Statistics: Introduction and Data Collection

What is Statistics?

Statistics is both a way of reasoning and a collection of tools and methods designed to help us understand the world through data. It involves analyzing variation, providing context, and making informed decisions based on data.

• Statistics (plural): Particular calculations made from data.

• Data: Values along with their context (who, what, when, where, why).

• Variation: Statistics is fundamentally about understanding why data varies.

Context in Data

• Respondents: Individuals who answer a survey.

• Subjects/Participants: People whom we experiment on.

• Observational/Experimental Units: Animals, plants, websites, or other objects studied.

• Records: Rows in a database.

• Variables: Characteristics recorded about each individual.

Sampling and Populations

• Sample: A subset of a larger population, ideally representative.

• Population: The entire group of individuals of interest.

Types of Variables

• Categorical variables: Not numerical (e.g., gender, color).

• Quantitative variables: Measured or counted numerical values.

  • Continuous variables: Can take any value within an interval (e.g., height).

  • Discrete variables: Fixed number of values (e.g., number of siblings).

• Identifier variable: Unique number to identify individuals.

• Ordinal variable: Values with a natural order (e.g., class rank).

Investigative Questions and Statistical Studies

• Good questions target populations difficult to study in entirety, requiring samples.

• Statistical studies use sample data to answer questions about populations.

• Data collection should be planned and new data should be collected for each study.

Common Pitfalls

• Variables can take on different roles depending on context.

• Always analyze context and be skeptical of data.

Describing Data with Tables and Graphs

Graphical Representation of Data

Visualizing data helps reveal patterns, trends, and relationships. Proper labeling and adherence to the area principle are essential for honest representation.

• Area Principle: The area occupied by a part of a graph should be proportional to the magnitude of the value it represents.

• Axes should be labeled with units, scales should be easy to read, and a key or caption should be provided.

Methods for Displaying Categorical Data

• Frequency Table: Lists categories and counts for each.

• Relative Frequency Table: Shows percentages instead of counts; must sum to 100%.

• Bar Chart: Compares discrete categories; bars separated by gaps.

• Pie Chart: Shows parts of a whole; best for proportions near 1/2, 1/4, or 1/8.

Contingency Tables and Conditional Distributions

Contingency tables summarize relationships between two categorical variables. Conditional distributions show the distribution of one variable for a subgroup defined by another variable.

• Marginal distributions: Overall, row, or column percentages.

• Side-by-side bar charts and segmented bar charts can visualize conditional distributions.

Simulations and Independence

• Statisticians often assume independence (null hypothesis) and simulate random chance to compare with real data.

• If real data differs significantly from simulation, a real relationship may exist.

Common Errors

• Never violate the area principle.

• Do not confuse similar-sounding percentages.

• Analyze totals in contingency tables.

• Simpson’s Paradox: Overall averages can be misleading when groups are unfairly averaged.

Describing Data Numerically

Graphical Methods for Quantitative Data

• Histogram: Bar graph without spaces; shows frequency or relative frequency.

• Stem-and-Leaf Display: Shows actual data values; good for small datasets.

• Dotplot: Each observation is a dot; highlights outliers and gaps.

• Boxplot: Based on five-number summary; ideal for comparing groups.

Comparison of Graphical Methods

Shape, Center, and Spread

• Shape: Modes (unimodal, bimodal, multimodal, uniform), symmetry, skewness.

• Center: Median (middle value), mean (average).

• Spread: Range, interquartile range (IQR), standard deviation.

• Outliers: Unusual values; can be identified using IQR * 1.5 rule.

Five-Number Summary and Boxplots

• Five-number summary: Max, Q3, Median, Q1, Min.

• Boxplot visualizes this summary; whiskers extend to values within 1.5 * IQR.

• Mean is powerful, but median is preferred if data is skewed or has outliers.

Comparing Groups

• Compare shape, center, spread, and unusual features across groups.

• Boxplots are effective for comparing several groups.

• Outliers should be investigated and corrected if errors are found.

Standardizing and the Normal Distribution

Shifting and Rescaling Data

• Shifting: Adding/subtracting a constant affects measures of center and position, not spread.

• Rescaling: Multiplying/dividing by a constant affects both center and spread.

Standardizing Data: z-Scores

Standardizing uses the standard deviation as a ruler to measure distance from the mean, creating z-scores. This allows comparison across different units and magnitudes.

• z-score: Tells how many standard deviations a value is from the mean.

• z-scores have a mean of 0 and a standard deviation of 1.

Formula for z-score (using sample statistics):

Formula for z-score (using population parameters):

The Normal Model and 68–95–99.7 Rule

• Normal model: Useful for unimodal, symmetric distributions.

• Parameters: Mean () and standard deviation ().

• Standard Normal model: (mean 0, SD 1).

• 68–95–99.7 Rule: In a Normal model, about 68% of values fall within 1 SD, 95% within 2 SDs, and 99.7% within 3 SDs of the mean.

Nearly Normal Condition: Distribution is nearly Normal if it is roughly unimodal and symmetric (checked via histogram or Normal probability plot).

Normal percentile: The percentage of values in a standard Normal distribution found at a given z-score or below.

Collecting Data: Sampling and Experimental Design

Sampling Principles

• Population: Entire group of interest.

• Sample: Subset of population.

• Sample survey: Questions asked to a sample to learn about the population.

• Bias: Over/underemphasizing certain characteristics; must be avoided.

• Randomization: Protects against known and unknown factors; ensures sample is representative.

• Sample size: Number of individuals matters more than fraction of population.

Sampling Methods

• Census: Attempt to survey entire population; often impractical.

• Simple Random Sample (SRS): Every individual has equal chance; selected via random numbers.

• Stratified Sampling: Population divided into groups (strata); samples taken from each group.

• Cluster Sampling: Population divided into clusters; entire clusters sampled.

• Systematic Sampling: Every nth individual selected, starting at random.

• Voluntary Response Sample: Individuals choose to respond; almost always biased.

• Convenience Sample: Sampling volunteers or easily accessible individuals; often biased.

Sampling Errors and Biases

• Bad sampling frame: Incomplete or inaccurate list of population.

• Undercoverage: Some population not sampled at all.

• Nonresponse bias: Sample members do not respond.

• Response bias: Survey design influences responses.

Experimental Design and Types of Studies

Types of Studies

• Sample survey: Estimates population parameters; requires representative sample.

• Observational study: No manipulation; can be retrospective or prospective.

• Experiment: Manipulates factor levels, randomly assigns subjects, compares responses.

Observational Studies

• Retrospective study: Subjects selected, previous conditions determined.

• Prospective study: Subjects followed to observe future outcomes.

Experimental Design Principles

• Control as many sources of variability as possible.

• Randomly assign experimental units to treatments.

• Replicate experiment on many subjects.

• Consider blocking to reduce variability from suspected sources.

• Use control groups, blinding, and placebo controls.

• Identify and address confounding and lurking variables.

Key Terms

• Factor: Variable manipulated in an experiment.

• Level: Specific value of a factor.

• Treatment: Combination of factor levels applied to subjects.

• Response variable: Outcome measured.

Statistical Inference: Confidence Intervals and Hypothesis Testing

Planning a Statistical Study

• Begin with a well-defined investigative question.

• Identify variables and their types before collecting data.

• Plan data description and analysis methods in advance.

Confidence Intervals

A confidence interval estimates a population parameter and provides a margin of error.

• Confidence interval: Interval of possible values for a parameter consistent with observed data.

• Margin of error: Amount by which parameter might differ from estimate.

• Confidence level: Probability that interval contains true parameter (commonly 90% or 95%).

Hypothesis Testing

Hypothesis tests determine whether observed patterns reflect real differences or random chance.

• Hypothesis test: Procedure to assess whether a pattern in data is due to chance.

• P-value: Probability that observed pattern could occur by random chance; small P-value suggests real effect.

Example: Testing whether the mean of a sample differs from the population mean.

Additional info: The notes cover foundational concepts for introductory statistics, including data collection, description, graphical and numerical summaries, sampling methods, experimental design, and basic statistical inference. These topics align with the first several chapters of a college-level statistics course.