统计系列（三）利用Python进行参数估计

统计系列（三）利用Python进行参数估计

点估计

样本均值估计为总体均值，样本比例估计为总体比例。

import numpy as np

x = [1, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 1, 1, 0, 1, 0, 0] # 1为成功，0为失败。估计成功失败比率

theta = np.mean(x)
h = theta/(1-theta)
print(h)

1.2500000000000002

区间估计

单正态总体均值区间估计

• 方差$\sigma$已知时$\mu$的置信区间

  Exp: $X\sim N(\mu ,0.6)$，样本x=[14.6, 15.1, 14.9, 14.8, 15.2, 15.1]，估计样本均值置信度为95%的置信区间

import numpy as np
import scipy.stats as ss

n = 6; p = 0.025; sigma = np.sqrt(0.6) 
x = [14.6, 15.1, 14.9, 14.8, 15.2, 15.1]
xbar = np.mean(x)
low = xbar-ss.norm.ppf(q=1-p)*(sigma/np.sqrt(n))
up = xbar+ss.norm.ppf(q=1-p)*(sigma/np.sqrt(n))

print([low,up])

[14.330204967695439, 15.569795032304564]

• 方差$\sigma$未知时$\mu$的置信区间

  Exp: $X\sim N(\mu ,{\sigma }^2)$，样本x=[99.3, 98.7, 100.5, 101.2, 98.3, 99.7, 99.5, 102.1, 100.5]，估计样本均值置信度为95%的置信区间

import numpy as np
import scipy.stats as ss
from scipy.stats import t

n = 9; p = 0.025;
x = [99.3, 98.7, 100.5, 101.2, 98.3, 99.7, 99.5, 102.1, 100.5]
xbar = np.mean(x)
s2 = np.var(x, ddof=1) # 样本方差除以n-1
s = np.sqrt(s2)
low = xbar-ss.t.ppf(1-p, n-1)*(s/np.sqrt(n))
up = xbar+ss.t.ppf(1-p, n-1)*(s/np.sqrt(n))

print([low,up])

[99.04599342616191, 100.90956212939363]

单正态总体方差区间估计

• $\mu$未知时方差${\sigma }^2$的置信区间

  实际中$\mu$已知的情况极少。Exp:样本量为16，样本均值为12.8，样本方差为0.0023。估计总体方差置信度为95%的置信区间

from scipy.stats import chi2

n=16; s2=0.0023; p=0.025
low = ((n-1)*s2)/chi2.ppf(1-p,n-1)
up = ((n-1)*s2)/chi2.ppf(p,n-1)
print([low,up])

[0.0012550751937877684, 0.005509300678006194]

双正态总体均值差区间估计

• ${\sigma }_1^2$、${\sigma }_2^2$已知，求$\mu 1-\mu 2$的置信区间

  Exp:样x=[628, 583, 510, 554, 612, 523, 530, 615]，且$x\sim N(\mu ,2140)$;样本y=[535, 433, 398, 470, 567, 480, 498, 560, 503, 426]，且$y\sim N(\mu ,3250)$，求两样本均值差的95%置信区间

import numpy as np
import scipy.stats as ss

x = [628, 583, 510, 554, 612, 523, 530, 615]
y = [535, 433, 398, 470, 567, 480, 498, 560, 503, 426]
n1 = len(x); n2 = len(y)
xbar = np.mean(x); ybar = np.mean(y)
x_s2 = 2140; y_s2=3250; p=0.025
low = xbar-ybar-ss.norm.ppf(q=1-p)*np.sqrt(x_s2/n1+y_s2/n2)
up = xbar-ybar+ss.norm.ppf(q=1-p)*np.sqrt(x_s2/n1+y_s2/n2)
print([low,up])

[34.66688380095825, 130.08311619904174]

• 两总体方差未知，但已知${\sigma }_1^2={\sigma }_2^2={\sigma }^2$，求$\mu 1-\mu 2$的置信区间

  Exp:样本x=[628, 583, 510, 554, 612, 523, 530, 615]，且$x\sim N(\mu ,{\sigma }^2)$;样本y=[535, 433, 398, 470, 567, 480, 498, 560, 503, 426]，且$y\sim N(\mu ,{\sigma }^2)$，求两样本均值差的95%置信区间

import numpy as np
import scipy.stats as ss

x = [628, 583, 510, 554, 612, 523, 530, 615]
y = [535, 433, 398, 470, 567, 480, 498, 560, 503, 426]
n1 = len(x); n2 = len(y)
xbar = np.mean(x); ybar = np.mean(y)
x_s2 = np.var(x); y_s2=np.var(y); p=0.025
s2 = ((n1-1)*x_s2+(n2-1)*y_s2)/(n1-1+n2-1)
low = xbar-ybar-ss.t.ppf(1-p, n1+n2-1)*np.sqrt(s2*(1/n1+1/n2))
up = xbar-ybar+ss.t.ppf(1-p, n1+n2-1)*np.sqrt(s2*(1/n1+1/n2))
print([low,up])

[32.65868498172926, 132.09131501827073]

双正态总体方差比区间估计

• $\mu$未知时方差${\sigma }_1^2/{\sigma }_2^2$的置信区间

  Exp:x=[20.5, 19.8, 19.7, 20.4, 20.1, 20.0, 19.0, 19.9]，且$x\sim N({\mu }_1,{\sigma }_1^2)$; y=[20.7, 19.8, 19.5, 20.8, 20.4, 19.6, 20.2]，且$y\sim N({\mu }_2,{\sigma }_2^2)$，试求两方差比的95%置信区间

import numpy as np
from scipy.stats import f

x = [20.5, 19.8, 19.7, 20.4, 20.1, 20.0, 19.0, 19.9]
y = [20.7, 19.8, 19.5, 20.8, 20.4, 19.6, 20.2]

x_s2 = np.var(x); y_s2=np.var(y)
n1 = len(x); n2 = len(y); p = 0.025
low = x_s2/y_s2*1/f.ppf(1-p, n1-1, n2-1)
up = x_s2/y_s2*1/f.ppf(p, n1-1, n2-1)

print([low,up])

[0.1421688673708112, 4.144622814076891]

单总体比例区间估计

Exp:随机变量x的转化率为15%，样本量为20，求总体转化率的95%置信区间

import numpy as np
import scipy.stats as ss

n = 60; p = 0.025; tr = 0.15 
low = tr-ss.norm.ppf(q=1-p)*np.sqrt(tr*(1-tr)/n)
up = tr+ss.norm.ppf(q=1-p)*np.sqrt(tr*(1-tr)/n)

print([low,up])

[0.05965012454920031, 0.24034987545079967]

总结

最常见的就是总体方差未知时，估计总体的均值u；总体服从二项分布，估计总体的比例p。如果遇到其他情形下的参数估计，同样只需要按照给定公式计算即可。

共勉～