Statistics for Decision Makers - 01.01 - Introduction

title

01.01 - Introduction

author

Bernard Szlachta (NobleProg Ltd) bs@nobleprog.co.uk

Inference and Decision。

• Inference is about finding facts
• Decisions are about taking courses of action, which can be based on inference findings

Is inference not followed by a decision idle?

Decisions and Rules。

• A Decision is a choice between alternatives
• Rules are the conditions which guide or determine decisions

BPMN, Decisions and Statistics。

Statistics, Decisions and Rules。

• The Decision Maker encounters a decision point in a business process
• Making those decisions is mostly a question of judgement
• Finding evidence using the scientific method
• The ultimate goal is to fully automate decision making by creating rules which use statistical methods

Hiring Example。

Business Process

Recruitment

Decision

Invite a candidate for interview or not?

Variables

Candidate has relevant experience

Salary expectation

Availability

Inference

What is the probability that the candidate will be hired and make a good employee?

Rules

 Rule "Should the candidate be interviewed?"
 When 
     P-value < 0.01 
 Then 
     Interview Candidate
 End

Traditional Rules Example。

 Rule "Should the candidate be interviewed?"
 When 
     ExpectedSalary < 100000 and
     AvailableDate < 2015-01-01
     ExperienceScore >= 1000
 Then 
     Interview Candidate
 End

When to use Statistical Inference to make decisions。

• Deterministic solutions are usually cheaper and easier to understand and maintain

Use Statistical Models if:

• the loss related to errors caused by deterministic solutions justifies the cost of maintaining a statistical module
• there are no easy deterministic solutions
• the model needs to learn (e.g. neural networks)
• the rules are in the heads of managers who change often and need to relearn the rules each time from experience

Technical Statistics。

• Database Indexes (e.g. Oracle)
• Search Engine
• Network Routers
• .....

Overview of the methods。

Describe	Sample	Measure	Predict
mean standard deviation distribution	simple random sampling stratified random sample determining sample size (from power)	validity reliability	hypothesis testing regression/forecasting

Questions to the audience。…

• Why things differ each time we measure them?
• Brownian Motion
• Heisenberg Uncertainty Principle

Quiz。

Please find the Quiz here

Quiz

1 A sentence "If the P-value is greater than 0.5 we consider the test inconclusive" is an example of

	A decision
	A rule

2 A company is surveying people about their propensity to recommend a product to a friend. It turned out that only a couple of people replied to the survey, but the difference in propensity increased 70%. What can the managers do?

	They must find more respondents and repeat the survey, otherwise the test is insignificant
	They can simply assume that the propensity increased 70%
	They should calculate what the chances are you would get such an increase by pure chance if there were no difference in reality (i.e. use a statistical method)