Chapter 3: Defining population and samples

LO 3.1 Defining pop and samples

Population:
- Group with something in common (parameter)
- Expensive/impossible to get all
- Parameter: characteristic of a population

VS Sample:
- Subset of a population
- Representative and focus on statistics = characteristics of a sample
- Used to make inferences (conclusion) about the characteristic of a pop
- To reduce cost = sample is drawn out of population
 Parameter and inference may match

Descriptive/Summary statistics: measures that describes visible component of population or

sample
Inferential statistics: measure calculated only using sample
Hypothesis: proposed explanation made on basis of sample/limited evidence
- Starting point for more investigation
- Uses inferential statistics
- T-test, Z-test, P-test, F-test

LO 3.2: Sampling methods data reduction and bias

1. Simple random sampling:

- all elements have equal chance of being selected into the sample to hope select
representative of whole population
- Doesn’t care about selecting subset, or when homogeneous data/similar
2. 3ified random sampling:
- Pop is categorized/ dividing population into subgroups or strata based on specific
characteristics or attributes  calculate proportion of population in each group and
random sample each group to ensure appropriate number of each group (stratum) is
represented
- Ex: ensure representation of both happy/unhappy customers
3. Cluster sampling:
- Diving into group, calculate proportion, select only few cluster relevant + proportionate
- Select few groups: geography, time zone
- More efficient and cost-effective
4. Convenience/non-probability sampling
- Ease of access and availability when limited and budgets at low
 - Some bias
 data already collected = simply subset data
 if need to be collected = distribute survey then stop collecting data once # reached

Data reduction: process of reducing size of data set to more manageable and suitable size for a
business analysis projects
- To retain meaningful information
- Focus on most interesting, critical, and abnormal items
- Speeds up analysis + reduce cost
How? Filtering:
 data  sort and filter  filter: removing rows that aren’t interesting/relevant/ choosing
interesting one
Decision process of using subset of data: need to consider purpose of analysis, time and cost.

Bias in business analytics:

- Prejudice in favor of or against
- Intentional vs unintentional
- During data collection, analysis, results
Types:
1. Nonresponse: sample not answering survey = potential distortion in results
 to prevent: inform sample about importance = sending gift card
- Partiality that results when respondents differ from non-respondents
2. Selection: analyst selected inappropriate sample
 when purposefully selects potions of participants/data that are likely to provide answers
aligning with analyst’s belief = not representative so not generalized
3. Confirmation: favor info that confirm pre-existing belief – do not reflect on truth =
inaccurate conclusions. Have already been collected
4. Outlier: extreme values influencing the interpretation of results. Have already been
collected

LO 3 Understanding statistics:

Probability distributions:
1. Random Variable: quantifies the outcomes of random occurrences
- To measure things that happen by chance
2. Data distribution: shows all possible values for a variable and how frequent
- Tells you what numbers could come up and how frequent each number is likely to appear
- Deals with hypothetical/predicted data
3. Probability distribution:
- Statistical function that describes possible values in population and likelihood that any
given observation (random variable) can take a given range/value
 - How probable an outcome is
- Deals with observed data you already have
- If discrete distribution = presented in ranges

Types of numerical data: determine types of probability distribution

1. Discrete data: whole number, finite set of values between 2 observations
Ex: inventory, vehicles
2. Continuous data: any numerical value, infinite set of value
Ex: Weight, currency

Measures of central tendency: describe the center point of data set + susceptible to outliers!
Mean: average = Sum/n – cannot be categorical + susceptible to outliers
Median: midpoint of data distribution – cannot be categorical (small – highest) – ave if pair – no
susceptibility to outliers
Mode: simplest+ most common value – most important for categorical - uni/bi/multi-modal
Symmetry = when all above are equal
Kurtosis: distribution shape + thickness of tails
- Closer to 0 = normal distribution
- Whether more clustered in peak or tails
Work with Skewness: help determine likelihood
of event falling in a tail
- Positive kurtosis : skewed right. More
extreme value peaked at center
- Negative Kurtosis: skewed left. Fewer extreme values so flatter around center
 Mean higher than median = implied some outliers are skewing it right

Measures of dispersion:
Range: Max – Min
- Affected by outliers
Interquartile range (IQR): 4 quartiles to determine shape
- Should be in sorted order
- Q1: lowest 25% of observations
- Q2: next 25% - from 25% to 50%(= median)
- Q3: median to 75%
- Q4: 75 and more
- “Inter”= implies not interest in 4 quartile but specific middle section Q2 and Q3
 Suppose you have a data set : 24,25,25,25,26,27,28,28,29
- Q2 value = median = 26
- Q1 value = median of lower half (red color) = 25+25/2 =25
- Q3 value = median of upper half in blue color = 28+28/2 =28
- Interquartile range : Q3 – Q1 = 28-25 = 3
 -
Variance: average squared deviation from the mean
- Measure how individual data points in a dataset differ from the average
- [X1 – Mean1]^2 + [….] / # of x
Standard deviation:
- Square root of Variance
- Same unit as data value
Coefficient of variation: o/u
- Measure of relative variability by comparing standard deviation to mean
- Useful when comparing different datasets or units or scales
- [SD/MEAN]*100

Continuous probability distributions:

Normal Gaussian Distribution:
- Bell shaped with most data point clustered near average/mean
- Natural occurring, symmetric = skewness is 0, but kurtosis is 3 b/c 68% data fall within
one SD, 95% within 2 SD, 99.7% falls within 3 SD
- Ex: weight, pop height, pop IQ, shoe size

Standard Normal distribution:

- Theoretical distribution: Doesn’t represent a real distribution – used only to make
comparison between distribution easier + calculated probabilities of individual
observation
- M-M-M = 0
- SD=1

Z score: standardized values

- Measures number of SD a data point is away from the mean
- Z = (x-mean)/ SD

Uniform distribution:
- Probability distribution that describes a set of continuous random variable where every
value within a given range is equally likely to occur:
- Flat and constant distribution – no peaks no tails
LO 3.4: Using Software tools to Create Summary Statistics:

Argument: value that the function uses to perform calculations (IQR exceptions)
Descriptive data: Data  Data analysis on right  DS  Ok = get data
Tableau Summary statistics – descriptive Statistics: Worksheet  Show summary

- Bin - Frequency

- [0-50) - 65
LO 3.5 Interpreting and Visualizing statistics:
Frequency distribution: - [50-100) - 63
- For numerical data b/c categorical we can just count
- Bins, classes, and intervals (categories in numerical - [100-150) - 30
data) - [150-200) - 8
- Table uses bins = to list frequent of various outcome
in sample
- Symmetrical = expect 5th/middle to contain most observation
- Skewed right = more observation in 1st
- Skewed left = more in last one

Histogram:
- Visual representation for frequency distribution
 - Size of bins can shape
- Shows stats like: mean, sun, count etc…
- No gaps in between unless need to indicate absence of data point for that interval
- Reference lines for mean and median
- Like vertical bar chart but bars are replaced by bins
- Y axis in bar graph for descriptive but in H for # of observations

Box Plot:
- shows dispersion in
terms of quartiles
Box: represents the IQR: 25th
Q1, 75th Q3
- Length: reflects spread
of middle 50% of the
data

Line/whisker: represent the

range of data from quartiles
(Q3+1.5IQR and Q1-
1.5IQR)
Median: horizontal bar
which divides the data
in two equal halves
Outliers: individual
points beyond