Why, when I calculate the skewness of a series with constants, do I not get NaN?

5 ビュー (過去 30 日間)
Angelavtc
Angelavtc 2021 年 10 月 11 日
コメント済み: Cris LaPierre 2021 年 10 月 11 日
Dear Matlab community,
I have a simulation of normally distributed numbers with a mean of 76 and a standard deviation of 0, which describes a series of constant numbers with the value 76. To this simulation, I apply a function that also yields constants and, therefore, its skewness should be also NaN. Why, when I calculate their skewness, do I not get NaN but 1? How do I correct this error?
Thanks in advance!
Np=20;
cb=2;
ab = 30*(Np/100)^(cb-1);
n=100;
Qd = normrnd(76,0,n,1);
Pw=ab*((Qd/Np).^(cb-1));
skewness(Pw)
skewness(Qd)
var(Qd)

採用された回答

Cris LaPierre
Cris LaPierre 2021 年 10 月 11 日
I suspect this has to do with how floating point numbers are stored in a computer. 76 can be stored exactly as is, but 22.8000 is not exactly 22.8 in your computer. If you round Pw, you will see the skewness is also NaN.
format long
Np=20;
cb=2;
ab = 30*(Np/100)^(cb-1);
Qd = normrnd(76,0,100,1);
Pw=ab*((Qd/Np).^(cb-1));
Pw(1)
ans =
22.799999999999997
skewness(Pw)
ans =
-1
var(Pw)
ans =
1.274926715508706e-29
skewness(round(Pw))
ans =
NaN
skewness(Qd)
ans =
NaN
var(Qd)
ans =
0
  3 件のコメント
Angelavtc
Angelavtc 2021 年 10 月 11 日
@Cris LaPierre However, this is creating another kind of problem for simulations with variance different than 0. For example when:
clear, clc;
Np=20;
cb=1.1;
ab = 30*(Np/100)^(cb-1);
n=10000;
Qd = normrnd(108,3,n,1);
Pw=ab*((Qd/Np).^(cb-1));
Skew1=skewness(Pw)
Skew1 = -0.0787
Skew2=skewness(round(Pw))
Skew2 = 33.2883
Skew1 and Skew2 are completely different, being Skew1 the one close to 0 (my correct answer). How can I correct this?
Thank you!
Cris LaPierre
Cris LaPierre 2021 年 10 月 11 日
Consider taking a look at the algorithm the skewness function is using.

サインインしてコメントする。

その他の回答 (0 件)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by