24 BMEN E4110 Lecture8

BMEN E4110
Biostatistics for Engineers

Lecture 8 | 2 October 2024
Overview
• Hypothesis testing for proportions

• Z-test and Chi-squared test
• Risks & odds
Overview

• Risks & odds
Loading…
Analysis of proportions: Summary statistics for population data
Claire’s sample data Xn = 0
Xn = 1 Proportion of ‘left-handed’ = Sum of 1’s

Total number of datapoints
Left-handed Right-handed
two subclasses
Population data Xn = 0
Xn = 1
• N = total datapoints in the population
• M = number of in population 1
• N-M = number in population 2
Population data
Xn = 0
Xn = 1
Loading…
•
•
N = total datapoints in the population
M = number of in population 1
Population data
Xn = 0

Xn = 1 • M = number of in population 1
-given
value
Mexu
Ex. 1

Ex. 2
P =
O=
200 ts
poin 100 Loch
N 100 • N = total datapoints in the population
For :GS • M = number of in population 1
Ex. 3

scan change p to
highestpossiblenumbe
a
Ex. 1 Ex. 2 Ex. 3
Xn = 0
G
=
Xn = 1 Xn = 0
Xn = 1 Xn = 1 Xn = 0
Left-handed Right-handed Left-handed Right-handed Left-handed Right-handed

↑
highest degree of dispensi an
Standard deviation of category data
What do Ex.1 – Ex. 3 live on this curve? EX2

A
ex3 Loading…
Highest variability
(equal chance of left-handed and right-handed)
I
No variability No variability
(no uncertainty) (no uncertainty)
ex1p
Hypothesis testing for proportions
• Now that we understand how to use an analysis of proportions to learn about Claire’s data
(e.g., proportions of left-handed vs right-handed patients our BCI experiment)
• Sandra: Do left-handed patients experience a greater chance of improvement with our BCI
system?
Claire’s sample data
How should Sandra plan data

collection for this new experiment?
system?

Left-handed Right-handed should eqe d

system?

system?
Sandra’s sample data

system?

Improved Same
-
E datu
must inchoe another colleges
system?

Improved Same
system?
How should Sandra plan data collection for this new experiment?
Left-handed data Right-handed data
(n = 100) (n = 100)
plan
data
collection
Improved Same Improved Same

Hypothesis testing for proportions: Case study 2
system?
How should Sandra plan data collection for this new experiment?
(n = 100) (n = 100)

system?
• H0: pleft-handed = pright-handed

(n = 100) (n = 100)

n = 11 n = 89 n = 10 n = 90
system?
• H0: pL = pR
thesame.
& ask isproportion improvence
of
(n = 100) (n = 100)
Improved Same & Improved Same

n = 11 n = 89 n = 10 n = 90
sound
Hypothesis testing for proportions: Case study 2 (USA)
system?
• H0: pL = pR

(n = 100) (n = 100)

n = 11 n = 89 n = 10 n = 90
Hypothesis testing for proportions: Case study 2 (Thailand)
system?
• H0: pL = pR

(n = 100) (n = 100)

n = 20 n = 80 n=6 n = 94
• Is the difference in the Sandra’s Thailand sample big enough to infer a difference in the
underlying population?

(n = 100) (n = 100)

n = 20 n = 80 n=6 n = 94
Overview

• Risks & odds
• Two tests we can use:
• Z-test
• Chi-squared test

(n = 100) (n = 100)

n = 20 n = 80 n=6 n = 94
• Two tests we can use:
• Z-test
• Chi-squared test

(n = 100) (n = 100)

n = 20 n = 80 n=6 n = 94
Case study 2 (Thailand): Z-test
• H0: pL = pR us
results are
similar
• H0: pL = pR
Loading…
I t
• Compute z-value: pp
a
• H0: pL = pR
• Compute z-value:
• H0: pL = pR
• Compute z-value:
Sample-wide
proportion
• H0: pL = pR
• Compute z-value: Gives us an idea of how

extreme our sample is under H0
Sample-wide
proportion
Gives us an idea of how

Sample-wide
proportion
t-distribution
Probability distribution function (pdf)

depending a
d
dage
Frieden
impacts
distribute
Key features: shape (heavier tails), degree of

freedom (v), centered at zero t-value (no. SD from the mean)
wikipedia.org
normal distribution
z-distribution is a t-distribution with infinite degrees of freedom
Probability distribution function (pdf) more data point,

I should approach
more
normal
Key features: shape (heavier tails), degree of

freedom (v), centered at zero z-value (no. SD from the mean)
wikipedia.org

Sample-wide
proportion
we can locker critical z value

Sample-wide
proportion
no night,
magdipt
as
nat
X = 0 . OS
8
-
alway met crit


Sample-wide Okay, but how extreme are

proportion Sandra’s samples?
1 96
Zarit = .
US A Thailand sample
improve same total
improve Sam a
to
800 694100
all 21 179 200
all 26 174 200
= = 0 11
.
-= = 02
.
n= = 0 10 .
Pr = 0. = 0
:impo
Zu
=
2 . 94
improve same total
Y
all
• H0: pL = pR
• Compute z-value: Gives us an idea of how

Sample-wide
proportion
,CT Ho
when 2Zorit
• How extreme are Sandra’s samples?: What’s the chance of observing the computed z-value?
z-value = 2.94
p = 0.003
*
exact p-value
can get
to put it together
z-value = 2.94
p = 0.003
• The chance of getting a

sample as extreme as the
one Sandra got is 0.003 (if
we assume H0).
• We can infer that there’s a

difference in BCI-induced
improvement in left-handed
vs. right-handed patients.
z-value = 2.94
p = 0.003

we assume H0).
• We can infer that there’s a

difference in BCI-induced
improvement in left-handed
vs. right-handed patients.
Chi-squared(x2) to i
have not done this yet

• Compare actual counts vs. those expected under the H0.
I
• Expected counts computed based on column and row totals.

the hull
flexb (more than two
under
-
Chi is
groups
hypothesis
• Compare actual counts vs. those expected under the H0.
• Expected counts computed based on column and row totals.
• Statistic is a function of squared differences in counts.

Overall Total
Degrees of freedom = (r-1) x (c-1)

it
improve same total
h
C =
002
100
-
·
200
&
174
13

Sample-wide
proportion
• H0: Handedness and BCI-induced improvement are independent.
• This means there is no association between handedness and BCI-induced improvement.
The proportion of patients experiencing improvement (‘improved’) or no improvement
(‘same’) is the same for both left-handed and right-handed groups.
Any observed differences (e.g., more left-handed patients showing improvement) are due
to random chance.
• In other words, H0: The distribution of BCI-induced improvement is the same for left-handed
and right-handed patients
DF = (r-1) x (c-1)
findcrit of Chi-still
DF = (r-1) x (c-1)
z-value = 2.94
p = 0.003

we assume H0).
• We can infer that

handedness and BCI-
induced improvement are
associated (not
independent).
z-value = 2.94
p = 0.003
Overview

• Risks & odds

24 BMEN E4110 Lecture8

Uploaded by

Copyright:

Available Formats

24 BMEN E4110 Lecture8

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

24 BMEN E4110 Lecture8

Uploaded by

Copyright:

Available Formats

BMEN E4110

Biostatistics for Engineers

• Hypothesis testing for proportions

• Hypothesis testing for proportions

Claire’s sample data Xn = 0

Xn = 1 Proportion of ‘left-handed’ = Sum of 1’s

• N = total datapoints in the population

• N = total datapoints in the population

• N = total datapoints in the population

Ex. 1 Ex. 2 Ex. 3

Left-handed Right-handed Left-handed Right-handed Left-handed Right-handed

What do Ex.1 – Ex. 3 live on this curve? EX2

Claire’s sample data

How should Sandra plan data

Claire’s sample data

How should Sandra plan data

Left-handed Right-handed should eqe d

Claire’s sample data

How should Sandra plan data

Sandra’s sample data

How should Sandra plan data

Sandra’s sample data

How should Sandra plan data

Sandra’s sample data

How should Sandra plan data

Improved Same Improved Same

Improved Same Improved Same

Sandra’s sample data

Improved Same Improved Same

Improved Same & Improved Same

Sandra’s sample data

Improved Same Improved Same

Sandra’s sample data

Improved Same Improved Same

Left-handed data Right-handed data

Improved Same Improved Same

• Hypothesis testing for proportions

Left-handed data Right-handed data

Improved Same Improved Same

Left-handed data Right-handed data

Improved Same Improved Same

• Compute z-value: Gives us an idea of how

Gives us an idea of how

Probability distribution function (pdf)

Key features: shape (heavier tails), degree of

Probability distribution function (pdf) more data point,

Key features: shape (heavier tails), degree of

Gives us an idea of how

Gives us an idea of how

alway met crit

Gives us an idea of how

Sample-wide Okay, but how extreme are

improve same total

• Compute z-value: Gives us an idea of how

• The chance of getting a

• We can infer that there’s a

• The chance of getting a

• We can infer that there’s a

have not done this yet

• Expected counts computed based on column and row totals.