You are on page 1of 23

BIOSTATISTICS

Dr. Kanchan Chitnis

DEFINITION
Collection, organization, summarization and analysis of data, and Drawing inferences Data derived from biological sciences and medicine is analyzed = Biostatistics

CHARACTERISTICS
Quantitative Discrete Ex: blood cell count Continuous Ex: height, weight Qualitative Nominal Ex: male/female, dead/alive Ordinal Ex: mild/moderate/severe better/same/worse

POPULATION AND SAMPLE


Sample Statistics Mean = x bar SD = s Population Parameters Mean = SD =

Characteristics of representative sample


Precision =Nearness of sample statistics to population parameters, n = 4(sd)2 L2 L= allowable error Problem: Mean pulse rate of population is 70/ min, with SD of 8 beats. Calculate size of sample to verify this if L = +2 beats, and when L = +1 beat.
5

Characteristics of representative sample


Control and experimental group must be there if possible

Sample should be unbiased Ex: Survey of prevalence of TB in a well to do locality does not give correct picture

NORMAL DISTRIBUTION
Large number of observations, small group intervals Mean + 1 SD limits include 68% of obs Mean + 1.96 SD limits include 95% of obs Mean + 2.58 SD limits include 99% of obs Binomial and Poisson distribution

STANDARD ERROR
Chance variation from sample to sample or from sample to population In any sampling distribution; + 1 SE limits include 68% of obs
+ 1.96 SE limits include 95% of obs + 2.58 SE limits include 99% of obs

This sampling distribution forms the basis of tests of significance


8

ESTIMATION OF POPULATION PARAMETER


Point estimate: Single statistic like mean Confidence interval: Range with attached probability Ex: Mean +1.96 SE =95% confidence interval and population mean would lie in that range in 95% cases Any values outside that range are rare, probability of their occurrence by chance is 5% ( p= 0.05)
9

TESTING STATISTICAL HYPOTHESES


Null hypothesis (Ho): of no difference Alternate hypothesis (H1): of significant difference Zone of rejection Zone of acceptance

10

TYPE I AND TYPE II ERRORS


Type I error ( error): Ho rejected when estimate falls in zone of acceptance of Ho We change level from 5 to 6 to 8 % Serious error Minimize by lowering level of significance Type II ( error): Ho accepted when estimate falls in zone of rejection We change level from 5 to 4 to 3 % Not serious error Needs confirmation by changing level and increasing size of sample

11

TO SUMMARIZE
Actual Situation Actual Situation Decision Ho rejected Ho accepted Ho is true Type I error No error Ho is false No error Type II error

12

Type II error and Power of a test


= type II error 1- =power of a test = probability that we reject false Ho = we will take correct action when Ho is false

13

To determine sample size to control type II errors


n=[(Zo + Z1) /(0 1)]2 Problem

14

To determine sample size to estimate means


Size of sample depends on , degree of reliability and desired interval width( d) d= reliability coefficient x SE If sampling is with replacement, d= z /n; n = z2 2/ d2..(1) If sampling is without replacement, n =N z2 2/ d2(N-1)+ z2 2 ..(2)
15

Sources of estimates for 2


A pilot sample may be drawn from population and variance computed may be used Estimates may be available from previous studies Problem

16

To determine sample size to estimate proportions


p=proportion in population possessing the characteristic of interest q=1-p = not possessing n = z2pq/d2 Problem

17

TESTS OF SIGNIFICANCE
Parametric tests Student`s t test Z-test Pearson correlation ANOVA Non parametric tests Chi square Fischer`s exact test Spearman`s r Wicoxon signed rank

18

Z test
Z p 1.6 0.1 2.0 2.3 2.6

0.05 0.02 0.01

Sample size more than 30 Problem

19

T test
Unpaired t test Experiment and control problem Paired t test Before and after treatment problem

20

Chi square test


For discrete data To find association of two events, goodness of fit to theoretical values Problem

21

OTHER TESTS of SIGNIFICANCE


Correlation coefficient tells strength of association between two continuous variables Regression coefficient helps to predict change in dependent variable F ratio ( ANOVA) used to compare means of more than 2 samples

22

Thank you

23

You might also like