Medical Statistics, Informatics and Health Economics

INNSBRUCK MEDICAL UNIVERSITY

Basics in Statistics

(Descriptive Statistics and Data Management)

Study Design
and
Estimation

Michael Edlinger
1
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

First session
Descriptive Statistics and Data Management

Statistics is about random events

Working with samples
Data types: nominal, ordinal, numerical
Description (summarising):
• graphs
• measures: central tendency and variability
Data management
Statistical programmes

2
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Objectives
Study Design and Estimation

Understanding of basic medical statistical and study design notions

Interpreting statistical specifications in papers

Appreciating basics of statistical inference

Applying statistical software at a basic level

3
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Literature
Median or mean?
Measures of variablility
Quantiles, cumulative distribution function, and box-plot
The 2 by 2 table
What is a confidence interval?
What is a significance test? General issues
Common significance tests

4
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Contents
Probability theory

Estimation

Study design

References and links

Exercises

Test (separate)

5
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Probability theory
Basics of sampling theory (warnings)

"All the techniques and theory [that we will discuss] apply to

random samples, and do not necessarily hold for any data
collected in any way" (Hays, 1972)

"Virtually any study can be made to show significant results if one

uses enough subjects, regardless of how nonsensical the content
may be" (Hays, 1973)

6
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Probability
The probability (p) that a certain event (A) will take place: p(A)

The event depends on chance, e.g. throwing a 5 with a dice

0 ≤ p(A) ≤ 1

p() = 1

If (BC) = , then p(BC) = p(B) + p(C)

7
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Probability
Algorithms:
• p() = 1 - p(A)
• p(BC) = p(B) + p(C) - p(BC)

Conditional probability: p(B|C)

e.g. probability of a certain diagnosis given a particular symptom
drawing a ticket in a lottery (without replacement)

Two events are independent of each other when:

p(B|C) = p(B)

8
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Bayes' theorem
Satz von Bayes

p(B|C) = p(BC) / p(C)

= p(C|B) * p(B) / p(C)

Applicable e.g. in medical diagnostics

9
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Bayes' theorem
Satz von Bayes

Example

With test T und disease D and

p(T+|D+) = 0.80
p(T+|D-) = 0.30
p(D+) = 0.01  p(D-) = 0.99

Then if T is positive and

p(D+|T+) = p(T+|D+) * p(D+) / p(T+) and
p(T+) = p(T+|D+) * p(D+) + p(T+|D-) * p(D-)

 p(D+|T+) = 0.80 * 0.01 / ((0.80*0.01) + (0.30*0.99)) = 0.03

10
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Bayes' theorem
Satz von Bayes

More examples

With test T und disease D and

p(T+|D+) = 0.80
p(T+|D-) = 0.30
p(D+) = 0.01  p(D-) = 0.99

p(D+) = 0.10  p(D-) = 0.90

 p(D+|T+) = 0.80 * 0.10 / ((0.80*0.10) + (0.30*0.90)) = 0.23

p(D+) = 0.30  p(D-) = 0.70

 p(D+|T+) = 0.80 * 0.30 / ((0.80*0.30) + (0.30*0.70)) = 0.53

11
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Distributions
Discrete probability distributions:
• Poisson
• Bernoulli
• binomial
• geometric
• negative binomial

Continuous probability distributions:

• Normal
• exponential
• log-normal
• chi-squared
• Student's t
• F
12
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Normal distribution

f(x) 0,25

0,20

0,15

0,10
f(x)
0,05

0,00

13
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Normal distribution
N(,2): - < X < 

Expected value =  and variance = 2

Symmetrical: mean = median = mode

Standardisation:
X 
Z 

Many practical applications, important characteristics, basis for
various test distributions

14
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

15
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Estimation
Exploratory versus confirmatory research
Descriptive versus inferential research

Medical research: extrapolation of findings from a sample of

individuals to the population of all similar individuals,
to draw general conclusions (inference)

Two basic approaches:

1. hypothesis testing
2. estimation

16
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Hypothesis testing:
• H0: the null hypothesis that there is no effect
• H1: the alternative hypothesis that there is an effect

What is the probability that we could have obtained the observed

data (or data that were more extreme) if H0 were true?

Testing (truth unknown): this probability (p-value) is used to

decide whether or not the H0 can be rejected

17
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Interpretation of p-values is problematic

Samples give results that differ from what is true in the population

The variability among samples decreases as the sample size

increases

The test of H0 is whether or not the p-value lies below the chosen
cut-off point

If the p-value exceeds the critical value, H0 is not rejected;

but we cannot say that H0 is true,
there is just not enough evidence to reject it

18
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Significance

The cut-off point is usually 0.05 (sometimes 0.01)

When p < cut-off the result is called statistically significant,

when p ≥ cut-off it is called not significant;
therefore "significance test"

Statistical significance versus clinical relevance

19
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Significance

Cut-off is arbitrary; no specific importance

Whether in a study p = 0.0450 or p = 0.0449, the result should not

be interpreted differently

Too often the custom: based on the p-value, the analysis is treated
as a process for making a decision

20
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Significance

Two possible errors:

1. Type I error: there is a significant result (thus H0 is rejected),
but H0 is actually true (false positive finding)
2. Type II error: there is a non-significant result (thus H0 is not
rejected), but H0 is actually not true (false negative finding)

Probability of Type I error = p(α): usually determined at max. 5%

Probability of Type II error = p(β)

dependent upon:
• effect size
• sample size

21
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
The problem

Publication bias: papers especially report significant findings

Results vary because of sampling variation;

studies that show larger effects are more likely to be significant,
thus more likely to be published

When H0 is true, one study out of 20 will give a significant result at

the 5% level

Statistical significance = "success", non-significance = "failure"

A small study may fail to detect (as significant) a real difference

22
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Hypothesis testing
Misinterpretations of p-value:
• the probability of the data having arisen by chance
• the probability that the observed effect is not a real one

missing: ... when H0 is true !

The observed effect in the sample is genuine,

but we do not know what is true in the population

One-sided or two-sided p-values?

N.B. H1 is the alternative hypothesis that there is an effect;
the effect could be positive or negative
One-sided tests are rarely appropriate, but the decision to use a
one-sided test must be made before the data are analysed
23
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

24
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Effect size
Measurement of the occurrence of disease:
• prevalence: n / N at a particular point in time
• incidence:
- cumulative incidence: n in time period / population at risk
- incidence rate: n per 1000 person-years

Data analysis: occurrence relation between determinant and outcome

Measurement of the occurrence relation for two treatment groups:

• outcome is continuous: difference
• outcome is discrete: absolute / relative risk

Estimation of the effect size and a measure of precision

25
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Precision
In a sample:


se 
n

where:
se = standard error of the mean
 = standard deviation in the population
n = sample size

Standard error: a measure of the uncertainty of a single sample

mean as an estimate of the population mean

 can usually be estimated by s, the standard deviation in the sample

26
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Precision
Continuous variable

Standard error of a sample mean:

s
se( x ) 
n

Standard error of the difference between two sample means:

s12 s22
se( x1  x2 )  
n1 n2

27
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Precision
Dichotomy

Standard error of a sample proportion (p):

p  (1  p)
se( p) 
n

Standard error of the difference between two proportions:

p1  (1  p1 ) p2  (1  p2 )
se( p1  p2 )  
n1 n2

28
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Precision
Two features:
• the mean observed in a sample:
best estimate of the true value in the population
• the distribution of the means obtained in several samples:
approximately Normal-distributed for large samples

 confidence interval:
a range of values which very likely includes the true value

95% confidence interval for the sample mean: a range of values

which contains the true population mean with a probability 0.95

Thus, there is a small chance of missing the true population mean !

29
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Precision
95% confidence interval typically:

between
x  1.96  se
and
x  1.96  se

Confidence interval also with any other estimate, e.g.:

• proportion
• risk estimate (OR, RR, HR)
• median
• sensitivity / specificity

30
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Estimation and testing

Hypothesis testing ?

31
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Estimation and testing

For clinical relevance the results must be quantified by estimates
of what is actually observed (effect size, consistency of the effect);
a single p-value conveys too little information

Confidence intervals and p-values are based on the same kind of

calculations

Confidence interval: contains information on the scale of the

measurement itself

The confidence interval shows the uncertainty (lack of precision) in

the estimate of interest: more useful information than the p-value !

Strange notion: "result is highly significant"

32
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Statistical modelling
Statistical model: a mathematical relationship, between 2 or more
variables, that gives an approximate description of the data;
a simplification compatible with the data

Important:
1. certain assumptions must be fulfilled to be able to use a model
2. how well does the model fit the data?:
– systematic discrepancies
– soundness of prediction for an individual

Common method is regression modelling:

• linear: continuous outcome
• logistic: dichotomous outcome
• Cox: time-to-event outcome
33
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Statistical modelling
Non-parametric methods

Estimation and hypothesis testing are different approaches, but both

based on the same statistical model and the same assumptions:
the theoretical distributions are described by the parameters
mean and standard deviation

Other methods without distributional assumptions: non-parametric;

also "distribution-free" or "rank" methods

Relevant for skewed data or score data

These methods usually more suited for hypothesis testing

(best known exception: the median)
34
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

35
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Analysis method
Scale:
• quantitative: discrete, continuous
• qualitative: dichotomous, nominal, ordinal

If quantitative scale, distribution:

• Normal
• Normal after transformation
• non-Normal

If Normal distribution, then parametric method

If non-Normal, then non-parametric method

36
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Parametric methods
One group of observations:
• mean and confidence interval
• one sample t-test

Two groups of paired observations:

• difference between means and confidence interval
• paired t-test

Two independent groups of observations:

• difference between means and confidence interval
• two sample t-test

More than two groups of observations: analysis of variance (ANOVA)

37
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Non-parametric methods
One group of observations: Wilcoxon signed rank sum test

Two groups of paired observations: Wilcoxon signed rank sum test

Two independent groups of observations: Mann-Whitney test

More than two groups of observations: Friedman; Kruskal-Wallis

38
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Qualitative variables
One variable:
• proportion and confidence interval
• z-test

Two variables, number of categories:

• 2 * 2: special case ---> later
• 2 * k nominal scale: χ2-test (with k-1 d.f.)
• 2 * k ordinal scale:
- χ2-test for trend (with 1 d.f.)
- Mann-Whitney test
• k1 nominal * k2 nominal scale: χ2-test (with [r-1]*[c-1] d.f.)
• k1 ordinal * k2 nominal scale: Kruskal-Wallis; Mann-Whitney
• k1 ordinal * k2 ordinal scale: rank correlation; paired Wilcoxon

39
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Qualitative variables - special case

2 * 2 categories; dependent samples:
• difference between proportions and confidence interval
• McNemar test

2 * 2 categories; independent samples:

a. cohort study
b. case-control study

40
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Qualitative variables - special case

2 * 2 categories; independent samples:

a. cohort study:
• difference between proportions and confidence interval
• z-test
• relative risk (RR) and confidence interval
• χ2-test (with 1 d.f.)
• Fisher's exact test

b. case-control study:
• odds ratio (OR) and (exact) confidence interval
• χ2-test (with 1 d.f.)
• Fisher's exact test

41
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Correlation - warning

r = 0.816

wikipedia
42
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

43
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Study design
Most studies aim to answer simple questions

Epidemiological research: basically about occurrence relations

The design must fit the research question to achieve valid an

precise results

The data from a good study can be analysed in many ways, but not
much can afterwards compensate for study design problems

Classifications:
• experimental - observational
• prospective - retrospective: the way the data are collected
• longitudinal - cross-sectional: numerous observations or once
44
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Variation
Care is needed to make samples "representative" of or fitting for
the population

In comparative studies the groups should be made similar for

known sources of variation

Bias: a disturbed representation of a true occurrence relation;

when not controlled, validity is compromised

45
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Classification
Experimental study: to investigate the effects of an intervention,
e.g.:
• clinical trials
• animal studies
• laboratory studies
Strong inferences, comparisons between groups
Experimental research is often prospective and longitudinal

Observational study: information collected, no influence on events,

e.g.:
• surveys
• epidemiological studies
Observational research can take on many forms

46
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Trials
Randomisation: at random allocation of treatments to patients

Blinding: prevention of observer effects; placebo-effect intact

Repeated measurements: to counteract high variability or poor

accuracy of measurement

The patient group selected should be as similar to the relevant

population as possible

Larger samples increase precision

47
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Trials
Random allocation:
• to prevent bias (confounding)
• statistical theory is based on the idea of random sampling

Types of randomisation:
• simple
• block
• stratified
• cluster

Structure: within subject and between subject comparisons

Factorial design: all combinations of factors are evaluated

48
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Observational studies
Randomisation often not an option
(unethical, unpractical, expensive)

Two main types in clinical research:

• case-control study
• cohort study

49
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Observational studies
Case-control study

A group of subjects with a certain disease is identified along

with a group of unaffected subjects;
in retrospect the exposure(s) are compared

If cases have greater exposure than the controls:

possibly the exposure is causally related to the disease

Advantages:
• simple, thus quick and cheap
• valuable when the condition is very rare
Main disadvantage:
• possible biases
50
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Observational studies
Case-control study

Main sources of bias:

• selection of the control group not easy: matching can help in
some circumstances
• selection of cases (affects the degree of generalisability)
• recall bias: differences between cases and controls (under-
reporting of exposure in control group)
• inaccuracy of retrospective data (underestimation)
• ascertainment bias (detection bias): e.g. oral contraceptive pill
and cervical cancer (OCP-users have more cervical smears, so
more likely to have cervical cancer detected)

51
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Observational studies
Cohort study

A group of subjects is identified and followed prospectively:

unaffected individuals observed until enough events

Advantages:
• more possibilities to focus on subgroups and determinants
• data recording can be carefully controlled
Disadvantages:
• time-consuming, thus expensive
• selection of subjects not always obvious (clinic-based versus
population-based studies; natural history of disease)
• loss to follow-up: bias (loss related to outcome/determinant)
• surveillance bias: more intensive investigation high risk group

52
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Observational studies
Cross-sectional study

One time subject contact; usually descriptive, often surveys

Advantages:
• no recall bias
• no loss to follow-up
• relatively cheap and easy
Problems:
• sample selection: representativeness
• response rates often quite low (volunteer bias)
• sequence in time: what is cause and what is effect?

53
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

54
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Clinical epidemiologic research

Research question comes first !

Four major types of research:

• intervention
• aetiological
• diagnostic
• prognostic

55
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Clinical epidemiologic research

Intervention research

Causal

Aim:
to causally explain course of disease as influenced by treatment

Relevance: research and drug development/registration

Randomisation and blinding will minimise any influence of

extraneous determinants

56
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Clinical epidemiologic research

Aetiological research

Causal

Aim:
to causally explain occurrence of a disease from determinant

Research relevance; may indicate means of prevention/intervention

57
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Clinical epidemiologic research

Diagnostic research

Descriptive

Aim:
to predict the probability of presence of a disease
from clinical and non-clinical profile

Relevance for patient and physician to establish diagnosis and

management guidance

58
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Clinical epidemiologic research

Prognostic research

Descriptive

Aim:
to predict the course of a disease
from clinical and non-clinical profile

Relevance for patient and physician to learn about the future

diagnosis and management guidance

Individual patient's prognosis

59
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

SPSS session
• Compare means
• Crossstabs (incl. OR)
• Non-parametric tests

60
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

References and links

Altman: "Practical Statistics for Medical Research", Chapman and
Hall, London
Altman, Machin, Bryant, Gardner: "Statistics with confidence;
confidence intervals and statistical guidelines", BMJ Books, Bristol
Grobbee, Hoes: "Clinical Epidemiology; Principles, Methods, and
Applications for Clinical Research", Jones and Bartlett, London

http://www.bmj.com/about-bmj/resources-readers/publications
• How to read a paper
• Epidemiology for the uninitiated
• Statistics at square one

61
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Literature
Median or mean?
Measures of variability
Quantiles, cumulative distribution function, and box-plot
The 2 by 2 table
What is a confidence interval?
What is a significance test? General issues
Common significance tests

62
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Consultation

63
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Test

Eur J Clin Inv 2008;38:372-380

64
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Questions ?

65
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Exercises
Formulate a research question with an exposure and an outcome

Explore the two variables separately (optionally also sex):

graph, frequency, parameters

Choose an appropriate analysis method and apply

Consider a new research idea (based on your results):

research question, study design

Interpretation, reflection

66
Basics Statistics 2, 2016/2017, ME
 Medical Statistics, Informatics and Health Economics
INNSBRUCK MEDICAL UNIVERSITY

Thank you for your attention

67
Basics Statistics 2, 2016/2017, ME