The Cauchy distribution looks like a normal curve but has much heavier tails, so extreme values are far more frequent. The Cauchy distribution calculator takes a location, a scale and two bounds and gives the probability between the bounds, the probabilities below and above, the median, the interquartile width, the peak density and the chance of being far from the centre, compared with a normal distribution.
How the calculation works
- Cumulative probability F(x) = ½ + arctan((x − x0) ÷ γ) ÷ π. The probability between a and b is F(b) − F(a).
- Median and mode both equal the location x0. The mean and the variance of a Cauchy distribution do not exist.
- Quartiles are at x0 − γ and x0 + γ, so the middle half of the probability lies in a band of width 2γ. The peak density is 1 ÷ (π × γ).
- Probability of being more than k·γ from the centre = 1 − 2 × arctan(k) ÷ π. For a normal distribution with the same quartiles, the same probability is much smaller.
Example
For a Cauchy distribution with location 2 and scale 3, the probability between 0 and 5 is (arctan(1) − arctan(−0.667)) ÷ π = (45° + 33.69°) ÷ 180° = 43.72%. The median is 2 and the peak density is 1 ÷ (π × 3) = 0.1061. The chance of falling more than 10 scale units from the centre is 1 − 2 × arctan(10) ÷ π = 6.34%, whereas a normal distribution with the same quartiles would give only about 0.000002%.
Why it matters
The distribution models resonance lines in physics, the ratio of noisy measurements and outliers in data. It is also the standard example of a distribution without a mean, which warns against applying averages to heavy-tailed data.
Good to know
Results are exact for an ideal Cauchy distribution. The location and scale must be estimated from data with the median and the interquartile range, since the sample mean is not reliable. The normal comparison uses a normal distribution that has the same quartiles, to make the heavy tail visible.
Related tools
The margin of error calculator and the standard error calculator assume well-behaved data, which a Cauchy distribution is not.