Biostatistics M1-1

Dr.
Nguyen Thi Van Anh

PMAB department
Content
• Introduction: parametric and non-parametric tests
• Check normality of data
• Independent t-test
• Paired t – test
• Pearson and Spearman correlation test

Introduc<on
Descriptive statistics Inferential statistics
• Collecting • Analyzing data
• Summarizing • Conclusion about the
• Presenting data population

How can use statistics to extrapolate from
sample to population?
Gaussian (normal) Non-gaussian (non-normal)

distribution
distribution (bell-shaped)
Parametric tests Non-parametric tests
Ex: t-test, ANOVA test Ex: Mann-Whitney test,

Wilcoxon test
Introduc<on
• Assume normal distribution • No assumed normal distribution

• Handles Interval/Ratio data • Handles Ordinal/Nominal data
• Results can be significantly • Results cannot be significantly
affected by ouliers affected by ouliers
• More statistical power • Less statistical power
Measurement and measurement scales
Measurement scale: 4 types of scale ordered in that all later scales
have all the properties of the earlier scales, plus new properties.
ü Norminal scale: use number just to distinguish categories
ü Ordinal scale: observations can be ranked
ü Interval scale: distance (not ratio) between 2 measurements is known
ü Ratio scale: equality of ratios and intervals may be determined

Examples of measurement scales
• Nominal scale:
Gender: male = 1, female = 2
“Scale” simply labels object

• Ordinal scale
Numbers used to place subjects in orders
But, No information on differences (intervals) between categories
• Interval scale:
Temperature: Celsius or Fahrenheiht
Same difference between 5°C (41°F) and 15°C (59°F) as between

20°C (68°F) and 30°C (86°F)
v The interval difference is meaningful

v BUT, we can not defend ratio relationships
v Zero point does not mean a true zero or absence of quantity
0°C does not mean lack of heat

We can’t say: 80°C is twice as hot as 40°C
Ratio scale
Height, weight, length, time…
We can say “100 kg is double 50 kg”
0 is meaningful (“absence of characteristics”)
v Have a true zero

v Ratio is meaningful
Variable
(Numerical)
(Quanlita<ve)
Parametric and Non-parametric test

Make assumptions about Make no assumptions
parameters of the
population
Check the normality of data

Normal distribu:on
The distribu:on gives informa:on about

• a typical value (a center) which data are spread
• the variability of values (the spread of distribu:on)
• Shape of distribu:on (whether a distribu:on is symmetric or
skewed,…)
Normal distribu:on
Symetric distribution: right half is a mirror image of left half
Asymetric distribution = skewed distribution
Normal distribu:on
Check the normality
• Graphical tests
• Analysis of Skewness and Kurtosis
• Statistical tests
• Transformation
Check the normality – Graphical tests
Histogram (a simple check):
Histogram – smaller samples:
QQ plot: scatter diagram
Box plot:
Check the normality –
Skewness and Kurtosis test
P < 0.05 : non-normal

P > 0.05: normal
Compare your data with a known distribution (normal distribution)

Check the normality – Transforma<on
• When a sample is not normally distributed
è Transform data è transformed data is more normal
• Ex: f(x) = log(x)

Check the normality – Transforma<on
• When a sample is not normally distributed
è Transform data è transformed data is more normal
• Ex: f(x) = log(x)

T- tests
• Independent sample t – test
• Paired t - test
The null – alterna:ve hypothesis
• Null hypothesis (Ho):
- What we attempt to find evidence to against our hypothesis
- P-value < α (significance level) è Reject Ho
- P-value > α è Fail to reject Ho
The null – alterna:ve hypothesis
• Alternative hypothesis (HA):
- What we attempt to demonstrate our hypothesis
- Reject Ho è accept HA
- Fail to reject Ho è do not accept HA
T-test
• Comparing two population means
• Examples:
2 methods to purify water: Method A: use a carbon filter
Method B: use a certain enzyme
µ1: mean bacteria count (method A)
µ2: mean bacteria count (method B)
We want to check if µ1 > µ2 è µ1-µ2 > 0
T-test
Are the data independent samples or dependent samples?

Independent/dependent samples
Are the data are independent samples or dependent samples?
Independent samples:
- The sample collected from one population has no relationship with
the sample collected from the other population
Dependent samples:
- Each measurement in 1 sample is matched or paired with
a particular measurement in the other sample
Independent/dependent samples
Example: Compare whether people give a higher taste rating to
Coke or Pepsi. To avoid psychological effects, people should
taste the drink blind.
A. Randomly assign half of the subjects to taste Coke and the
other half to taste Pepsi
B. All subject taste both Coke and Pepsi. The drinks should be
given in a random order.
Independent sample t-test
• Evaluate whether 2 means from 2 samples of the same
dependent variable are significantly different from one another
• Used only with interval or ratio data but not norminal nor ordinal
• Used when the data are reasonably normally distributed

Independent sample t - test
• Null hypothesis H0 “There is no significant difference between
the two groups in terms of dependent variable”
• Alternative hypothesis HA “There is a significant difference

between the two groups in terms of dependent variable”
Ex: Same dependent variable – baby birth weight
Independent variable – 2 groups of expectant mothers
- Group 1: drink < 2 bottles of water per day
- Group 2: drink > 2 bottles of water per day

Ex: Ho: No significant difference in the baby birth weight
between two mother groups
HA: There is significant difference in the baby birth weight

between two mother groups
Daily water drinking
Expectant group
The formula for the independent samples t-test
t-value =
6 10
t – value =
df = N1 + N2 -2
Sp =
Determine cri<cal t value:

Degree of Standard t value to determine
freedom significant difference or not
between 2 means
(1-α)100% Confidence interval for the difference between
two means:
• Let’s begin with Computa<on 1
• Let’s begin with Computa<on 1
Specific version
Simplified version
Need to calculate: sample size, variance

Ho: µ1 - µ2 ≤ 0; Ha: µ1 - µ2 > 0
If N1 = 30 = N2, variance = 2.0; Mean 1 = 10, mean 2 = 6
è SE = 0.395
è t = 10.95
Look for t-critical value: t (α)
(df= N1+N2-2)
• t > t-critical: reject Ho
• T < t-critical: fail to reject Ho
α = 0.05, df = 58 è t-critical value = 1.671

t = 10.95 > 1.671 è reject Ho
The mean weight of babies whose mothers drink less than 2 bottles
of water per month is statistically significantly greater than the mean
weight of babies whose mothers drink more than 2 bottles of water
per month
• Same variance
Subtracting distributions with similar variance yield more stable results
Variances must be tested for similarity

• Different variance
Subtracting distributions with similar variance yield lest stable results

Exercise: In a packing plant, a machine packs cartons with jars. It is
supposed that a new machine will pack faster on the average than
the machine currently used. To test that hypothesis, the times it
takes each machine to pack ten cartons are recorded.
Can we conclude that

the new machine
packs faster?
(α = 0.05)
Assumption 1: Independent samples? Yes
Assumption 1: Normal population? n1, n2 < 30

Assumption 3: Same variances? Yes
Rule of thumb: s1/s2 = [0.5 – 2]: same variance

≥
Example: Calculate 99% CI for the difference between the mean time
it takes the new machine to pack 10 cartons and that of the new
machine to pack 10 cartons?
df = 18
Interpret the results:

Biostatistics M1-1

Uploaded by

Copyright:

Available Formats

Biostatistics M1-1

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

Biostatistics M1-1

Uploaded by

Copyright:

Available Formats

Dr.

Nguyen Thi Van Anh

• Check normality of data

• Pearson and Spearman correlation test

Descriptive statistics Inferential statistics

• Collecting • Analyzing data

• Summarizing • Conclusion about the

• Presenting data population

Gaussian (normal) Non-gaussian (non-normal)

Parametric tests Non-parametric tests

Ex: t-test, ANOVA test Ex: Mann-Whitney test,

• Assume normal distribution • No assumed normal distribution

ü Norminal scale: use number just to distinguish categories

ü Ordinal scale: observations can be ranked

ü Interval scale: distance (not ratio) between 2 measurements is known

ü Ratio scale: equality of ratios and intervals may be determined

Gender: male = 1, female = 2

“Scale” simply labels object

Same difference between 5°C (41°F) and 15°C (59°F) as between

v The interval difference is meaningful

0°C does not mean lack of heat

v Have a true zero

Parametric tests Non-parametric tests

Check the normality of data

The distribu:on gives informa:on about

• Analysis of Skewness and Kurtosis

P < 0.05 : non-normal

Compare your data with a known distribution (normal distribution)

• When a sample is not normally distributed

è Transform data è transformed data is more normal

• Ex: f(x) = log(x)

è Transform data è transformed data is more normal

• Ex: f(x) = log(x)

Are the data independent samples or dependent samples?

• Used when the data are reasonably normally distributed

• Alternative hypothesis HA “There is a significant difference

Independent variable – 2 groups of expectant mothers

- Group 1: drink < 2 bottles of water per day

- Group 2: drink > 2 bottles of water per day

HA: There is significant difference in the baby birth weight

Daily water drinking

Determine cri<cal t value:

Need to calculate: sample size, variance

α = 0.05, df = 58 è t-critical value = 1.671

Subtracting distributions with similar variance yield more stable results

Variances must be tested for similarity

Subtracting distributions with similar variance yield lest stable results

Can we conclude that

Assumption 1: Independent samples? Yes

Assumption 1: Normal population? n1, n2 < 30

Assumption 3: Same variances? Yes

Rule of thumb: s1/s2 = [0.5 – 2]: same variance

Interpret the results:

You might also like