In Uniform Distribution we explore the continuous version of the uniform distribution where any number between α and β can be selected. There is also a discrete version of this distribution where α and β are integers and only integer values between these parameters can be selected. In fact, if we let N = β – α + 1, then the discrete uniform distribution determines the probability of selecting an integer between 1 and N at random. The probability density function f(x) and cumulative distribution function F(x) for this distribution are clearly
f(x) = 1/N F(x) = x/N
for x in the set {1, 2, …, N}.
The inverse cumulative distribution function is
F–1(p) = INT(Np)
Other key statistical properties are:
- Mean = (N + 1) / 2
- Median = (N + 1) / 2
- Mode = any x, 1 ≤ x ≤ N
- Variance = (N2 – 1) / 12
- Skewness = 0
- Kurtosis = -6(N2+1)/(5(N2-1))
Reference
Wikipedia (2019) Discrete uniform distribution
https://en.wikipedia.org/wiki/Discrete_uniform_distribution
I believe the variance is (N^2 – 1)/12, not (N-1)^2/12.
To demonstrate, enter {1,2,…,20} into A1:A20.
Then VARP(A1:A20) is 33.25. And so is ((A20-A1+1)^2 – 1)/12.
But ((A20-A1+1)-1)^2/12 = (A20-A1)^2/12 is 30.0833333333333.
I wrote: “I believe the variance is (N^2 – 1)/12, not (N-1)^2/12.“
I should say: according to the cited wikipage, namely https://en.wikipedia.org/wiki/Discrete_uniform_distribution .
However, when I try to calculate α and β in a manner similar to that demonstrated for the continuous uniform distribution at https://www.real-statistics.com/other-key-distributions/uniform-distribution, var=(N-1)^2/12, which is equivalent to var=(β-α)^2/12, seems to work better.
Is the Discrete_uniform_distribution wikipage wrong?
Hi Joe,
No, I don’t think that the Wikipedia page is wrong. I just made a mistake, possibly a typo.
If N = 2, then (N-1)^2/2 = 1/2, while (N^2-1)/12 = 1/4. To check that 1/4 is correct, I placed the formula =RANDBETWEEN(1,2) in cell A1 and then highlighted range A1:Q206 and pressed Ctrl-R and Ctrl-D. Finally, I calculated =VAR.P(A1:Q206) and obtained the value .249992, indicating that 1/4 is the correct value.
Charles
Thanks for the confirmation. I forgot that I had demonstrated the correctness of
var = (N^2-1)/12 in my first comment. Klunk!
But then can you help me with calculating α and β.
Based on the corrected formula for var, I believe that:
β = mean + (SQRT(var+1) – 1)/2
α = 2*mean – β, or equivalently α = mean – (SQRT(var+1) – 1)/2
But with {1,2,…,20} in A1:A20, we calculate:
B1: =AVERAGE(A1:A20) is 10.5
B2: =VARP(A1:A20) is 33.25
B5 (β): =B1+(SQRT(B2+1)-1)/2 is 12.9261749776799
B6 (α): =2*B1-B5 is 8.07382502232009
Even if we use VAR (sample var) instead of VARP, α=8 and β=13.
Obviously, α and β are wrong, because we know that the distribution is 1 to 20.
On the other hand, if we use the __continuous__ uniform distribution
var = (β-α)^2/12, or equivalently sd = (β-α)/SQRT(12), I believe that:
β = mean + sd*SQRT(12)/2
α = 2*mean – b, or equivalently α = mean – sd*SQRT(12)/2
Then we calculate:
B3: =STDEVP(A1:A20) is 5.7662812973354
B9 (β): =B1+B3*SQRT(12)/2 is 20.4874921777191
B10 (α): =2*B1-B9 is 0.512507822280909
If we use STDEV (sample sd) instead of STDEVP, as you do, α=0.253049234040404 and β=20.7469507659596.
In either case, α and β are about right for the actual distsribution of 1 to 20.
Am I miscalculating α and β with the discrete var?
Or is there some reason why we must always use the continuous var/sd to derive α and β?
Hello Joe,
I just emailed you a response to this comment and your email.
Charles
Thanks.
In your email, you note that I foolishly dropped the multiply-var-by-12 in my first derivation. Klunk!
Hello Joe,
Thanks for catching this error. I have just made the correction on the webpage.
Charles