Percentiles, Interquartile Range, Z-Scores, and Boxplots Study Guide

Percentiles and Quartiles

  • Overview of Location Measurements:

    • Standard statistical metrics such as the sample mean and standard deviation are sensitive to extreme values (outliers).
    • Extreme values inflate the standard deviation and pull the mean toward the tail of the data.
    • Percentiles serve as measurements of location within an ordered dataset, offering position metrics relative to the overall distribution.
  • Percentiles (PkP_k):

    • Denoted by PkP_k, where kk is an integer from 11 to 9999 inclusive (1≤k≤991 \le k \le 99).
    • A percentile represents a value or group of values accounting for approximately 1%1\% of the data set.
    • Definition: The kk\text{th} percentile (PkP_k) is the value or group of values larger than or equal to approximately k%k\% of all data in the dataset.
    • Dual Perspectives of Interpretation:
    • Greater-Than-or-Equal Perspective: A score at the kk\text{th} percentile is greater than or equal to approximately k%k\% of all values.
    • Less-Than Perspective: A score at the kk\text{th} percentile is strictly less than (100−k)%(100 - k)\% of all values in the dataset.
    • Example: An SAT Math score of 680680 out of 800800 at the 82nd82\text{nd} percentile (P82P_{82}) means the score is greater than or equal to approximately 82%82\% of all scores, and lower than approximately 18%18\% (100%−82%100\% - 82\%) of all test scores.
    • Example: The 78th78\text{th} percentile (P78P_{78}) is greater than or equal to 78%78\% of values and smaller than 22%22\% of values.
    • Example: The 22nd22\text{nd} percentile (P22P_{22}) is greater than or equal to 22%22\% of values and smaller than 78%78\% of values.
  • Group Representation in Large Datasets:

    • A percentile does not need to correspond to a single unique numerical score; it represents a range or group of values accounting for 1%1\% of data.
    • In large datasets, distinct scores (e.g., 680680 and 660660) may both reside within the 82nd82\text{nd} percentile because percentiles group values into whole integer percentage bands.
  • Non-Universality of Percentile Estimation Procedures:

    • There is no universally agreed-upon standard algorithm among statisticians or textbook curricula for calculating precise percentiles or quartiles.
    • Computational techniques vary slightly depending on the statistical software or method implemented.
  • Quartiles (Q1Q_1, Q2Q_2, Q3Q_3):

    • Quartiles split an ordered dataset into four equal parts, corresponding to three primary percentiles:
    • First Quartile (Q1Q_1): The 25th25\text{th} percentile (P25P_{25}). Greater than or equal to 25%25\% of data; smaller than the remaining 75%75\%.
    • Second Quartile (Q2Q_2): The 50th50\text{th} percentile (P50P_{50}). Greater than or equal to 50%50\% of data; smaller than the remaining 50%50\%. Equivalent to the dataset median.
    • Third Quartile (Q3Q_3): The 75th75\text{th} percentile (P75P_{75}). Greater than or equal to 75%75\% of data; smaller than the remaining 25%25\%.
  • Procedure for Calculating Quartiles:

    • Data Preparation: Arrange data values in ascending chronological order from smallest to largest.
    • Finding Q2Q_2 (Median):
    • For an odd number of data values (nn is odd), the median is the single central value at position n+12\frac{n+1}{2}.
    • For an even number of data values (nn is even), the median is the arithmetic mean of the two central values at positions n2\frac{n}{2} and n2+1\frac{n}{2} + 1
    • Finding Q1Q_1 and Q3Q_3:
    • Q1Q_1 is the median of the lower half of data (all observations positioned strictly below the overall median location).
    • Q3Q_3 is the median of the upper half of data (all observations positioned strictly above the overall median location).
  • Detailed Calculation Example (n=20n = 20 Data Points):

    • Consider 2020 test scores ordered from smallest to largest.
    • Calculating Q2Q_2:
    • Dataset size n=20n = 20 is even. The central pair consists of the 10th10\text{th} value (7575) and the 11th11\text{th} value (8080).
    • Q2=75+802=77.5Q_2 = \frac{75 + 80}{2} = 77.5
    • Calculating Q1Q_1:
    • The lower half contains 1010 values (an even set size). The central pair of the lower half consists of the 5th5\text{th} value (6060) and the 6th6\text{th} value (7070).
    • Q1=60+702=65Q_1 = \frac{60 + 70}{2} = 65
    • Calculating Q3Q_3:
    • The upper half contains 1010 values (an even set size). The central pair of the upper half consists of the 5th5\text{th} value of the top row (8585) and the 6th6\text{th} value of the top row (9090).
    • Q3=85+902=87.5Q_3 = \frac{85 + 90}{2} = 87.5
    • Note on subsets: If dataset size n=30n = 30, the overall median splits the data into two halves of 1515 values. Since 1515 is odd, Q1Q_1 and Q3Q_3 are found as single middle values (8th8\text{th} observation in each half).

Measures of Dispersion and the Interquartile Range (IQR)

  • The Middle 50%50\% of Data:

    • The interval between Q1Q_1 (25th25\text{th} percentile) and Q3Q_3 (75th75\text{th} percentile) bounds the central 50%50\% of data, referred to as the middle half.
  • Definition and Formula for Interquartile Range (IQRIQR):

    • The Interquartile Range (IQRIQR) is a measurement of dispersion quantifying the spread of the middle 50%50\% of data.
    • Formula:     IQR=Q3−Q1IQR = Q_3 - Q_1
    • Calculation using the n=20n = 20 test score dataset:     IQR=87.5−65=22.5IQR = 87.5 - 65 = 22.5
  • Outlier Resistance Properties:

    • Standard deviation and overall range (Maximum−Minimum\text{Maximum} - \text{Minimum}) are non-resistant to outliers; extreme values alter their metrics significantly.
    • IQRIQR is an outlier-resistant measure of dispersion because it evaluates only the internal 50%50\% of data, ignoring extreme high or low values.
  • **Paired Statistical Applications (