Name the tail, not the tallest bar
A distribution has positive, or right, skew when its tail extends toward larger values. Negative, or left, skew means the tail extends toward smaller values. The main concentration of observations often lies on the opposite side. Start with a dot plot or histogram rather than labeling skew from the location of the peak alone.
Why mean and median can differ
The mean uses every value and its numerical size. The median uses middle positions after sorting. A few very large observations can pull the mean upward without moving the median by the same amount; very small observations can pull it downward. Thus mean above median often accompanies right skew, while mean below median often accompanies left skew. This is a descriptive tendency, not a logical equivalence for every dataset.
New examples with both tail directions
Consider the illustrative values . Their mean is 4, median is 2, and mode is 2. The isolated value 12 gives a long right tail.
Now reflect every value about 10, using . Sorted, the reflected data are . Their mean is 16, median is 18, and mode is 18. The low value 8 creates the corresponding left tail. The transformation makes the tail direction reverse while keeping the center calculations transparent.
Symmetry requires matching shape
Exact symmetry about a center requires values equally far to its left and right to have matching frequencies. Equal minimum and maximum distances are insufficient. Likewise, equal mean and median alone do not prove symmetry. For the symmetric dataset , the mean and median are both 4, but it has no unique mode. Do not assume every symmetric distribution has one central peak.
Numerical skewness and small samples
One moment-based measure uses cubed deviations:
It requires nonzero variance. Positive values indicate positive moment skewness and negative values indicate negative moment skewness under this definition. Software may report a bias-adjusted sample coefficient instead, so identify the definition when comparing values. Cubing makes this measure sensitive to extreme observations.
A small, discrete, or multimodal sample may not fit a simple one-tail description well. Report its peaks, gaps, and tail pattern, and use numerical coefficients as additional evidence. Rebinning a histogram can change its visual impression without changing the data. Mean, median, and a shape plot should be read together rather than turned into an automatic rule.
Related question
Apply this knowledge
Use the concept guide to understand the reasoning, then return to the complete question and worked answer.
Positive vs Negative Skew: Letter-Count Dot Plots and CentersSources
These references support the core concepts and interpretation boundaries explained above.