# MA250 Probability and Statistics Nazar Khan PUCIT Lecture

- Slides: 24

MA-250 Probability and Statistics Nazar Khan PUCIT Lecture 1

What is Statistics and Probability • Statistics is the ‘art’ of – understanding the “real” world as it is and not “how we think” it is, – intelligently summarizing large amounts of data, – making numerical guesses for puzzling questions. • Probability is the ‘tool’ – to work with statistics, – to make conclusions/predictions from statistics.

Statistics • Even a lifeless calculator can give you statistics by plugging numbers into formulae. • But the true meaning of those statistics requires careful thinking. • One aim of this course is to make you think like a statistician, not like a calculator!

Probability • One of the more important branches of mathematics. • Can be a bit unintuitive. • Has its own terminology. • Every probability problem requires thinking. – Fortunately, there are some tricks. • One aim of this course is to make you develop thinking skills that help solve probability problems!

Applications of Probability and Statistics • • • Computer Networks Machine Learning Computer Vision, Image Processing, Graphics Algorithms Data Mining

Applications of Probability and Statistics • • • Politics Economics Social Sciences Medicine Everything involves probability and statistics!

Applications of Probability and Statistics • Every two days we create as much data as we did from the beginning of mankind till 2003. • The only way to deal with such large amounts of data is to summarize it. • Statistics is the method of summarization.

The Scientific Method 1. Define the question 2. Background research, observation – Have others tried to answer this earlier? 3. Formulate a hypothesis – If we do X, then Y will happen. 4. Design and run an experiment 5. Analyze the results 6. Communicate the results • Experimental measurements are noisy (randomness). • Statistics is critical in the steps 4 and 5!

Design is more important than the experiment itself DESIGN OF EXPERIMENTS

Is Polio vaccine effective? • Some one makes a vaccine for Polio. • You need to find if it is effective or not. • How will you go about finding an answer to this question? – Follow the scientific method

Is Polio vaccine effective? • Follow the scientific method 1. 2. Define the question Background research, observation – 3. Formulate a hypothesis – 4. Do X Analyze the results – – 6. If we do X, then Y will happen. Design and run an experiment – 5. Have others tried to answer this earlier? Did Y happen? So what do we conclude? Communicate the results

Is Polio vaccine effective? • Which hypothesis is better? 1. Children that get vaccinated will have lesser polio cases 2. Children that get vaccinated will have lesser polio cases compared to children that don’t get vaccinated. • What will each hypothesis prove?

Is Polio vaccine effective? • Compare – those that use the vaccine – treatment group – those that don’t – control group • If the treatment group has lesser percentage of polio, then the treatment is effective • Otherwise the treatment is useless. Outcome Treatment Control Conclusion Less Polio More Polio Effective

How to lie with statistics • What if the treatment and control groups are different? – Treatment group is from people that are more immune to Polio – poor children – Control group is from people that are less immune to Polio – rich children – Statistics will falsely show that the vaccine is effective. Outcome Treatment Control Conclusion Less Polio More Polio Effective

How not to lie with statistics • The treatment and control groups must be similar. • How to ensure that? – Pick randomly • Statistical studies can mix up hidden factors. – Polio is a disease of hygiene. This factor must be accounted for in the treatment and control groups.

Questions that statistics can answer • Is Homeopathy effective? – Bad study: ask those that use Homeopathy. – Good study: perform a controlled experiment with randomly selected treatment and control groups. • How important is Misbah-ul-Haq to the team? – Bad study: look at his averages. – Good study: look at the circumstances of his innings.

Statistics is also an ‘art’ • Blind statistics: Plugging numbers into formulae without thinking. • Proper statistics: Controlling for the hidden factors. • Blind application of statistics can be disastrous. – Terming a medical treatment effective when, in fact, it is harmful. – Terming a medical treatment ineffective when, in fact, it is effective.

Confounding Factor • The treatment and control groups should differ from each other only in terms of the treatment. • If they differ with respect to some other factor, then this is a confounding factor. – Are the results due to treatment or due to the confounding factor?

The Randomized Controlled Experiment • Random selection of treatment and control groups • Elimination of confounding factors – Placebo effect – some people are cured by the idea of treatment. So give both groups the impression that they are being treated. – Double blinding – neither the patient nor the doctors know which group the patient is in. So the doctors don’t give different treatment.

The Observational Study • Is smoking harmful? – Can we do a randomized controlled experiment to answer this? • We can observe smokers and non-smokers over time.

Is Smoking Harmful? • If smokers are less healthy compared to nonsmokers, then “yes, smoking is harmful”. • Wrong!! • Association is not causation! • What about confounding factors? – For example, a gene that causes lung cancer and also causes people to smoke. • Careful studies have concluded that there are no confounding factors. Smoking really is harmful.

Association is not Causation • Many people die in hospitals. • Do hospitals cause death? • Hospitals and deaths are associated with each other. • But hospitals don’t cause deaths (in general). • Blind statistics will tell you that hospitals lead to deaths.

Randomized Controlled Experiment vs. Observational Study • Observational studies prove association but not causation. • Confounding factors can be at work. • Randomized controlled experiments try to minimize the effects of confounding factors. • So, wherever possible, a randomized controlled experiment should be performed to understand the real world.

Next Lecture • Descriptive statistics – Histograms – Mean, Standard Deviation – The Normal Curve • Read Ch. 3 -5.