Comprehensive Guide to Describing Quantitative and Categorical Data Distributions

Measures of Central Tendency and Order

• Median Placement and Definition:

  • The median represents the exact middle position of an ordered data set, where $50\%$ of the data falls above the value and $50\%$ falls below it.
  • To calculate the position index of the median in a full-sized data set with $n$ observations, use the formula:         $\frac{n + 1}{2}$
  • The median measures central tendency strictly based on relative position/order within the data set.

• Mean vs. Median Comparison:

  • Mean: The numerical average of all data point values combined. It is always a measure of center, but it is not always the most appropriate measure of center for a given data distribution.
  • Median: The middle observation value. It is the preferred measure of center for skewed distributions or distributions containing extreme outliers because of its mathematical resistance.
  • Symmetrical Distributions: When data is symmetric and mound-shaped (following a bell curve or normal distribution), the mean and median are approximately equal, making the mean an excellent measure of center.

Features of Distributions and Hedged Language

• Unusual Features:

  • Outliers: Observations that lie far away from the main body of the data.
    • Example: Earning mostly $100\%$ or $A$ grades on coursework and subsequently receiving a grade of $0$ on a single test creates an extreme outlier that severely pulls down the overall grade average.
  • Gaps: Intervals or ranges across a numeric scale where no data observations occur (frequency of $0$).
  • Clusters: Groups of observations that gather very tightly together in a localized section of the numeric scale.

• Sample vs. Population Distinction and Non-Definitive Language:

  • Unless full numerical verification is performed or data represents an entire population parameter, never state definitively that a feature is an outlier, gap, or cluster.
  • When analyzing sample statistics, use non-definitive/hedged wording such as "there appears to be a possible outlier", "possible gap", or "possible cluster".
  • Sampling Error Rationale: In sample data, unobserved population members might fall into apparent gaps or near isolated points. Extending beyond the sample scope to the broader population could reveal additional data points in those regions. Only complete population data allows absolute, definitive categorization without room for sampling error.

Shapes of Quantitative Distributions

• Unimodal / Symmetric:

  • Mound-shaped distribution featuring a single distinct peak ($1$ peak).
  • The left and right sides of the distribution form approximate mirror images of each other.

• Skewed Right:

  • A distribution where the majority of data points are concentrated/bunched on the left side, while a long tail end extends outward to the right.

• Skewed Left:

  • A distribution where the majority of data points are concentrated/bunched on the right side, while a long tail end extends outward to the left.

• Uniform:

  • A distribution with approximately equal frequencies across all bins or categories, appearing completely flat across the entire horizontal axis.

• Bimodal:

  • A distribution displaying two prominent distinct peaks ($2$ peaks) across its numeric range.

• Mandatory Language Rule for Shapes:

  • Always describe distribution shapes using qualifying modifiers such as "approximately symmetric", "approximately skewed right", or "possibly bimodal" to accommodate visual estimation and sample variability.

Spread and Variability Measures

• Definition of Spread:

  • Spread quantifies the variability of the data, measuring how far observations extend from the minimum value to the maximum value and how dispersed data points are relative to the center.

• Measures of Spread:

  • Range: The total difference between the highest value (maximum) and lowest value (minimum).         $\text{Range} = \text{Maximum} - \text{Minimum}$
    • The range is extremely sensitive to extreme values or outliers.
  • Interquartile Range (IQR):
    • Represents the range of the middle $50\%$ of the ordered data set ($IQR = Q_3 - Q_1$).
    • The IQR is mathematically resistant to extreme values.
  • Standard Deviation:
    • Measures the typical distance or dispersion of individual data observations away from the calculated mean ("give or take" relative to average).
    • Standard deviation is non-resistant to extreme values.
  • Variance:
    • The average squared deviation from the mean, representing another non-resistant measure of variability.

Resistance vs. Non-Resistance in Statistics

• Resistant Measures:

  • Measures that are not strongly affected by extreme outliers or skewness.
  • Examples: Median, Interquartile Range (IQR).
  • Explanation: The median depends entirely on positional ordering. Changing extreme values at the high or low ends does not alter the middle position index.

• Non-Resistant Measures:

  • Measures that are strongly affected by extreme outliers or skewness.
  • Examples: Mean, Range, Standard Deviation, Variance.
  • Explanation: An outlier (such as a single test score of $0$ amidst high scores) directly alters the arithmetic sum used to compute the mean, expands the maximum-minimum span of the range, and increases standard deviation and variance.

Framework for Describing Quantitative Distributions (CUSS & BS)

• The BS Rule (Be Specific / Context):

  • Every statistical description must include explicit context. FRQ answers must be written in complete sentences so that a reader can fully understand the scenario without referring back to the prompt.

• The CUSS Acronym:

  • C — Center: State the numerical center using the median (for skewed data/general distributions) or the mean (for symmetric distributions).
  • U — Unusual Features: Explicitly identify any possible outliers, possible gaps, or possible clusters. If no unusual features exist, explicitly state that there are no unusual features.
  • S — Shape: State whether the distribution is approximately symmetric, unimodal, bimodal, uniform, skewed left, or skewed right.
  • S — Spread: Provide the numerical variability boundaries (e.g., specifying range boundaries from minimum to maximum, IQR, or standard deviation).

Worked Examples of Quantitative Distribution Descriptions

• Example 1: Symmetric Distribution Centered at $35$:

  • Center: The center (mean or median) is approximately $35$.
  • Unusual Features: There are no unusual features (no outliers, gaps, or clusters).
  • Shape: Approximately symmetric and unimodal (mound-shaped).
  • Spread: Spans approximately from $23$ to $47$ (estimated values using hedged language).

• Example 2: Symmetric Distribution Centered at $70$:

  • Center: The mean or median center is approximately $70$.
  • Unusual Features: No outliers, gaps, or clusters are present.
  • Shape: Approximately symmetric and unimodal.
  • Spread: Spans approximately from $57$ to $82$.

• Example 3: Right-Skewed Distribution:

  • Center: Must use the median due to skewness; median is approximately $40$ (located near the peak/bulk of the data).
  • Unusual Features: No unusual features present.
  • Shape: Approximately skewed to the right.
  • Spread: Stated as an approximate range from the lowest observed value to the highest observed value.

• Example 4: UT Football Wins Since $1975$ Season (Dot Plot):

  • Context: UT football season win counts recorded annually since $1975$
  • Center: The center of the distribution of UT football season wins is a median of $9$ wins.
  • Unusual Features: There are no unusual features present in the distribution.
  • Shape: The shape of the distribution of UT football wins is approximately bimodal.
  • Spread: The spread of UT football season wins extends from $4$ to $13$ wins (or recorded span) since the $1975$ season.

Categorical vs. Quantitative Data Displays

• Categorical / Qualitative Data:

  • Graphical Displays: Bar graphs, Pie charts.
  • Properties: Data is sorted into distinct categories with no innate numerical order or priority.
  • Strict Prohibition: NEVER use quantitative shape descriptors (e.g., skewed left, skewed right, symmetric, unimodal, bimodal, uniform) to describe categorical/qualitative data distributions.

• Quantitative Data:

  • Graphical Displays: Dot plots, Histograms, Stem plots (stem-and-leaf plots), Cumulative relative frequency plots.
  • Properties: Plotted on a continuous numerical scale; fully described using CUSS and numerical descriptors.

• Histogram Construction and Bandwidth Sensitivity:

  • Histograms group continuous quantitative data into fixed range intervals called bins.
  • Bandwidth Effect: Altering bin widths changes visual perception. Narrower bin ranges (e.g., bin width of $1$ or $2$ units) expose subtle choppiness and localized peaks, whereas wider bin ranges smooth out variations and can obscure bimodal features.

Frequency, Relative Frequency, and Cumulative Relative Frequency

• Frequency: The exact count of individual data observations falling within a specific category or bin.

• Relative Frequency: The calculated proportion or percentage of observations within a bin relative to the total number of observations:     $\text{Relative Frequency} = \frac{\text{Frequency}}{\text{Total Observations}}$

  • In standard statistical practice, relative frequencies are computed and rounded to $4$ decimal places.

• Cumulative Relative Frequency:

  • The accumulated running sum of relative frequencies at or below a given interval boundary.
  • Calculated sequentially down a frequency table:
    1. Row $1$ Cumulative = Row $1$ Relative Frequency.
    2. Row $2$ Cumulative = Row $1$ Relative + Row $2$ Relative.
    3. Row $n$ Cumulative = Sum of all relative frequencies from Row $1$ through Row n$.\n * The final cumulative value at the last category must equal 1.0000$($100\%).\n * To find cumulative counts across an interval range (e.g., from 3$to$7), sum the frequencies/relative frequencies of all contained bins.\n\n\n# Questions and Audience Interactions\n\n* **Question on Gaps**: Does a gap require absolutely zero observations, or can an interval with a tiny amount of data still be called a gap?\n * *Clarification*: A gap is defined primarily as a section with no data (0$$ observations) at that specific point or interval.

• Question on Outlier Categorization: At what exact range can a data point be called a definitive outlier rather than a possible outlier?

  • Clarification: Range boundaries vary across datasets. Unless analyzing an entire population parameter, data from a sample must always be referred to as a "possible outlier at about [value]" to account for sampling error and unobserved population values.

• Question on Non-Definitive Modifiers: Is it mandatory to use modifiers like "about" or "possibly" for shape and spread features?

  • Clarification: Yes, hedged language is necessary to leave room for estimation error, especially when reading visual plots or working with sample data.

• Question on Visual Mean vs. Median Determination: Can the relationship between the mean and median (e.g., mean less than median) be definitively determined directly from a visual graph?

  • Clarification: Visual graphs alone do not allow definitive assertions because changes in bin width and scaling affect visual perception. Approximate language must be used unless precise numerical summary metrics are explicitly computed.