hypothesis testing lectures

20
HYPOTHESIS TESTING PRESENTED BY: DR SANJAYA KUMAR SAHOO PGT,AIIH&PH,KOLKATA

Upload: sanjaya-sahoo

Post on 05-Dec-2014

93 views

Category:

Health & Medicine


0 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Hypothesis testing lectures

HYPOTHESIS TESTING PRESENTED BY:

DR SANJAYA KUMAR SAHOO PGT,AIIH&PH,KOLKATA

Page 2: Hypothesis testing lectures

HYPOTHESIS TESTING:

OBJECTIVE:

To test whether evidence for assumption or statements we make

about our research objectives(i.e alternative hypothesis) against the

previous or existing history of that particular research objective(i.e

null hypothesis)

Page 3: Hypothesis testing lectures

What is hypothesis?

• Hypothesis is a belief concerning a parameter (i.e statistical constant) such as prevalence,incidence,population mean,correlation etc.

• Example: incidence of TB is higher among low SES groups as compared to

higher SES groups.Incidece of lung cancer is higher among smokers as compared to non-

smokers.

Page 4: Hypothesis testing lectures

5

TESTING OF HYPOTHESES:

• states a hypothesis to be tested,• formulates an analysis plan,• analyzes sample data according to

the plan, and • accepts or rejects the hypothesis,

based on results of the analysis.

The researcher

Page 5: Hypothesis testing lectures

The scientific hypothesis, is of 2 types:

1.null hypothesis (denoted by H0)

2.alternative hypothesis (denoted by H1).

Thus the alternative hypothesis is the assertion accepted when null hypothesis is rejected

Page 6: Hypothesis testing lectures

NULL HYPOTHESIS (H0):

Null hypothesis is generally a statement that assumes that, there is “NO”

difference between two sets of values.

Examples:

• There is no difference between 2 drugs A & B.

• There is no association between lung cancer and smoking.

• Mean cholesterol level in normals =mean cholesterol level in HTN patients.

Page 7: Hypothesis testing lectures

ALTERNATIVE HYPOTHESIS(H1)

• It states that there is difference between two sets of values.

Examples:

• There is difference between 2 drugs A & B.

• There is an association between lung cancer and smoking.

• Mean cholesterol values in normals ≠ mean cholesterol in HTN patients.

Page 8: Hypothesis testing lectures

TYPES OF ERRORS:TYPE 1 ERROR:

When null hypothesis H0 is true but still it is rejected type 1 error(alpha error) is committed .

It is represented as α(alpha) or p-value or level of significance.

Usually P is less than 5 in a hundred(p<0.05)

TYPE II ERROR: When null hypothesis H0 is false but it is accepted/fails to reject type-2 error (beta error) is

committed.

Power of test (1-β) is the ability of test to correctly reject H0(i.e no diff between the drugs)

in favour of H1(i.e there is a difference) when H0 is false.

Page 9: Hypothesis testing lectures

H0 is true

H0 rejected

H0 is false

H0 accepted

Page 10: Hypothesis testing lectures

H0 -> μAtorva(A)= μRosuva(B)

Ha->μAtorva(A)≠ μRosuva(B) (2 tailed)

Ha->μAtorva(A)> μRosuva(B) (1 tailed)Ha->μAtorva(A)< μRosuva(B) (1 tailed)

Page 11: Hypothesis testing lectures

The power (1-β) of a study   The probability that it correctly

rejects the null hypothesis(A=B) when the null hypothesis is false.

accepts the alternative hypothesis(A≠B) when the alternative hypothesis is

true –(It assumes that there is a difference)

identifies a significant difference or effect or association in the sample that

exists in the population.

The larger the sample size, the study will have greater power• “Power” is the antithesis of “risk of Type II error”

• Risk of Type II error = 1 – power

• Power = 1 - Risk of Type II error

Page 12: Hypothesis testing lectures

Type(Alpha) Error Z=standardized normal deviate

P Z(alpha)

0.05 1.96

0.01 2.57

0.001 3.29

Type(Beta) Error

p Z(beta)

0.10 1.282

0.15 1.037

0.20 0.842

0.25 .675

Page 13: Hypothesis testing lectures

Hypothesis testing

When mean is the parameter of the study and

σ (sigma)→ pooled standard deviation/common sd

between 2 groups

d → clinically meaningful difference between two

groups/difference between 2 means

Zα → Z value for α level (type I error)

Zβ → Z value for β level (type II error)

Page 14: Hypothesis testing lectures

Estimation of differences between two means (on quantitative variable like BP, blood sugar, serum cholesterol etc.)  

Requirements:

Estimate of variables of individual values (means & standard deviations)

Magnitude of differences that is desired to detect(limit of accuracy required)

Degree of confidence requiredThe value of the power desired (1-)

Larger the difference, smaller the trial size.

Greater the Power, greater the trial size.

Larger the s.d., greater the trial size. 

Page 15: Hypothesis testing lectures

Simple formula When mean is the parameter of the study :

Sample size in each group (assumes equal sized groups)

Represents the desired power (typically .84 for 80% power).

Represents the desired level of statistical significance (typically 1.96).

Common Standard deviation between two groups

σ2 =(n1-1)σ12 +(n2-1)σ2

2/(n1+n2)-2

Effect Size (the difference in means)

2

2/2

2

difference

)Z(2

Zn

Page 16: Hypothesis testing lectures

Hypothesis testing:

When proportion is the parameter of the study and

p1 → proportion in the first group

p2 → proportion in the second group

P

d (p1 -

p2 )

→ clinically meaningful difference between proportion of two

groups

Zα → Z value for α level (type I error)

Zβ → Z value for β level (type II error)

Average of P1 & P2

Page 17: Hypothesis testing lectures

Estimation of difference between two proportions

Requirements:

• An estimate of response rate in two groups

• Difference in response rates

• Level of statistical significance ()• The value of the power desired (1-)• Whether the test should be one sided or two sided

1.Larger the difference, smaller the trial size2. Larger the Power, larger the trial size3. Absolute value of average P, also affects trial size

Page 18: Hypothesis testing lectures

Simple formula when proportion is the parameter of the study

221

2/2

)(p

)Z)(1)((2

p

Zppn

Sample size in each group (assumes equal sized groups)

Represents the desired power (typically .84 for 80% power).

Represents the desired level of statistical significance (typically 1.96).

A measure of variability (similar to standard deviation)

Effect Size (the difference in proportions)

Page 19: Hypothesis testing lectures

EFFECT SIZE Crucial factor : comparative studies

Smallest measured difference between the comparison groups

Unknown, selection of a reasonable difference derived from pilot studies/lit review/past experience

ES or the clinically meaningful difference is determined by the investigator, not by statistician

For small ES sample size is large

The investigator may choose a minimum expected difference of 10%(.1) and this is effect size.

21

Page 20: Hypothesis testing lectures

THANK YOU