Transcript Do I need statistical methods?

Do I need statistical
methods?
Samu Mäntyniemi
Learning from experience
 Which way a bottle cap is going to land?
 Think, and then write down your opinion
about the probability that in the next toss the
cap will end up upside down.
 After seeing the result of the first toss, reevaluate your opinion
 Continue this process for 20 times..
 Did your opinion change during the
accumulation of evidence?
Scientific?
 Explain your existing knowledge
 What do you know about bottle caps and how this
knowledge is brought into this process of learning
more?
 Explain your experiment
 How did you conduct the experiment?
 What kind of bottle cap was used?
 Reveal your logic
 How did you combine your existing knowledge and
observations to come up with your conclusion?
Problems in explanation
 In an earlier study, the cap was found upside down in
6 out of 10 cases, and the cap and the experiment
are identical, therefore the initial guess for the first
toss is 0.6
 In our first toss the cap was upside down, then for the
next toss the probability was changed to 0.63…
 After 19 tosses observing the cap upside down did
not affect the probability that much anymore because
larger number of tosses had already taken place
 It is difficult to replicate the logic and conclusions of
the study based on the verbal explanation!
How will statistical methods help?
 Provides mathematical presentation of
knowledge


common language for more precise
expression of uncertainty
Results of previous analysis can be directly
accounted for in a new study
 Provides precise and transparent logic for
updating and combining knowledge

Conclusions can be replicated and sources of
information affecting them can be isolated
Subjectivity in science
 Subjective : depends on the person



Each scientist has slightly different
background: the knowledge about any given
question is slightly different
Interpretation of new evidence is slightly
different
In this context subjectivity does not include
personal opinions affected by fear, hope,
political orientation etc.
Objectivity
 Objective : independent of the person
 State of the world for which there is no room for
interpretation
 For example, the position of the bottle cap: it is upside
down or not, regardless of the person assessing the
position.
 Data: objective set of facts
 Can scientific reasoning be objective?
 Yes, when reporting the data
 No, when drawing conclusions about unobserved
quantities: subjective interpretation can not be avoided
Role of statistical inference
 There is an objective true state of the world
 We do not know exactly what it is
 If it cannot be simply observed by a measurement,
then collect indirect evidence about the unknown
state (data)
 Use statistical methods to formulate existing
knowledge about the unknown state and then
combine with interpretation of new evidence
 “Given my past knowledge and my interpretation of
the observed data set, what can I say about the true
state of the world?”
Why to use statistical inference?
 More precise description of your logic and inference -
> your message is more clear
 Can be very valuable in case of large and complex
problems


Setup your logic in small local pieces, then pull
everything together in a consistent way
-> Modern computing power can solve problems you
could not
 Editor or referee is asking you to
 Is this a good reason?
 Are you describing or making inference?