Statistics for Data Science

Inferential Statistics
Agenda – Inferential Statistics
1. Inferential Statistics
2. Some fundamental terms first
a. Random Variables
b. Distribution and its types
3. Binomial Distribution
4. Uniform Distribution
5. Normal Distribution
6. Sampling and Inference
a. Simple Random Samples
b. Sampling Distribution
c. Central Limit Theorem
7. Estimation
a. Point Estimation
b. Interval Estimation

2
Descriptive vs. Inferential Statistics

Summaries from Inferential

Is that enough?
data Statistics

Summaries give a sense of Typically, we work with Inferential statistics helps

central tendency only ‘samples’ of data tackle that challenge

Tell a lot about ‘what’s It is not rough to make There are powerful
happening’ generalized statement methods to draw
about the population reasonable conclusions
Mean, Std Dev, IQRs etc. about the population from
Challenge - how to learn an observed sample
about population from the
sample? This becomes extremely
critical in business
decision making

4
Role of distributions in inferential statistics

5
Real World Problems

Is the new manufacturing process better/more reliable than the old

Quality Testing process?

What is the likelihood that temp will be more than 20 degree celsius
Meteorology on a specific day?

Human Does training the workforce improve sales?

Resources

Digital What is the likelihood that the conversion rate of the website will be
Marketing above x% next month?

... ...
6
We will learn about...
1. Some famous distributions & their characteristics
2. Central Limit Theorem - Powerful idea that enables a lot of inferential statistics
3. Estimations and Confidence Intervals
4. Hypothesis Testing

7
Some fundamental terms first

8
Random Variable
Suppose there are 1000 students in the university. What is the probability that 500
students will pass the upcoming exam?

There is a 50-50 chance that each student will pass or fail

The total number of students who pass can range from 0 to 1000

A random variable assigns a numerical value to each outcome of an experiment. It

assumes different values with different probability.

9
Discrete Random Variable

You work for an Auto insurance company. Suppose the number of insurance claims filed by
a driver in a month is a random variable (X) described by

0, with prob 0.95

1, with prob 0.04
X=
2, with prob 0.008
3, with prob 0.002

All probabilities must be non-negative and sum to 1.

When all possible values the random variable can take can be listed,
we call it a discrete random variable

10
Continuous Random Variable
Suppose the volume of soda in a bottle is described by a random variable.

Can we list all possible values?

498 mL, 499 mL, 500 mL, … … … What about 499.2129415 mL?

Sometimes it is just not possible to list all values a random variable can take

If the random variable can take any value in a given range, we call it a
continuous random variable

11
Probability Distribution

Probability Distribution
Describes the values that a random variable can take,
along with the probabilities of those values

Discrete Probability Distribution Continuous Probability Distribution

Arises from discrete random variables Arises from continuous random variables

Has an associated probability mass Has an associated probability density

function, which gives the probability with function, which helps determine the
which the random variable takes a probability with which the random
particular value variable lies between two given numbers

12
Probability Distribution: Example
A company tracks the number of sales new employees make each day during a 100-day
probationary period. The results for one new employee are shown. Construct and plot a
probability distribution.

Sales #Days Sales #Days Relative

Frequency
0 16
0 16 0.16
1 18
1 18 0.18
2 15
2 15 0.15
3 21
3 21 0.21
4 11
4 11 0.11
5 10
5 10 0.10
6 9
6 9 0.09
13
Distributions around us (commonly occuring)

Company has introduced a new drug to cure a disease, it either cures

Bernoulli
the disease (it's successful) or it doesn't (it's a failure)

Binomial The number of defective products in a production run.

Uniform The number of of microwave oven sold daily at a electronic store

Income distribution of a country : middle-class population is a bit

Normal
higher than the rich and poor population in the country

14
Basic distributions - Binomial

15
Bernoulli Distribution
Success and failure are non-
judgemental. Any one outcome
may be termed as success

It has only two possible outcomes, namely 1 (success) and 0 (failure), of one single trial.

1, with prob p
X=
0, with prob 1-p

Very useful in many scenarios:

Manufacturing defective parts Outcome of medical test

16
Binomial Distribution

Suppose we ask any adult who uses the app TikTok if he/she has ever posted a video on the app

The answer can be Yes or No (success or failure)

We can use the Bernoulli distribution to model this scenario

Now let us extend this into a survey of 25 adults chosen at random

We can define a random variable X which counts the number of successes

(say, the number of adults who responded Yes)

17
Binomial Distribution

In many situations an experiment may have only two outcomes - success and failure

These experiments can be modelled using the Binomial probability distribution.

Bernoulli Distribution is a special case of Binomial Distribution with a single trial.

Probability Mass Function

18
Binomial Distribution : Assumptions

Number of trials (n) is fixed.

Each trial is independent of the other trials.

There are only two possible outcomes (success or failure) for each trial.

The probability of a success (p) is the same for each trial.

19
What happens if these assumptions are violated?
In a month of 30 days, what is the probability that it will rain on more than 10 days, if on
average the chance of rain on a given day is 20%?

If we assume that:

1. The event of rain on a particular day is independent of it raining on the previous day.
2. The chance of rain does not increase or decrease over the duration of the month.

Then we can use the binomial distribution with n=30 and p=0.2 to calculate the probability.

Assumptions 1 and 2 in the example are not strictly valid, but they allow for a direct calculation
that may be good enough for practical purposes.

20
Basic distributions - Uniform

21
Uniform Distribution
Suppose we roll a die. The outcomes of this event can be 1,2,3,4,5,6

All of the outcomes have an equal probability of occurrence and are

mutually exclusive https://pixabay.com/vectors/di
ce-games-play-1294902/

We can say that the probabilities of occurrence is uniformly distributed.

This is referred to as Uniform Distribution

Useful when we are interested in unbiased selection

22
Uniform Distribution
There are two types of Uniform Distribution

Discrete Uniform Distribution: Can take a finite number (m) of values and each value has
equal probability of selection.
For example: Number of books sold by a bookseller per day can be uniformly distributed
between 100 to 300.

Continuous Uniform Distribution: Can take any value between a specified range.
For example: Tomorrow’s temperature in United states can be
uniformly Distributed between 12 degree Celsius to 17 degree celsius

23
Basic distributions - Normal

24
Normal Distribution : Introduction
Normal distribution is the most common and useful continuous distribution

It is characterized by a symmetric bell-shaped curve having two parameters - mean (μ)and

standard deviation (𝜎).

25
Normal Distribution : Why Normal

Why is it called the normal distribution?

They are commonly found everywhere starting from nature to industry

Many useful datasets are approximately normally distributed

For example the height and weight of the adults, IQ scores, measurement errors,
quality control test results etc.

26
Normal Distribution : Properties

1. The graph of the normal 1. About 68% of the data fall

distribution is called the within one standard
normal curve deviation from the mean
2. Normal curve is symmetric 2. About 95% of the data fall
around the mean within two standard
3. Mean, Median and Mode of the deviation from the mean
normal distribution are equal 3. About 99.7% of the data
4. Total area under the normal fall within three standard
curve is 1 deviation from the mean
68%
95%
99%

𝜇-3𝜎 𝜇-2𝜎 𝜇-𝜎 𝜇+𝜎 𝜇+2𝜎 𝜇+3𝜎

27
Normal Distribution : Example
Assume that a food delivery service provider A has a mean delivery time of 40 minutes and
a standard deviation of 10 minutes.

Using the Empirical Rule, we can determine that

● About 68% of the delivery times are

between 30-50 minutes (40 ± 10)

68%
● About 95% of the delivery times are
between 20-60 minutes (40 ± 2(10)) 95%
99%
● About 99.7% of the delivery times are
between 10-70 minutes (40 ± 3(10)) 10 20 30 40 50 60 70

This property is known as Empirical rule.

28
Normal Distribution : Area under Density Curve

As with any continuous probability distribution, the area under the density curve between
two points indicates the probability that the variable will fall within that interval.

𝜇 and 𝜎 are the parameters that decide the center and spread of the normal curve

To find the area, we need Calculus

But, there is an easier way to do it in Python (or other softwares).

It provides us the necessary functions to calculate the area.

29
Normal Distribution : Standard Normal Distribution
A standard normal distribution is used to compare two normal distributions with different
parameters (μ, σ)

The standard normal variable is denoted by Z and the distribution is also known as Z
distribution

It always has a mean of 0 and standard deviation of 1

Standardize

Normal Distribution Standard Normal Distribution

30
Normal Distribution : Z-Score

A normal variable can be converted to standard normal variable by subtracting the

mean (𝜇) and dividing the standard deviation (𝜎):

where,
● X is the observed data point
● Z (Z-Score/Standard score) is the measure of the number of
standard deviations above or below the mean that data point
falls

31
Sampling and Inference

32
Revisiting the need for sampling..

In many of the situations, what we have available to us is a sample of data.

The data we have is finite.

Till now, the goal was to find ways of describing, summarizing and visualising
the sample data only

Moving ahead, we want to make inferences about the

“entire” population using the sample data.

33
Sampling : Simple Random Sampling

A sampling technique where every item in the population has an equal chance
of being selected

Allows all the entities in the population to have

Why are simple random an equal chance of being selected and so the
samples important? sample is likely to be representative of the
population

34
Sampling Distribution
The sampling distribution of a statistic is the probability distribution of that
statistic when we draw many samples
For example sampling distribution of the mean, sampling distribution of variance etc.
To a great extent, statistical inference techniques are based on sampling distribution of a statistic

Samples of
size n

Population Sampling Distribution

Distribution of means

35
Some Important Points

Suppose we are sampling from a population with mean 𝝁 and standard deviation 𝞂. Let 𝑿
ഥ
be the random variable representing the sample mean of n independent observations.

The mean of 𝑿
ഥ is equal to 𝝁

The standard deviation of 𝑿

ഥ is equal to 𝞂/√n (Also called the ‘standard error’ of 𝑿
ഥ)

Even the population is not normally distributed, then for sufficiently

large n 𝑿
ഥ is also normally distributed.

36
Central Limit Theorem

The sampling distribution of the sample means will approach

normal distribution as the sample size gets bigger, no matter
what the shape of the population distribution is.

Assumptions

Data must be randomly sampled Sample values must be independent of each other

Samples should come from the same distribution Sample size must be sufficiently large (≥30)

37
Central Limit Theorem

Large sample size provides better estimate of

the population mean.

For sample size n = 5, the mean of sample

means pile up around the population mean.

For sample size n = 30, the mean of sample

means are much closer to the population
mean.

38
Estimations

39
Estimation

Estimation
Make inference about a population parameter
based on sample statistic

Point Estimation Interval Estimation

Single point estimation of the population A range of values within which the
parameter population parameter lies with some
(x%) confidence
E.g. Population mean as estimated from
the sample mean is $40 E.g. Population mean should lie between
$38-42, with 95% confidence (x = 95)
40
Point Estimation
A point estimate of a population parameter is a single value of a statistic

For example: The sample mean X̅ is a point estimate of the population mean μ. Similarly, the sample
standard deviation s is a point estimate of the population standard deviation σ.

Point estimates vary from sample to sample. Often an interval is used to provide
a range of values the parameter can take, instead of a single point estimate

41
Interval estimation - Confidence interval

Confidence interval provides an interval, or a range of values, which is expected

to cover the true unknown parameter.
Confidence limits
True Value
Estimation
(unknown)
The upper and lower limits of
the interval are determined
using the distribution of the
sample mean and a multiplier
which specifies the ‘confidence’
Confidence level

42
Confidence Interval for Mean 𝜇

Interpretation of 95% Confidence Interval

- The interpretation of a 95% confidence interval is that, if the process is repeated a

large number of times, then the intervals so constructed, will contain the true
population parameter 95% of times.

Why not 100% Confidence Interval?

- A 100% confidence interval will include All possible values.

- Hence there will be no insight into the problem.

43