Biostatistics
description
Transcript of Biostatistics
Dr. Priya narayanPost graduate studentDepartment of oral pathology & microbiologyRajarajeswari dental college & hospital.
Biostatistics
Contents:-
Introduction
Measures of central tendency
Measures of dispersion
The normal curve
Tests of significance
References
Introduction
‘Statistics’ – Italian word ‘statista’ meaning ‘statesman’ or the
German word ‘statistik’ which means ‘a political state’.
Originated from 2 main sources:
Government records
Mathematics
Registration of heads of families in ancient Egypt & Roman census
on military strength, births & deaths, etc.
John Graunt (1620-1674) – father of health statistics
STATISTICS : is the science of compiling, classifying and
tabulating numerical data & expressing the results in a
mathematical or graphical form.
BIOSTATISTICS : is that branch of statistics concerned with
mathematical facts & data related to biological events.
Uses of statistics
To assess the state of oral health in the community and to
determine the availability and utilization of dental care facilities.
To indicate the basic factors underlying the state of oral health by
diagnosing the community and solutions .
To determine success or failure of specific oral health care
programs or to evaluate the program action.
To promote health legislation and in creating administrative
standards.
MEASURES OF CENTRAL TENDENCY
A single estimate of a series of data that summarizes the data -
the measure of central tendency.
Objective:
To condense the entire mass of data.
To facilitate comparison.
PROPERTIES
Should be easy to understand and compute.
Should be based on each and every item in the series.
Should not be affected by extreme observations.
Should be capable of further statistical computations.
Should have sampling stability.
The most common measures of central tendeny that are used
in dental sciences are :
Arithmetic mean – mathematical estimate.
Median – positional estimate.
Mode – based on frequency.
Arithmetic mean Simplest measure of central tendency.
Ungrouped data: =
Grouped data:
=
Mean Sum of all the observations in the data
Number of observations in the data
Mean Sum of all the variables multiplied by the
corresponding frequency in the data
Total frequency
MEDIAN:-
Middle value in a distribution such that one half of the units in
the distribution have a value smaller than or equal to the median
and one half have a value greater than or equal to the median.
All the observations are arranged in the order of the magnitude.
Middle value is selected as the median.
Odd number of observations : (n+1)/2.
Even number of observations: mean of the middle two values is
taken as the mean.
MODE
The mode or the modal value is that value in a series of
observations that occurs with the greatest frequency.
When mode is ill defined, it can be calculated using the relation
Mode = 3 median – 2 mean
Most commonly used: arithmetic mean.
Extreme values in the series : median.
To know the value that has high influence in the series: mode.
Measures of dispersion
Dispersion is the degree of spread or variation of the variable
about a central value.
Measures of dispersion used:
To determine the reliability of an average.
To serve as basis for control of variability.
To compare two or more series in relation to their variability.
Facilitate further statistical analysis.
RANGE
It is the simplest method, Defined as the difference between the
value of the smallest item and the value of the largest item.
This measure gives no information about the values that lie
between the extremes values.
Subject to fluctuations from sample to sample.
MEAN DEVIATION
It is the average of the deviations from the arithmetic mean.
M.D = ƩX – Xi , where Ʃ ( sigma ) is the sum of, X is the
n arithmetic mean, Xi is the value of each
observation in the data, n is the number of observation in the
data.
STANDARD DEVIATION(SD)
Most important and widely used.
Also known as root mean square deviation, because it is the
square root of the mean of the squared deviations from the
arithmetic mean.
Greater the standard deviation, greater will be the magnitude of
dispersion from the mean.
A small SD means a higher degree of uniformity of the
observations.
CALCULATION
For ungrouped data: Calculate the mean(X) of the series. Take the deviations (d) of the items from the mean by : d=Xi – X,
where Xi is the value of each observation. Square the deviations (d2) and obtain the total (∑ d2) Divide the ∑ d2 by the total number of observations i.e., (n-1) and
obtain the square root. This gives the standard deviation. Symbolically, standard deviation is given by:
SD= √ ∑ d2 /(n-1)
For grouped data with single units for class intervals:
S = √∑(Xi - X) x fi / (N -1)
Where,
Xi is the individual observation in the class interval
fi is the corresponding frequency
X is the mean
N is the total of all frequencies
• For grouped data with a range for the class interval:
S =√ ∑(Xi - X) x fi / (N -1)
Where,
Xi is the midpoint of the class interval
fi is the corresponding frequency
X is the mean
N is the total of all frequencies
COEFFICIENT OF VARIATION(C.V.)
A relative measure of dispersion.
To compare two or more series of data with either different units
of measurement or marked difference in mean.
C.V.= (Sx100)/ X
Where, C.V. is the coefficient of variation
S is the standard deviation
X is the mean
Higher the C.V. greater is the variation in the series of data
NORMAL DISTRIBUTION CURVE
Gaussian curve
Half of the observations lie above and half below the mean
– Normal or Gaussian distribution
Properties
Bell shaped.
Symmetrical about the midpoint.
Total area of the curve is 1. Its mean zero & standard deviation 1.
Height of curve is maximum at the mean and all three measures of
central tendency coincide.
Maximum number of observations is at the value of the variable
corresponding to the mean, numbers of observations gradually
decreases on either side with few observations at extreme points.
Area under the curve between any two points can be found out in
terms of a relationship between the mean and the standard
deviation as follows:
Mean ± 1 SD covers 68.3% of the observations
Mean ± 2 SD covers 95.4% of the observations
Mean ± 3 SD covers 99.7% of the observations
These limits on either side of mean are called confidence limits.
Forms the basis for various tests of significance .
TESTS OF SIGNIFICANCE
Different samples drawn from the same population, estimates
differ – sampling variability.
To know if the differences between the estimates of different
samples is due to sampling variations or not – tests of
significance.
Null hypothesis
Alternative hypothesis
NULL HYPOTHESIS
There is no real difference in the sample(s) and the
population in the particular matter under consideration
and the difference found is accidental and arises out of
sampling variation.
ALTERNATIVE HYPOTHESIS
Alternative when null hypothesis is rejected.
States that there is a difference between the two groups
being compared.
LEVEL OF SIGNIFICANCE
After setting up a hypothesis, null hypothesis should be either
rejected or accepted.
This is fixed in terms of probability level (p) – called level of
significance.
Small p value - small fluctuations in estimates cannot be
attributed to sampling variations and the null hypothesis is
rejected.
STANDARD ERROR
It is the standard deviation of a statistic like the mean, proportion
etc
Calculated by the relation
Standard error of the population = √(p x q)/ n
Where,
p is the proportion of occurrence of an event in the sample
q is (1-p)
n is the sample size
TESTING A HYPOTHESIS
Based on the evidences gathered from the sample
2 types of error are possible while accepting or rejecting a null
hypothesis
Hypothesis Accepted Rejected
True Right Type I error
False Type II error Right
STEPS IN TESTING A HYPOTHESIS
State an appropriate null hypothesis for the problem.
Calculate the suitable statistics.
Determine the degrees of freedom for the statistic.
Find the p value.
Null hypothesis is rejected if the p value is less than 0.05,
otherwise it is accepted.
TYPES OF TESTS :-
PARAMETRIC
i. student’s ‘t’ test.
ii. One way ANOVA.
iii. Two way ANOVA.
iv. Correlation coefficient.
v. Regression analysis.
NON- PARAMETRIC
i. Wilcoxan signed rank test.
ii. Wilcoxan rank sum test.
iii. Kruskal-wallis one way
ANOVA.
iv. Friedman two way ANOVA.
v. Spearman’s rank correlation.
vi. Chi-square test.
CHI- SQUARE(ᵡ2) TEST
It was developed by Karl Pearson.
It is the alternate method of testing the significance of
difference between two proportions.
Data is measured in terms of attributes or qualities.
Advantage : it can also be used when more than two groups
are to be compared.
Calculation of ᵡ2 –statistic :-
ᵡ2 = Ʃ ( O – E )2
EWhere, O = observed frequency and E = expected frequency.
Finding the degree of freedom(d.f) : it depends on the number of columns & rows in the original table.
d.f = (column -1) (row – 1). If the degree of freedom is 1, the ᵡ2 value for a probability of
0.05 is 3.84.
CHI-SQUARE WITH YATE’S CORRECTION
It is required for compensation of discrete data in the chi-square
distribution for tables with only 1 DF.
It reduces the absolute magnitude of each difference (O- E) by half
before squaring.
This reduces chi- square & thus corrects P( i.e., result significance).
Formula used is :
ᵡ2 = Ʃ[ ( O – E ) – ½]2
E
It is required when chi-square is in borderline of significance.
LIMITATIONS :-
It will not give reliable result if the expected frequency in any one cell is less than 5.
In such cases, Yates’ correction is necessary i.e , reduction of the (O-E) by half.
X2 = ∑[(O-E) – 0.5]2
E The test tells the presence or absence of an association between
the two frequencies but does not measure the strength of association.
Does not indicate the cause & effect. It only tells the probability of occurrence of association by chance.
STUDENT ‘T’ TEST :-
When sample size is small. ‘t’ test is used to test the hypothesis. This test was designed by W.S Gosset, whose pen name was
‘student’. It is applied to find the significance of difference between two
proportions as, Unpaired ‘t’ test. Paired ‘t’ test. Criterias : The sample must be randomly selected. The data must be quantitative. The variable is assumed to follow a normal distribution in the
population. Sample should be less than 30.
PAIRED ‘T’ TEST
When each individual gives a pair of observations.
To test for the difference in the pair values.
Test procedure is as follows:
Null hypothesis
Difference in each set of paired observation calculated : d=X1 – X2
Mean of differences, D =∑d/n, where n is the number of pairs.
Standard deviation of differences and standard error of difference
are calculated.
Test statistic ‘t’ is calculated from : t=D/SD/√n
Find the degrees of freedom(d.f.) (n-1)
Compare the calculated ‘t’ value with the table value for
(n-1) d.f. to find the ‘p’ value.
If the calculated ‘t’ value is higher than the ‘t’ value at 5%,
the mean difference is significant and vice-versa.
ANALYSIS OF VARIANCE(ANOVA) TEST :-
When data of three or more groups is being investigated.
It is a method of partionioning variance into parts( between &
within) so as to yield independent estimate of the population
variance.
This is tested with F distribution : the distribution followed by the
ratio of two independent sample estimates of a population
variance.
F = S12/ S2
2 .The shape depends on DF values associated with S12 &
S22 .
One way ANOVA : if subgroups to be compared are
defined by just one factor.
Two way ANOVA : if subgroups are based on two
factors.
Miscellaneous :- Fisher’s exact test :
A test for the presence of an association between categorical
variables.
Used when the numbers involved are too small to permit the use
of a chi- square test.
Friedman’s test :
A non- parametric equivalent of the analysis of variance.
Permits the analysis of an unreplicated randomized block design.
Kruskal wallis test : A non-parametric test. Used to compare the medians of several independent samples. It is the non-parametric equivalent of the one way ANOVA.
Mann- whitney U test : A non-parametric test. Used to compare the medians of two independent samples.
Mc Nemar’s test : A variant of a chi squared test, used when the data is paired.
Tukey’s multiple comparison test :
It’s a test used as sequel to a significant analysis of variance test,
to determine which of several groups are actually significantly
different from one another.
It has built-in protection against an increased risk of a type 1 error.
Type 1 error : being misled by the sample evidence into rejecting
the null hypothesis when it is in fact true.
Type 2 error : being misled by the sample evidence into failing to
reject the null hypothesis when it is in fact false.
REFERENCES
Park K, Park’s text book of preventive and social
medicine, 21st ed, 2011, Bhanot, India; pg- 785-792.
Peter S, essential of preventive and community dentistry,
4th ed; pg- 379- 386.
Mahajan BK, methods in biostatistics. 6th edition.
John j, textbook of preventive and community dentistry,
2nd ed; pg- 263- 68.
Mahajan BK, methods in biostatistics. 6th edition.
Prabhkara GN, biostatistics; 1st edition.
Rao K Visweswara, Biostatistics – A manual of
statistical methods for use in health, nutrition &
anthropology. 2nd edition.2007.
Raveendran R, Gitanjali B, A practical approach to PG
dissertation.2005.