7
66.1

Weight (kg)

0.04

BMI (kg/m2)

60

10.2
23.1 3.5

888

13.7
23.0 3.5

0.88

Systolic Pressure

61

110.0

894

114.5

0.001

(mmHg)
Diastolic Pressure

61

10.0
67.0 6.3

894

12.7
68.9 7.9

0.03

(mmHg)
Mean Arterial

61

81.3 6.5

894

84.1 8.3

0.002

Pressure (mmHg)
Heart Rate

60

74.3

890

76.4

0.15

(beats/min)

10.7

11.6

Data are from a total of 958 participants. P values are the outcomes of
Students unpaired t-test and are shown in bold if 0.05. SD = standard
deviation, BMI = body mass index.

Table 5: Characteristics of participants according to their usual level of physical activity.


________________________________________________________________________________________________________________
No Exercise
1-2 TPW
3-5 TPW
> 5 TPW
Variable
n
Mean
n
Mean
n
Mean SD n
Mean SD
P
SD
SD
________________________________________________________________________________________________________________
Age (years)

155

20.7 1.5

347

20.7 1.9

Height (cm)

156

346

Weight (kg)

154

BMI (kg/m2)

154

165.2
9.6
61.5
14.9
22.3 3.9

166.7
9.1
63.4
13.0
22.7 3.5

344
343

31
5
31
7
31
4
31
4

20.8 2.1

136

20.8 1.5

0.88

169.5
8.9***
67.9
12.4***
23.5
3.4***

137

175.3
8.6***
73.1
12.3***
23.7
3.2***

<0.0
01
<0.0
01
<0.0
01

136
136

Systolic Pressure
156
111.1
345
112.5
31
115.2
137
119.3
<0.0
(mmHg)
11.8
12.8
6
12.5***
11.1***
01
Diastolic Pressure
156 70.1 8.1 345 69.3 7.8 31
68.4 7.8
137
67.0
0.004
(mmHg)
6
7.1***
Mean Arterial
156 83.8 8.4 345 83.7 8.5 31
84.0 8.2
137
84.5 7.1
0.81
Pressure (mmHg)
6
Heart Rate
155
79.6
343
78.3
31
74.9
136
70.6
<0.0
(beats/min)
11.5
11.2
5
10.4***
12.5***
01
__________________________________________________________________________________________________________________
Data are from a total of 958 participants. P values are the outcomes of one-way analysis of variance and are shown
in bold if 0.05. *P 0.05, **P 0.01, ***P 0.001 for post hoc comparisons with no exercise (Dunnetts test).
TPW = times per week, SD = standard deviation, BMI = body mass index.

10

Table 6: Ordinary least products regression analysis for the


relationships depicted in Figure 4.
___________________________________________________________________________
Dependent Variable
r2
P
intercept
Slope
___________________________________________________________________________
Systolic Arterial Pressure

0.17

<0.001

32.7(27.0-

3.53(3.28-3.78)
2.22(2.04-2.40)

Diastolic Arterial Pressure

0.04

<0.001

38.5)
17.6(13.4-

Mean Arterial Pressure

0.11

<0.001

21.7)
30.6(25.6-

2.31(2.14-2.48)

0.43

34.5)
152(145-

-3.27(-3.56 to

Heart Rate

0.001

158)
-2.97)
___________________________________________________________________________
r2 = Pearson product-moment correlation coefficient. P = probability that
no relationship exists between body mass index and the dependent
variable. Intercept = a, slope = b for Y = a + b*X, expressed as mean and
95% confidence intervals.

11

Table 7: Outcomes of analysis of covariance.


___________________________________________________________________________
Dependent Variable
PGender
PBMI
PGender*BMI
___________________________________________________________________________
Systolic Arterial Pressure
0.11
<0.001
0.53
Diastolic Arterial Pressure
0.44
<0.001
0.44
Mean Arterial Pressure
0.21
<0.001
0.81
Heart Rate
0.003
0.88
0.01
___________________________________________________________________________
Dependent Variable
PActivity
PBMI
PActivity*BMI
___________________________________________________________________________
Systolic Arterial Pressure
Diastolic Arterial Pressure
Mean Arterial Pressure
Heart Rate

0.98
0.85
0.92
0.005

<0.001
<0.001
<0.001
0.24

0.78
0.92
0.96
0.07

Dependent Variable
PDiet
PBMI
PDiet*BMI
___________________________________________________________________________
Systolic Arterial Pressure
Diastolic Arterial Pressure
Mean Arterial Pressure
Heart Rate
Values 0.05 are bolded.

0.15
0.99
0.48
0.78

<0.001
0.005
<0.001
0.45

0.06
0.78
0.27
0.67

12

Figure 1: Characteristics of participants according to gender. Columns


and error bars represent the means standard deviation of data from
356-360 males and 590-598 females. P values are the outcomes of
Students unpaired t-test.

13

Figure 2: Characteristics of participants according to diet (vegetarian or


non-vegetarian). Columns and error bars represent the means standard
deviation of data from 60-61 vegetarians and 888-896 non-vegetarians. P
values are the outcomes of Students unpaired t-test.

14

Figure 3: Characteristics of participants according to usual level of


physical activity. 1 = no regular exercise (n = 154-156), 2 = regular
exercise 1-2 times per week (n = 343-347), 3 = regular exercise 3-5 times
per week (n = 314-317), 4 = regular exercise > 5 times per week (n =
136-137). Columns and error bars represent mean standard deviation. P
values are the outcomes of one way analysis of variance. * P 0.05, ** P

15

0.01, *** P 0.001 for post hoc comparisons with no regular exercise
(Dunnetts test).

16

Figure 4: Relationships between body mass index and arterial pressure


and heart rate. Symbols represent data from individuals (n = 958). Lines
of best fit were determined by ordinary least squares (model 1) or
ordinary

least

products

(model

2)

regression

analysis.

Regression

coefficients for the ordinary least products analysis are provided in Table
6.

17

18

Figure 5 Relationships between body mass index and arterial pressure


and heart rate according to gender. Symbols represent data from
individual males (blue, n = 358) and females (pink, n = 592). Lines of best
fit were determined by ordinary least products (model 2) regression
analysis. Analysis of covariance (Table 7) showed a significant effect of
gender on the relationship between heart rate and body mass index, so
separate lines of best fit are shown for the two genders. For other

19

variables no significant effects of gender were detected, so the line of best


fit for the combined data is shown.

20

Figure 6 Relationships between body mass index and arterial pressure


and heart rate according to usual level of physical activity. Symbols
represent data from individuals who do not participate in regular exercise
(blue, n = 154), who exercise regularly 1-2 times per week (green, n =
343), who exercise regularly 2-5 times per week (red, n = 314) and who
exercise regularly more than 5 times per week (pink, n = 136). Analysis of
covariance (Table 7) did not detect significant effects of usual level of
activity on systolic, diastolic or mean arterial pressure, so the lines of best
fit (ordinary least products regression analysis) for the combined data are
shown. However, the relationship between heart rate and body mass

21

index did differ significantly according to usual level of activity, so lines of


best fit are shown for the four categories separately.

You might also like