Chapter 1: Overview and Descriptive Statistics

Devore — Probability and Statistics for Engineering and the Sciences

Introduction
Statistics helps us make informed decisions in the presence of uncertainty and
variation. If nothing varied, a single observation would tell you everything.

Two branches work together: probability (reasons from population → sample,

“deductive”) and inferential statistics (reasons from sample → population, “inductive”).
You need probability first because you must understand sampling uncertainty before
you can draw conclusions from a sample.

1.1 Populations, Samples, and Processes

Key vocabulary

Population: the entire well-defined collection of objects/individuals under study.

Census: collecting data on the whole population (often impractical).

Sample: a subset of the population, selected in some prescribed way.

Variable: a characteristic that can differ from one object to another (categorical, e.g.,
pass/fail, or numerical, e.g., diameter).

Data types:

Univariate — one variable observed

Bivariate — two variables observed together

Multivariate — more than one variable observed together

Branches of statistics

Descriptive statistics: summarizing/organizing data — graphs (histograms, boxplots,

scatterplots) and numerical summaries (mean, standard deviation, correlation).

Inferential statistics: using sample data to draw conclusions (make inferences) about
the population — point estimates, confidence intervals, hypothesis tests.
Populations can be:

Concrete (an actual existing group, e.g., all drivers in a city)

Conceptual/hypothetical (e.g., “all possible measurements under similar conditions”

when there’s no literal finite population to sample from)

Enumerative vs. analytic studies (distinction due to W. E. Deming):

Enumerative study: a fixed, identifiable population exists (a sampling frame is available),

e.g., sampling serial numbers of manufactured units.

Analytic study: no sampling frame exists; you’re studying a process to improve future
output (e.g., testing prototype turbines). Conclusions about future production require
caution and subject-matter judgment.

Collecting data

Target population (who you want to conclude about) may differ from the population
actually sampled — a common source of error.

Simple random sample: every subset of the given size is equally likely to be chosen.

Stratified sampling: split the population into non-overlapping groups (strata) and
sample from each — ensures every subgroup is represented.

Convenience sample: units selected without systematic randomization — risks being

unrepresentative.

1.2 Pictorial and Tabular Methods in Descriptive Statistics

Stem-and-leaf display: splits each value into a “stem” (leading digit(s)) and “leaf” (trailing
digit) — quick way to see shape, spread, and outliers while keeping the actual data values
visible.

Dotplot: each observation is shown as a dot above its value on a number line — useful for
smaller data sets.

Discrete vs. continuous variables (important because it affects how you build a
histogram):

Discrete: possible values are finite or countable (e.g., number of defects).

Continuous: possible values form an interval (e.g., a measurement like weight or

length).
Histograms

For discrete data: plot frequency (or relative frequency) vs. each possible value.

For continuous data: divide the range into class intervals, then plot frequency/relative
frequency (or density = relative frequency ÷ class width) per interval. Using density
means the total area of the histogram equals 1 (or equals the relative frequency total),
which allows fair comparison across differently-sized data sets.

Shape vocabulary:

Unimodal: one peak. Bimodal: two peaks. Multimodal: more than two.

Symmetric: left and right halves mirror each other.

Positively skewed: long right tail (upper tail stretched out).

Negatively skewed: long left tail.

Qualitative (categorical) data: summarized with frequency/relative frequency tables and

bar charts (histogram analog for categories) or Pareto diagrams (bars ordered by
descending frequency, common in quality control).

1.3 Measures of Location (Center)

Sample mean $$\bar{x} = \frac{\sum x_i}{n}$$

Represents the “balance point” of the data.

Sensitive to outliers — a single extreme value can pull it substantially.

The population mean is denoted µ (Greek mu); $\bar{x}$ is used to estimate µ.

Sample median ($\tilde{x}$)

Order the data. If n is odd, it’s the middle value; if n is even, it’s the average of the two
middle values.

Resistant/insensitive to outliers — unaffected by how extreme the tail values are, only
by their rank.

Population median denoted $\tilde{\mu}$.

Quartiles and percentiles

Quartiles divide data into 4 equal parts (2nd quartile = median).

 The kth percentile separates the lowest k% from the rest.

Trimmed mean

A compromise between mean (0% trimmed) and median (max trimmed): eliminate the
smallest and largest p% of the data, then average what remains.

Reduces outlier sensitivity while still using more of the data than the median does.

Categorical data — sample proportion

For a two-category (dichotomous) population, if x observations fall in category 1, the

sample proportion is $x/n$.

1.4 Measures of Variability (Spread)

Deviations from the mean: $(x_i - \bar{x})$ — these always sum to zero, so raw deviations
don’t measure spread directly; hence we square them.

Sample variance $$s^2 = \frac{\sum (x_i - \bar{x})^2}{n-1} = \frac{S_{xx}}{n-1}$$

Divides by n − 1 (not n) — this makes it an unbiased estimator of the population variance

σ².

Sample standard deviation $$s = \sqrt{s^2}$$

Same units as the original data (unlike variance, which is in squared units) — makes it
interpretable as a “typical” deviation from the mean.

Fourths and fourth spread (resistant alternative to variance)

Split ordered data into lower/upper halves (median included in both if n is odd); the
lower fourth = median of the lower half, upper fourth = median of the upper half.

Fourth spread: $fs = \text{upper fourth} - \text{lower fourth}$ — resistant to outliers

(similar in spirit to the interquartile range).

Boxplots

Built from the five-number summary: minimum, lower fourth, median, upper fourth,
maximum.

Box spans lower fourth to upper fourth (width = fourth spread); a line marks the median;
whiskers extend to the extreme non-outlier values.

Outlier rule: a point farther than 1.5 × fs from the nearest fourth is an outlier; farther
 than 3 × fs is an extreme outlier (otherwise “mild”).

Comparative boxplots (side by side) are useful for comparing several samples/groups at
once.

Quick-Reference Summary Table

Concept Formula / Rule Sensitive to outliers?

Mean $\bar{x} = \sum x_i / n$ Yes

Median Middle value(s) of ordered data No

Trimmed mean Mean after removing top/bottom p% Partially

Variance $s^2 = \sum(x_i-\bar{x})^2/(n-1)$ Yes

Std. deviation $s = \sqrt{s^2}$ Yes

Fourth spread Upper fourth − Lower fourth No

Outlier (boxplot) > 1.5·fs from nearest fourth —

Big-picture takeaways
1. Chapter 1 sets up the language (population, sample, variable, data types) used
throughout the book.

2. It draws the key contrast between descriptive statistics (chapter’s focus) and
inferential statistics (the rest of the book).

3. Every graphical/numerical tool introduced has a trade-off: mean vs. median, variance
vs. fourth spread — richer use of data vs. robustness to outliers.