Analysis
Descriptive statistics Inferential statistics
Central Tendencies Estimation
Deviations Test of Significance
Locations
Correlation
Measures of Central Tendencies
Arithmetic Mean
Simple average (µ, x)
Median
Middle value
Mode
Peakness of data
Measures of Deviations
Absolute Relative
Range Coefficient of Range
Quartile Deviation Coefficient of Q.D.
Mean Deviation Coefficient of M.D.
Standard Deviation (σ) Coefficient of Variance
Obtain Mean Median Mode Range SD
and CV for the following Hb data
3.2 7.5 9.3 2.9 5.6 7.4 8 9.1
Arranging ascending order
2.9 3.2 5.6 7.4 7.5 8 9.1 9.3
Correlation
• Strength of association between two numerical
(quantitative) variables(or more variables)
Correlation
Pearson’s
(Product moment)
Spearman’s (Rank Correlation)
“r” “”
Obtain correlation coefficient for the
following data.
Height 140 143 157 160 164 170
weight 40 44 41 50 53 60
x y x2 y2 xy
140 40 19600 1600 5600
143 44 20449 1936 6292
157 41 24649 1681 6437
160 50 25600 2500 8000
164 53 26896 2809 8692
170 60 28900 3600 10200
934 288 146094 14126 45221
= 0.8452
Classification of Correlation Coefficient
• Correlation coefficient denoted by ‘r’
• ‘r’ value always lies between -1 to +1
• If ‘r ‘is
Between 0 to 0.25 No correlation
Between 0.25 to 0.50 Low correlation
Between 0.50 to 0.75 Moderate correlation
Between 0.75 to 1 High correlation
+1 or -1 Perfect Correlation
Scatter Diagram
• Graphical method to find correlation between two variables
65
60 60
55
53
Weight
50 50
45
44
41
40 40
35
135 140 145 150 155 160 165 170 175
Height
Confidence Intervals (CI)
• for population prediction (using SE)
• for sample illustration (using SD)
SE = SD/ n
• 68% (mean ± SD)
• 95% (mean ± 2*SD)
• 99% (mean ± 3*SD)
Population Prediction
Eg. From our research we found mean DPB in 3
trimester of pregnant women was 88 mg/dl.
Now we interested to generalize/forecast this result
to our whole population with 95% or 99%
confidence…
95% (80 to 96)
99% (78 to 102)
Sample Illustration
• 68% (mean ± SD)
• 95% (mean ± 2*SD)
• 99% (mean ± 3*SD)
Mean Heights and Standard Errors (SE) of 50
boys and 50 girls are given below. Find out
the 95% confidence limits within which the
height of all boys and girls would lie.
Mean S.E
Boys 151.6cm 0.89cm
Girls 147.4cm 0.93cm
Test of Significance
• To check the significance between two or
more statistics from different samples or
sample statistic with corresponding
population parameter
Two samples
Population 1 Population 2
Mean 1 or Mean 2
Sample 1 Sample 2
One Sample
Population 1 ParameterPopulation 2
Statistic or Parameter
Statistic
Sample 1
One Sample (two observations)
Population 1 Population 2
Mean1 or Mean2
Mean 1
Intervention
Treatment
Sample Mean 2
Leukaemia
Total
Present Absent
Computer Yes
Work No
Total
Is there any association between computer work
and leukaemia.
Some tests namely
t-test
F-test
Chi-square test
ANOAVA
Wilcoxon test
Mann-whitney test
Kruskal Wallis test, etc….
Statistical Tests
Parametric Non-parametric
Non-Numeric
Non-Normal
Non-Random
Hierarchies of Testing of Hypothesis
• Hypothesis setting from research problem
• Test value from appropriate test
• Tabulated value from statistical tables
• Comparison between test and table values
• Interpretation
Chi-Square ( 2)test
• To find the association between two factors
• To find the significance between two or more
proportions
• To check the homogeneity of a variable
• To check the goodness of fit
A study was conducted, 1000 persons
working with computers and 2000
matched individuals not so working
were followed up for 20 years to find
out how many developed leukaemia.
Ten in former group and five in latter
group developed the disease. Did
leukaemia and computer works are
associated? (table value 2(1) 5%=3.84)
H0: There is no association between computer work
and leukaemia.
Leukaemia
Total
Present Absent
Computer Yes 990
Work No
1995
Total 2985
Test statistics
O E O-E (O-E)2
10 5 5 25 5
990 995 -5 25 0.0251
5 10 -5 25 2.5
1995 1990 5 25 0.0126
3000 3000 7.5377
From the chi-square distribution table
Here,
Calculated χ2 > Table χ2
7.5377 > 3.86
Therefore H0 is rejected.
Thus, there is an evidence for association
between Computer work and leukaemia
t-test
• To compare sample mean with population
mean (standard value).
• To compare two sample mean values.
• To compare paired sample mean values.
• Mean/ Corr. Coeff./ Proportion
t-test for single mean
• The mean weight of 26 pre school children was
17.8kg with a standard deviation of 1.25kg. Does
this mean weight differ from the population mean
weight of 18.5? (t(25)5% =2.060)
Given,
H0: There is no significant difference between the sample
and the population mean values.
Test Statistics
1.625
S
Tabulated value from t-distribution table,
t(n-1)5% t(26-1)5%
t(25)5%
2.060
Here, Calculated “t” > tabulated “t”
i.e. 2.7999 > 2.060
we reject the H0.
Thus, the sample mean and population mean are
significantly differ.
t-test for two sample means
• Details of study on arm circumference (cm) of American
and Indian pre-school children are given below. Can we
infer that arm circumferences is differ between American
and Indian children? (t(50)5% =2.014)
American Indian
Details
children children
Sample size 26 26
Mean 20.5 19.1
SD 4.3 3.1
H0: There is no significant difference between
the two sample mean values.
Test Statistics
\We accept the H0.
Thus, the two sample mean values are not
significantly differing.
Paired t-test
• In a sample of 6 subjects who were on high intake
of calories, the gain in calf-circumference as
observed subject wise is given below:
Subject 1 2 3 4 5 6
Gain in Calf
11 10 9 7 12 8
Circumference (cm)
Can we conclude that there is a significant gain in calf
circumference against nil effect? (t(5)5% =2.571)
H0: There is no significant gain in calf circumference
Test Statistics
d
11 1.5 2.25
10 0.5 0.25
9 -0.5 0.25
7 -2.5 6.25
12 2.5 6.25
8 -1.5 2.25
57 Total 17.5
Tabulated value from t-distribution table,
t(n-1)5% t(6-1)5%
t(5)5%
2.571
Here, Calculated “t” > tabulated “t”
i.e. 12.4386 > 2.571
we reject the H0.
Thus, there is a significant gain in calf circumference.
Sampling Error
Reality / State of Nature
Hypothesis Testing Null Null
Hypothesis is Hypothesis is
True False
Fail to Reject
Null Correct Type II Error
(1-α) β
Hypothesis
Our research
decision
Reject Null
Hypothesis Type I Error Correct
α (1-β)
“α” is the size of type I error. Also know as level of significant,
critical region, region of rejection.
“1-α” is known as confidence level.
“β” is the size of type II error.
“1-β” is known a power level
[Link]