Beruflich Dokumente
Kultur Dokumente
a hypothesis
Learning material:
Chapters 9-11: Confidence Interval
Estimation
Chapters 12-14: Testing hypotheses
Learning Objectives
Clinical
observation
IDEAS
Imagination
Reasoning
Hypotheses
(Proposed answers)
}
Experimental studies
Observational studies
Tests of
hypothesis
EVIDENCE-BASED INFORMATION
4
Examples:
Mechanisms of disease at the molecular level
Drugs against antibiotic-resistant bacteria are developed through
knowledge of the mechanism of resistance
Learning Objectives
The p-value
Example
H0 : The coin is fair (i.e. will produce as many heads as tails)
=> population proportion =0.5
Outcome: 42 heads out of 100 times tossing =>
sample proportion p=0.42.
Is the difference between 0.5 and 0.42 statistically significant
or due to chance? How do we decide what proportion of heads
we might expect to get if the coin is fair?
Matter of interest: P(heads 42 | coin is fair)=0.067 > 0.05
=> the coin is most likely fair
p-value
P(heads 41 | coin is fair) = 0.044 < 0.05
=> the coin is most likely not fair
p-value
if outcome: 41 heads
10
Definition
The p-value is defined as the probability of obtaining the
result, or a more extreme result, if the null hypothesis is true.
Interpretation of p-value:
We usually say, if p 0.05, then the results are
statistically significant (or unlikely due to chance), and
there is strong enough evidence against the null hypothesis.
11
Example
Odds Ratios of antimicrobial resistance in hospitalacquired versus community-acquired infections
<0.05
>0.05
For Ceftazidim the difference in Odds Ratios for Denmark (OR=15.3)
and Israel (OR=3.4) is statistically significant (p<0.05).
For Ceftriaxone there is no evidence to conclude that the difference
(between OR=8.0 and OR=3.7) is statistically significant (p>0.05).
12
Power
14
To define Power
Power depends on:
significance level ()
sample size
minimum size of the effect that would be clinically
useful
Fix (= 5%) and then try to minimize (maximize 1 - )
Keeping other variables constant, the best way to increase
power is to increase sample size
(Studies with small sample sizes tend to have low power)
Typical power is 80% or greater
You can compute the power for a study that you are about
to run (good idea!), or you can compute post-hoc power
after an experiment is done.
15
Example
Conclusion:
16
> 0.05 (= )
probability of observing 5% difference or more,
when the null hypothesis is true
p>5% => do not reject null hypotesis
17
18
Test statistic
19
x 0
t
s/ n
difference in means
difference in means
variability
statistical difference
0
variability
20
Rejection area
H0: = 0
one sample t-test statistic
x 0
t
s/ n
t*
critical value
Ha: >0
=.05
Two-tailed
Ha: 0
=.025
=.025
22
Collect samples data and determine sample statistics (e.g., sample mean,
sample proportion,).
If p-value < significance level then reject H0; otherwise do not reject H0.
23
Statistical test
There are many different statistical tests. The choice
of which test to use depends on several factors:
24
Assumptions:
Random sample
Metric sample data
Normally distributed sample data
(although this assumption is less critical when the sample size 30)
H0: = 0
test statistic=
x 0
t
s/ n
25
Sample:
_
random sample
metric data
x 1.2 mg / dL
s = 0.352 mg / dL
se= 0.088 mg / dL
checking normality:
2-tailed test
Test statistics: t = (1.2-0.9) / 0.088 = 3.4
(=> the results are 3.4 standard error units higher the expected value).
Use a statistic table to look up the t-distribution and find
critical value t*.
28
df
t distribution depends on
degrees of freedom: df = n-1
29
Calculate p-value:
t=3.4
t=-3.4
t*
P-value < a
31
Learning Objectives
32
Alternatively: Estimation
Population
Mean, , is
unknown
Random Sample
Mean
X = 1.2
I am 95%
confident that
is between
1.01 & 1.38.
Sample
X 1.96
n
34
X t * s
n
35
Graphical presentation of a CI
37
39
40
Learning Objectives
41
42