Judging Whether a Normal Model Fits

Start with Shape, Not Just a Formula

In Interpreting Simulation Results Against a Model, you compared simulated results with what a stated probability model predicts. Here the question comes earlier: do the data’s shape and summary statistics make a normal model reasonable in the first place? This matters because a normal model is useful only when its shape is a credible representation of the variable being studied.

As covered in Features of a Normal Distribution, a normal distribution is symmetric, single-peaked, and bell-shaped. Its mean is at the center, and its two sides mirror one another. Real data do not need to match a perfect curve, but a noticeably long tail or strong asymmetry can make a normal model a poor choice.

The word “tend” is important. The relationship between mean and median is a clue, not a rule that classifies every data set. A few unusual values can pull the mean, and different shapes can produce similar means and medians. Use several pieces of evidence together rather than relying on one statistic.

Use Several Summary Statistics as Clues

The mean uses every observation, so unusually large or small values can affect it substantially. The median depends on the middle position in the ordered data and is less affected by extreme values. When a few high observations stretch the right tail, the mean is often pulled above the median. When a few low observations stretch the left tail, the mean is often pulled below it.

Quartiles add another useful comparison. The first quartile, \(Q_1\), marks the approximate 25th percentile; the third quartile, \(Q_3\), marks the approximate 75th percentile. Compare the distance from \(Q_1\) to the median with the distance from the median to \(Q_3\). If one side is much longer, that can point to asymmetry. Also note unusually distant minimum or maximum values: a long stretch toward the largest values is a clue of right skew.

A standard deviation describes spread around the mean, but it does not tell you whether the data are symmetric. Two distributions can have similar means and standard deviations while having very different shapes. This is why, in Checking Whether Data Are Approximately Normal, histograms, empirical-rule comparisons, and normal probability plots were treated as complementary evidence. Use summary statistics to guide your judgment, and use a graph when the data are available.

Context matters, too. A group of heights drawn from people of similar ages and other relevant characteristics might have a roughly symmetric, single-peaked distribution. Combining distinct groups can change the shape. Income distributions often have a practical lower limit but can extend far upward, because a small number of people may have much higher incomes than most. That pattern can produce a long right tail and a mean above the median. These are expectations to investigate, not assumptions to impose on every data set.

Worked Example: A Sample of Heights

Worked Example: A Right-Skewed Income Sample

Worked Example: Equal Mean and Median, but Not a Normal Shape

Make a Judgment, Not a Proof

These examples show why a normal-model judgment should draw on several clues. A mean noticeably above the median, uneven quartile spacing, and unusually high values point toward right skew. A mean noticeably below the median and a long stretch toward small values point toward left skew. Similar mean and median, balanced quartile spacing, and no striking extremes are more consistent with symmetry, but they do not guarantee a single-peaked, bell-shaped distribution.

There is no universal cutoff for how far apart the mean and median may be before a normal model is rejected. Their difference must be considered in relation to the variable’s units and spread. For example, a difference of 8.5 thousand dollars in the income example is meaningful alongside the median and the extreme observation; the number is not a general rule for all income data.

A normal model can also be unsuitable for reasons that summaries do not reveal. The data might have two clusters, a gap, or a shape that changes across subgroups. Conversely, a modest amount of irregularity in a sample does not automatically rule out a useful approximation. Use the context and the distribution’s overall shape, not a single summary statistic, to make the decision.

Common Mistakes and AP Exam Tips

• Treating mean and median as a formal test. A mean close to the median is a clue of symmetry, not proof of normality. Full-credit reasoning describes it as evidence and checks other features.
• Ignoring the direction of the tail. In a right-skewed distribution, high values tend to pull the mean above the median. In a left-skewed distribution, low values tend to pull it below. Name the direction and connect it to the context.
• Assuming standard deviation measures shape. Standard deviation measures spread around the mean. It does not tell whether the distribution is symmetric, skewed, or multi-peaked.
• Assuming all measurements are normal. Heights may be roughly symmetric within a suitable group, but mixing different ages or other populations can alter the shape. Describe the group being modeled.
• Calling an unusual value proof of skewness or model failure by itself. An extreme observation is a reason to investigate shape and context. Explain how it relates to the rest of the distribution.
• Using summaries while ignoring visible clusters. Equal mean and median can occur in a distribution that is not single-peaked. Look at the ordered data, histogram, or normal probability plot when available.

Key Takeaway

Summary statistics can reveal clues about a distribution’s symmetry and tails. The mean-median relationship, quartile spacing, and extreme values are useful together, especially when interpreted in context. A normal model is most plausible when the evidence supports a roughly symmetric, single-peaked shape; summaries can support that judgment, but they cannot replace examination of the distribution.