Scores, distributions & statistical meaning

START HERE

The essential idea

Nominal data classify without order; ordinal data indicate order without guaranteeing equal intervals. Interval scales have equal units without a meaningful absolute zero, while ratio scales also have an interpretable zero. The scale helps determine what summaries and comparisons make sense.

The mean is sensitive to extreme values; the median locates the middle ordered value; the mode describes the most frequent value. Variability describes spread rather than central location. In a skewed distribution, a typical value may be poorly represented by the mean alone. Always examine how scores were obtained and who is represented.

A z score expresses distance from a reference mean in standard deviation units. If a reference mean is 50 and its standard deviation is 10, a score of 60 has z = 1. This does not mean the person is “one unit healthier.” The interpretation depends on what the instrument measures, score direction, and supporting evidence.

A percentile describes relative standing in a reference distribution; it is not percentage correct. Measurement error limits precision, so avoid treating a small numerical difference as a meaningful clinical change without appropriate evidence. Reliability concerns consistency; validity concerns support for an intended interpretation and use.

BUILD THE REASONING · 01

Start with the actual values

Five fictional appointment waits are 2, 3, 3, 4, and 18 days. The mean is (2+3+3+4+18)/5 = 6 days; the median is 3 days; the range is 16 days. The mean captures the total waiting burden spread across people, while the median describes the middle observed wait. Neither alone shows that one person waited 18 days. Report the distribution and the question you are answering instead of selecting the nicer-looking statistic.

The sample standard deviation is the square root of the sum of squared deviations from the sample mean divided by n−1. It describes spread in the original units, not uncertainty about a particular individual’s true score. For heavily skewed values, a median and an interquartile range can complement a graph or listed values. Never remove an extreme value merely because it changes the average; first check whether it is a data error or a real experience.

BUILD THE REASONING · 02

Convert a score without changing its meaning

For a fictional symptom scale with reference mean 50 and standard deviation 10, a score of 65 has z = (65−50)/10 = 1.5. A conventional T-score transformation is T = 50 + 10z, so z = 1.5 corresponds to T = 65. These calculations state distance from the reference mean. They do not establish a diagnosis or indicate how much of a curriculum was mastered.

If the reference distribution is approximately normal, z = 1 corresponds to about the 84th percentile; z = 1.5 corresponds to about the 93rd. That conversion depends on the distribution assumption. An actual instrument’s norm table can use observed distributions and its own scoring rules. Percentiles are ranks, not equal-interval units: moving from the 50th to 60th percentile is not necessarily the same raw-score change as moving from the 90th to 100th.

BUILD THE REASONING · 03

Add precision to the explanation

Under a simple classical-test model, a standard error of measurement can be approximated as SEM = SD × √(1−reliability). If SD = 10 and a suitable reliability estimate is .84, SEM = 10 × .4 = 4 points. An illustrative normal-error interval around an observed score of 60 is approximately 60 ± 1.96×4, or 52.2–67.8. Real instruments may require conditional errors or other manual-specified intervals; use their procedures.

A change from 60 to 62 is not automatically meaningful when scores contain measurement error. Nor does an individual-score interval describe uncertainty in a group mean. Interpret change using suitable change-score reliability, practice effects, expected natural variation, functional evidence, and the instrument’s validated use. Precision helps explain what a number cannot settle.

FROM CONCEPT TO DECISION

Two scales, two different-looking numbers

1. 1

   Standardize each within its own reference

   Scale A: z = 15/10 = 1.5. Scale B: z = 4/4 = 1.0. The larger raw number alone was not the basis for comparison.

2. 2

   Limit the comparison

   A is farther above its own reference mean. Because the constructs and norm groups differ, this is not proof that A represents a more serious clinical problem.

3. 3

   Check distribution and precision

   Use the manuals’ norm tables and appropriate score intervals rather than assuming both scales are normal and equally precise.

4. 4

   Explain in context

   Connect the findings to the person’s report and functioning, and state which interpretation each instrument supports.

MAKE THE REASONING YOURS

Work through a new situation

READ WITH A PURPOSE

Best references

These readings deepen the specific models, evidence, and decisions in this lesson.

1. 01
   NIST: Measures of Location

   Compare mean and median when an extreme observation changes the shape of the data.

2. 02
   NIST: Normal Distribution

   Review standardization and the normal-distribution assumption behind approximate percentile conversions.

3. 03
   AERA/APA/NCME: Standards for Educational and Psychological Testing (2014)

   Read Chapters 1–3 on intended interpretations, precision, and fairness. Use the standards to evaluate evidence for a particular use, not to label an instrument universally valid.