幂律分布:all you need to know about power law distribution

For power law distribution
p k = ν − 1 k m i n ( k k m i n ) − ν p_k=\frac{\nu-1}{k_{min}}(\frac{k}{k_{min}})^{-\nu} pk=kminν1(kmink)ν
To estimate ν \nu ν by moment estimation method
F o r   m e t h o d   o f   m o m e n t s   E [ x ] = ν − 1 ν − 2 E [ x 2 ] = ν − 1 ν − 3 x m i n 2 x ^ m i n = ∑ x i 2 ( ν − 3 ) ∑ x i ( ν − 2 ) v ^ M M E = − 2 ( ∑ x i ) 2 + 2 n ∑ x i 2 ± − n ( ∑ x i ) 2 ∑ x i 2 + n 2 ( ∑ x i 2 ) 2 − ( ∑ x i ) 2 + n ∑ x i 2 F o r   m a x i m i z i n g   l o g − l i k e l i h o o d p ( x ) = ∏ i = 1 n ν − 1 x m i n ( x x m i n ) − ν l o g p ( x ) = n l o g ( ν − 1 ) − n l o g x m i n − ν ∑ i = 1 ∞ l o g x i x m i n ∂ l o g p ( x ) ∂ x m i n = − n x m i n + ν x m i n ∑ i = 1 ∞ 1 x i = 0 ∂ l o g p ( x ) ∂ ν = n ν − 1 − ∑ i = 1 ∞ l o g x i x m i n ν ^ M L E = 1 + n [ ∑ i = 1 n l n x i x m i n ] − 1 For ~method~of~moments\\ ~E[x]=\frac{\nu-1}{\nu-2}\\ E[x^2]=\frac{\nu-1}{\nu-3}x_{min}^2\\ \hat{x}_{min}=\frac{\sum x_i^2(\nu-3)}{\sum x_i(\nu-2)}\\ \hat{v}_{MME}=\frac{-2(\sum x_i)^2+2n\sum x_i^2 \pm \sqrt{-n(\sum x_i)^2 \sum x_i^2+n^2 (\sum x_i^2)^2}}{-(\sum x_i)^2+n\sum x_i^2}\\ For~maximizing ~log-likelihood\\ p(x)=\prod_{i=1}^n\frac{\nu-1}{x_{min}}(\frac{x}{x_{min}})^{-\nu}\\ logp(x)=nlog(\nu-1)-nlogx_{min}-\nu\sum_{i=1}^\infty log\frac{x_i}{x_{min}}\\ \frac{\partial logp(x)}{\partial x_{min}}=\frac{-n}{x_{min}}+\frac{\nu}{x_{min}}\sum_{i=1}^\infty \frac{1}{x_i}=0\\ \frac{\partial logp(x)}{\partial \nu}=\frac{n}{\nu-1}-\sum_{i=1}^\infty log\frac{x_i}{x_{min}}\\ \hat{\nu}_{MLE}=1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1}\\ For method of moments E[x]=ν2ν1E[x2]=ν3ν1xmin2x^min=xi(ν2)xi2(ν3)v^MME=(xi)2+nxi22(xi)2+2nxi2±n(xi)2xi2+n2(xi2)2 For maximizing loglikelihoodp(x)=i=1nxminν1(xminx)νlogp(x)=nlog(ν1)nlogxminνi=1logxminxixminlogp(x)=xminn+xminνi=1xi1=0νlogp(x)=ν1ni=1logxminxiν^MLE=1+n[i=1nlnxminxi]1
To find the statistical properties of these estimator including unbiasedness and consistency
E [ ν ^ M M E ] = − 2 ( n ν − 1 ) 2 + 2 n n 2 ( ν − 1 ) ( ν − 2 ) ± − n ( n ν − 1 ) 2 n 2 ( ν − 1 ) ( ν − 2 ) + n 2 ( n 2 ( ν − 1 ) ( ν − 2 ) ) 2 − ( n ν − 1 ) 2 + n n 2 ( ν − 1 ) ( ν − 2 ) w h e n   n → ∞ , E [ ν ^ M M E ] → 3   o r   1 E [ ν ^ M L E ] = 1 + n E [ ∑ i = 1 n l n x i x m i n ] − 1 ] = 1 + E [ n z ] = 1 + n ( ν − 1 ) n − 1 = n ν n − 1 − 1 n − 1 w h e n   n → ∞ , E [ ν ^ M L E ] → ν E[\hat{\nu}_{MME}]=\frac{-2(\frac{n}{\nu-1})^2+2n\frac{n^2}{(\nu-1)(\nu-2)}\pm \sqrt{-n(\frac{n}{\nu-1})^2\frac{n^2}{(\nu-1)(\nu-2)}+n^2(\frac{n^2}{(\nu-1)(\nu-2)})^2}}{-(\frac{n}{\nu-1})^2+n\frac{n^2}{(\nu-1)(\nu-2)}}\\ when~n \rightarrow \infin , E[\hat{\nu}_{MME}]\rightarrow 3~or~1\\ E[\hat{\nu}_{MLE}]=1+nE[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1}]\\ =1+E[\frac{n}{z}]\\ =1+\frac{n(\nu-1)}{n-1}\\ =\frac{n\nu}{n-1}-\frac{1}{n-1}\\ when~n \rightarrow \infin , E[\hat{\nu}_{MLE}]\rightarrow \nu\\ E[ν^MME]=(ν1n)2+n(ν1)(ν2)n22(ν1n)2+2n(ν1)(ν2)n2±n(ν1n)2(ν1)(ν2)n2+n2((ν1)(ν2)n2)2 when n,E[ν^MME]3 or 1E[ν^MLE]=1+nE[i=1nlnxminxi]1]=1+E[zn]=1+n1n(ν1)=n1nνn11when n,E[ν^MLE]ν
So MME estimator and MLE estimator are both biased estimator and MLE estimator is a consisent estimator while MME estimator is not.

For joint PDF
f ( x ; ν ) = ( ν − 1 ) n x m i n − n ( ν − 1 ) ∏ i = 1 n x i − ν 1 x i ≥ x m i n = c ( ν ) h ( x ) e x p [ ∑ j = 1 k q j ( ν ) t j ( x ) ] = 1 x i ≥ x m i n ( ν n x m i n n ( ν − 1 ) ) e x p [ − ν ∑ i = 1 n l n x i ] f(x;\nu)=(\nu-1)^n x_{min}^{-n(\nu-1)}\prod_{i=1}^nx_i ^{-\nu}1_{x_i\geq x_{min}}\\ =c(\nu)h(x)exp[\sum_{j=1}^k q_j(\nu)t_j(x)]\\ =1_{x_i\geq x_{min}}(\nu ^n x_{min}^{n(\nu-1)})exp[-\nu \sum_{i=1}^n ln x_i]\\ f(x;ν)=(ν1)nxminn(ν1)i=1nxiν1xixmin=c(ν)h(x)exp[j=1kqj(ν)tj(x)]=1xixmin(νnxminn(ν1))exp[νi=1nlnxi]
So ∑ i = 1 n l n x i \sum_{i=1}^n lnx_i i=1nlnxi is complete and sufficient.

Let $ Y=ln\frac{x}{x_{min}}$,so x = x m i n e Y x=x_{min}e^Y x=xmineY
f X ( x ) = ( ν − 1 ) x m i n ν − 1 x − ν 1 x ≥ x m i n f Y ( y ) = f X ( x m i n e Y ) ∣ d x d y ∣ = ( ν − 1 ) x m i n ν − 1 ( x m i n e Y ) − ν 1 x m i n e Y ≥ x m i n x m i n e Y = ( ν − 1 ) e − y ( ν − 1 ) 1 Y ≥ 0 f_X(x)=(\nu-1)x_{min}^{\nu-1}x^{-\nu}1_{x\geq x_{min}}\\ f_Y(y)=f_X(x_{min}e^Y)|\frac{dx}{dy}|\\ =(\nu-1)x_{min}^{\nu-1}(x_{min}e^Y)^{-\nu}1_{x_{min}e^Y\geq x_{min}}x_{min}e^Y\\ =(\nu-1)e^{-y(\nu-1)}1_{Y\geq 0}\\ fX(x)=(ν1)xminν1xν1xxminfY(y)=fX(xmineY)dydx=(ν1)xminν1(xmineY)ν1xmineYxminxmineY=(ν1)ey(ν1)1Y0
So y follows the exponential distribution with rate ν − 1 \nu-1 ν1
z = ∑ i = 1 n Y i = ∑ i = 1 n l n X i − l n X m i n z ∼ Γ ( n . ν − 1 ) E ( Y i ) = 1 ν − 1 E ( z ) = n ν − 1 z=\sum_{i=1}^nY_i=\sum_{i=1}^nlnX_i-lnX_{min}\\ z\sim \Gamma(n.\nu-1)\\ E(Y_i)=\frac{1}{\nu-1}\\ E(z)=\frac{n}{\nu-1}\\ z=i=1nYi=i=1nlnXilnXminzΓ(n.ν1)E(Yi)=ν11E(z)=ν1n
Therefore 1 n ∑ i = 1 n Y i = 1 ν − 1 \frac{1}{n}\sum_{i=1}^n Y_i=\frac{1}{\nu-1} n1i=1nYi=ν11 which means 1 n ∑ i = 1 n Y i \frac{1}{n}\sum_{i=1}^n Y_i n1i=1nYi is an unbiased estimator for 1 ν − 1 \frac{1}{\nu-1} ν11, we guess 1 + 1 1 n ∑ i = 1 n Y i 1+\frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i} 1+n1i=1nYi1 is an unbiased estimator for ν \nu ν.
E [ n z ] = E [ 1 1 n ∑ i = 1 n Y i ] = n ∫ 0 ∞ 1 z 1 Γ ( n ) ( ν − 1 ) n z n − 1 e − ( ν − 1 ) z d z = n ( ν − 1 ) Γ ( n − 1 ) Γ ( n ) ∫ 0 ∞ 1 Γ ( n − 1 ) ( ν − 1 ) n − 1 z n − 2 e − ( ν − 1 ) z d z = n ( ν − 1 ) n − 1 A c c o r d i n g   t o   L e h m a n n − S c h e f f e   T h e o r e m ∑ i = 1 n l n x i   i s   c o m p l e t e   a n d   s u f f i c i e n t ∵ T = n − 1 n 1 1 n ∑ i = 1 n Y i + 1   i s   u n b i a s e d   f o r   τ ( ν ) = ν E[\frac{n}{z}]=E[\frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i}]\\ =n\int_0^\infin \frac{1}{z}\frac{1}{\Gamma(n)}(\nu-1)^nz^{n-1}e^{-(\nu-1)z}dz\\ =n\frac{(\nu-1)\Gamma(n-1)}{\Gamma(n)}\int_0^\infin \frac{1}{\Gamma(n-1)}(\nu-1)^{n-1}z^{n-2}e^{-(\nu-1)z}dz\\ =\frac{n(\nu-1)}{n-1}\\ According~to~Lehmann-Scheffe ~Theorem\\ \sum_{i=1}^n lnx_i ~is~complete ~and~sufficient\\ \because T=\frac{n-1}{n} \frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i}+1~is~unbiased~for~\tau(\nu)=\nu\\ E[zn]=E[n1i=1nYi1]=n0z1Γ(n)1(ν1)nzn1e(ν1)zdz=nΓ(n)(ν1)Γ(n1)0Γ(n1)1(ν1)n1zn2e(ν1)zdz=n1n(ν1)According to LehmannScheffe Theoremi=1nlnxi is complete and sufficientT=nn1n1i=1nYi1+1 is unbiased for τ(ν)=ν
So an unbiased for ν \nu ν is ν ^ g o o d = n − 1 n 1 1 n ∑ i = 1 n Y i + 1 = n − 1 ∑ i = 1 n l n X i − l n X m i n + 1 \hat{\nu}_{good}=\frac{n-1}{n} \frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i}+1=\frac{n-1}{ \sum_{i=1}^n ln X_i-lnX_{min}}+1 ν^good=nn1n1i=1nYi1+1=i=1nlnXilnXminn1+1.

To find CRLB for ν \nu ν
C R L B ( ν ) = τ ′ ( ν ) 2 n E [ ∂ ∂ ν l n f ( x ; ν ) ] 2 τ ( ν ) = ν p k = ν − 1 k m i n ( k k m i n ) − ν l n f ( x ; ν ) = l n ( ν − 1 ) − ( ν − 1 ) l n x m i n − ν l n x ∂ ∂ ν l n f ( x ; ν ) = 1 ν − 1 − l n x m i n − l n x C R L B ( ν ) = 1 n E [ 1 ν − 1 − Y ] 2 E [ 1 ν − 1 − Y ] 2 = 1 ( ν − 1 ) 2 − 2 ν − 1 1 ν − 1 + V a r [ y 2 ] + E [ y ] 2 = 1 ( ν − 1 ) 2 ∴ C R L B ( ν ) = ( ν − 1 ) 2 n CRLB(\nu)=\frac{\tau'(\nu)^2}{nE[\frac{\partial }{\partial \nu}ln f(x;\nu)]^2}\\ \tau(\nu)=\nu\\ p_k=\frac{\nu-1}{k_{min}}(\frac{k}{k_{min}})^{-\nu}\\ ln f(x;\nu)=ln(\nu-1)-(\nu-1)lnx_{min}-\nu lnx\\ \frac{\partial }{\partial \nu}ln f(x;\nu)=\frac{1}{\nu-1}-lnx_{min}-lnx\\ CRLB(\nu)=\frac{1}{nE[\frac{1}{\nu-1}-Y]^2}\\ E[\frac{1}{\nu-1}-Y]^2=\frac{1}{(\nu-1)^2}-\frac{2}{\nu-1}\frac{1}{\nu-1}+Var[y^2]+E[y]^2\\ =\frac{1}{(\nu-1)^2}\\ \therefore CRLB(\nu)=\frac{(\nu-1)^2}{n}\\ CRLB(ν)=nE[νlnf(x;ν)]2τ(ν)2τ(ν)=νpk=kminν1(kmink)νlnf(x;ν)=ln(ν1)(ν1)lnxminνlnxνlnf(x;ν)=ν11lnxminlnxCRLB(ν)=nE[ν11Y]21E[ν11Y]2=(ν1)21ν12ν11+Var[y2]+E[y]2=(ν1)21CRLB(ν)=n(ν1)2

E [ ν ^ M L E ] = E [ 1 + 1 1 n ∑ i = 1 n Y i ] = E [ n z ] + 1 = n ( ν − 1 ) n − 1 + 1 E [ ν ^ M L E 2 ] = E [ 1 + 2 1 n ∑ i = 1 n Y i + ( 1 1 n ∑ i = 1 n Y i ) 2 ] = 1 + 2 n ( ν − 1 ) n − 1 + n 2 ∫ 0 ∞ 1 z 2 1 Γ ( n ) ( ν − 1 ) n z n − 1 e − ( ν − 1 ) z d z = 1 + 2 n ( ν − 1 ) n − 1 + n 2 ( ν − 1 ) 2 ( n − 1 ) ( n − 2 ) V a r ( ν ^ M L E 2 ) = E [ ν ^ M L E 2 ] − E [ ν ^ M L E ] 2 = 1 + 2 n ( ν − 1 ) n − 1 + n 2 ( ν − 1 ) 2 ( n − 1 ) ( n − 2 ) − ( n ( ν − 1 ) n − 1 + 1 ) 2 = n 2 ( ν − 1 ) 2 ( n − 1 ) 2 ( n − 2 ) ∴ C R L B ( ν ) = ( ν − 1 ) 2 n ≤ n 2 ( ν − 1 ) 2 ( n − 1 ) 2 ( n − 2 ) = V a r ( ν ^ M L E 2 ) E[\hat{\nu}_{MLE}]=E[1+\frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i}]=E[\frac{n}{z}]+1\\ =\frac{n(\nu-1)}{n-1}+1\\ E[\hat{\nu}^2_{MLE}]=E[1+\frac{2}{\frac{1}{n}\sum_{i=1}^n Y_i}+(\frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i})^2]\\ =1+\frac{2n(\nu-1)}{n-1}+n^2\int_0^\infin \frac{1}{z^2}\frac{1}{\Gamma(n)}(\nu-1)^nz^{n-1}e^{-(\nu-1)z}dz\\ =1+\frac{2n(\nu-1)}{n-1}+\frac{n^2(\nu-1)^2}{(n-1)(n-2)}\\ Var(\hat{\nu}^2_{MLE})=E[\hat{\nu}^2_{MLE}]-E[\hat{\nu}_{MLE}]^2\\ =1+\frac{2n(\nu-1)}{n-1}+\frac{n^2(\nu-1)^2}{(n-1)(n-2)}-(\frac{n(\nu-1)}{n-1}+1)^2\\ =\frac{n^2(\nu-1)^2}{(n-1)^2(n-2)}\\ \therefore CRLB(\nu)=\frac{(\nu-1)^2}{n} \leq \frac{n^2(\nu-1)^2}{(n-1)^2(n-2)}=Var(\hat{\nu}^2_{MLE}) E[ν^MLE]=E[1+n1i=1nYi1]=E[zn]+1=n1n(ν1)+1E[ν^MLE2]=E[1+n1i=1nYi2+(n1i=1nYi1)2]=1+n12n(ν1)+n20z21Γ(n)1(ν1)nzn1e(ν1)zdz=1+n12n(ν1)+(n1)(n2)n2(ν1)2Var(ν^MLE2)=E[ν^MLE2]E[ν^MLE]2=1+n12n(ν1)+(n1)(n2)n2(ν1)2(n1n(ν1)+1)2=(n1)2(n2)n2(ν1)2CRLB(ν)=n(ν1)2(n1)2(n2)n2(ν1)2=Var(ν^MLE2)

V a r ( ν ^ U M V U E ) = V a r ( n − 1 n 1 1 n ∑ i = 1 n Y i + 1 ) = ( n − 1 ) 2 n 2 V a r ( ν ^ M L E ) = ( n − 1 ) 2 n 2 ∗ n 2 ( ν − 1 ) 2 ( n − 1 ) 2 ( n − 2 ) = ( ν − 1 ) 2 n − 2 ∴ C R L B ( ν ) = ( ν − 1 ) 2 n ≤ ( ν − 1 ) 2 n − 2 = V a r ( ν ^ U M V U E 2 ) Var(\hat{\nu}_{UMVUE})=Var(\frac{n-1}{n} \frac{1}{\frac{1}{n}\sum_{i=1}^n Y_i}+1)\\ =\frac{(n-1)^2}{n^2}Var(\hat{\nu}_{MLE})\\ =\frac{(n-1)^2}{n^2}*\frac{n^2(\nu-1)^2}{(n-1)^2(n-2)}\\ =\frac{(\nu-1)^2}{n-2}\\ \therefore CRLB(\nu)=\frac{(\nu-1)^2}{n} \leq \frac{(\nu-1)^2}{n-2}=Var(\hat{\nu}^2_{UMVUE}) Var(ν^UMVUE)=Var(nn1n1i=1nYi1+1)=n2(n1)2Var(ν^MLE)=n2(n1)2(n1)2(n2)n2(ν1)2=n2(ν1)2CRLB(ν)=n(ν1)2n2(ν1)2=Var(ν^UMVUE2)

To estimate ν \nu ν by confidence interval
Y = l n x x m i n w h i c h   f o l l o w s   e x p o n e n t i a l   d i s t r b u t i o n 100 ( 1 − α ) % = P ( Γ α 2 ( n , ν − 1 ) ≤ Y ≤ Γ 1 − α 2 ( n , ν − 1 ) ) = P [ n Γ 1 − α 2 ( n , ν − 1 ) + 1 ≤ ν ^ M L E ≤ n Γ α 2 ( n , ν − 1 ) + 1 ] C I = ( n Γ 1 − α 2 ( n , ν − 1 ) + 1 , n Γ α 2 ( n , ν − 1 ) + 1 ) Y=ln\frac{x}{x_{min}}which~follows ~exponential~distrbution\\ 100(1-\alpha)\% =P(\Gamma_{\frac{\alpha}{2}}(n,\nu-1)\leq Y\leq \Gamma_{1-\frac{\alpha}{2}}(n,\nu-1))\\ =P[\frac{n}{\Gamma_{1-\frac{\alpha}{2}}(n,\nu-1)}+1\leq \hat{\nu}_{MLE}\leq \frac{n}{\Gamma_{\frac{\alpha}{2}}(n,\nu-1)}+1]\\ CI=(\frac{n}{\Gamma_{1-\frac{\alpha}{2}}(n,\nu-1)}+1,\frac{n}{\Gamma_{\frac{\alpha}{2}}(n,\nu-1)}+1)\\ Y=lnxminxwhich follows exponential distrbution100(1α)%=P(Γ2α(n,ν1)YΓ12α(n,ν1))=P[Γ12α(n,ν1)n+1ν^MLEΓ2α(n,ν1)n+1]CI=(Γ12α(n,ν1)n+1,Γ2α(n,ν1)n+1)
According to CLT
ν ^ M L E = 1 + n [ ∑ i = 1 n l n x i x m i n ] − 1 ∑ i = 1 n X i n = e n ν ^ − 1 ∑ i = 1 X m i n n X ‾ − e n ν ^ − 1 ∑ i = 1 X m i n n S / n ∼ t ( n − 1 ) 100 ( 1 − α ) % = P ( − t 1 − α 2 ( n − 1 ) ≤ X ‾ − e n ν ^ − 1 ∑ i = 1 X m i n n S / n ≤ t 1 − α 2 ( n − 1 ) ) = P ( 1 + n l n X ‾ + t 1 − α 2 ( n − 1 ) S n ∑ i = 1 n X m i n n ≤ ν ^ ≤ 1 + n l n X ‾ − t 1 − α 2 ( n − 1 ) S n ∑ i = 1 n X m i n n ) C I = ( 1 + n l n X ‾ + t 1 − α 2 ( n − 1 ) S n ∑ i = 1 n X m i n n , 1 + n l n X ‾ − t 1 − α 2 ( n − 1 ) S n ∑ i = 1 n X m i n n ) \hat{\nu}_{MLE}=1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1}\\ \frac{\sum_{i=1}^nX_i}{n}=e^{\frac{n}{\hat{\nu}-1}}\frac{\sum_{i=1}X_{min}}{n}\\ \frac{\overline{X}-e^{\frac{n}{\hat{\nu}-1}}\frac{\sum_{i=1}X_{min}}{n}}{S/\sqrt{n}} \sim t(n-1)\\ 100(1-\alpha)\%=P(-t_{1-\frac{\alpha}{2}}(n-1)\leq \frac{\overline{X}-e^{\frac{n}{\hat{\nu}-1}}\frac{\sum_{i=1}X_{min}}{n}}{S/\sqrt{n}} \leq t_{1-\frac{\alpha}{2}}(n-1))\\ =P(1+\frac{n}{ln\frac{\overline{X}+t_{1-\frac{\alpha}{2}}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}} \leq \hat{\nu} \leq 1+\frac{n}{ln\frac{\overline{X}-t_{1-\frac{\alpha}{2}}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}})\\ CI=(1+\frac{n}{ln\frac{\overline{X}+t_{1-\frac{\alpha}{2}}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}}, 1+\frac{n}{ln\frac{\overline{X}-t_{1-\frac{\alpha}{2}}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}})\\ ν^MLE=1+n[i=1nlnxminxi]1ni=1nXi=eν^1nni=1XminS/n Xeν^1nni=1Xmint(n1)100(1α)%=P(t12α(n1)S/n Xeν^1nni=1Xmint12α(n1))=P(1+lnni=1nXminX+t12α(n1)n Snν^1+lnni=1nXminXt12α(n1)n Sn)CI=(1+lnni=1nXminX+t12α(n1)n Sn,1+lnni=1nXminXt12α(n1)n Sn)
For hypothesis test, test H 0 : ν > ν 0 , H 1 : ν < ν 0 H_0: \nu>\nu_0, H_1:\nu<\nu_0 H0:ν>ν0,H1:ν<ν0 at significant level α \alpha α.
α = P [ 1 + n [ ∑ i = 1 n l n x i x m i n ] − 1 ≥ ν 0 ] = P [ ∑ i = 1 n l n x x m i n ≤ n ν 0 − 1 ] = P [ Γ 1 − α ( n , ν − 1 ) ≤ n ν 0 − 1 ] I f   Γ 1 − α ( n , ν ^ − 1 ) ≤ n ν 0 − 1 , r e j e c t   H 0 β = P ( T y p e I I ) = P ( a c c e p t   H 0 ∣ H 1 ) = P [ Γ 1 − α ( n , ν ^ − 1 ) > n ν 0 − 1 ∣ ν < ν 0 ] \alpha = P[1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1}\geq \nu_0]\\ =P[\sum_{i=1}^nln\frac{x}{x_{min}}\leq \frac{n}{\nu_0-1}]\\ =P[\Gamma_{1-\alpha}(n,\nu-1)\leq \frac{n}{\nu_0-1}]\\ If~ \Gamma_{1-\alpha}(n,\hat{\nu}-1)\leq \frac{n}{\nu_0-1},reject~H_0\\ \beta=P(Type II)=P(accept~ H_0|H_1)\\ =P[\Gamma_{1-\alpha}(n,\hat{\nu}-1)> \frac{n}{\nu_0-1}|\nu<\nu_0]\\ α=P[1+n[i=1nlnxminxi]1ν0]=P[i=1nlnxminxν01n]=P[Γ1α(n,ν1)ν01n]If Γ1α(n,ν^1)ν01n,reject H0β=P(TypeII)=P(accept H0H1)=P[Γ1α(n,ν^1)>ν01nν<ν0]
Similarly, according to CLT
α = P ( X ‾ − e n ν ^ − 1 ∑ i = 1 X m i n n S / n ≥ t 1 − α ( n − 1 ) ) = P [ ν ≥ 1 + n l n X ‾ − t 1 − α ( n − 1 ) S n ∑ i = 1 n X m i n n ) ] I f   ν ^ ≥ 1 + n l n X ‾ − t 1 − α ( n − 1 ) S n ∑ i = 1 n X m i n n , r e j e c t   H 0 \alpha=P( \frac{\overline{X}-e^{\frac{n}{\hat{\nu}-1}}\frac{\sum_{i=1}X_{min}}{n}}{S/\sqrt{n}} \geq t_{1-\alpha}(n-1))\\ =P[\nu \geq 1+\frac{n}{ln\frac{\overline{X}-t_{1-\alpha}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}})]\\ If~\hat{\nu} \geq 1+\frac{n}{ln\frac{\overline{X}-t_{1-\alpha}(n-1)\frac{S}{\sqrt{n}}}{\frac{\sum_{i=1}^nX_{min}}{n}}},reject ~H_0 α=P(S/n Xeν^1nni=1Xmint1α(n1))=P[ν1+lnni=1nXminXt1α(n1)n Sn)]If ν^1+lnni=1nXminXt1α(n1)n Sn,reject H0

estimation methodfunctionunbiasednessconsistencyeffectiveness
measurementcalculate by function Fcalculate the E(F)calculate E(F) given n → ∞ n\rightarrow \infin ncompared with C R L B ( ν ) = ( ν − 1 ) 2 n CRLB(\nu)=\frac{(\nu-1)^2}{n} CRLB(ν)=n(ν1)2
MME − 2 ( ∑ x i ) 2 + 2 n ∑ x i 2 ± − n ( ∑ x i ) 2 ∑ x i 2 + n 2 ( ∑ x i 2 ) 2 − ( ∑ x i ) 2 + n ∑ x i 2 \frac{-2(\sum x_i)^2+2n\sum x_i^2 \pm \sqrt{-n(\sum x_i)^2 \sum x_i^2+n^2 (\sum x_i^2)^2}}{-(\sum x_i)^2+n\sum x_i^2} (xi)2+nxi22(xi)2+2nxi2±n(xi)2xi2+n2(xi2)2 biased E [ ν ^ M M E ] → 3   o r   1 E[\hat{\nu}_{MME}]\rightarrow 3~or~1 E[ν^MME]3 or 1bigger than ( ν − 1 ) 2 n \frac{(\nu-1)^2}{n} n(ν1)2
MLE 1 + n [ ∑ i = 1 n l n x i x m i n ] − 1 1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1} 1+n[i=1nlnxminxi]1 E [ ν ^ M L E ] = n ν n − 1 − 1 n − 1 E[\hat{\nu}_{MLE}]=\frac{n\nu}{n-1}-\frac{1}{n-1} E[ν^MLE]=n1nνn11 E [ ν ^ M L E ] → ν E[\hat{\nu}_{MLE}]\rightarrow \nu E[ν^MLE]ν n 2 ( ν − 1 ) 2 ( n − 1 ) 2 ( n − 2 ) ≥ ( ν − 1 ) 2 n \frac{n^2(\nu-1)^2}{(n-1)^2(n-2)}\geq \frac{(\nu-1)^2}{n} (n1)2(n2)n2(ν1)2n(ν1)2
unbised estimator 1 + n − 1 ∑ i = 1 n l n X i − l n X m i n 1+\frac{n-1}{ \sum_{i=1}^n ln X_i-lnX_{min}} 1+i=1nlnXilnXminn1 E [ ν ^ U M V U E ] = ν E[\hat{\nu}_{UMVUE}]=\nu E[ν^UMVUE]=ν E [ ν ^ U M V U E ] → ν E[\hat{\nu}_{UMVUE}]\rightarrow \nu E[ν^UMVUE]ν ( ν − 1 ) 2 n − 2 ≥ ( ν − 1 ) 2 n \frac{(\nu-1)^2}{n-2}\geq \frac{(\nu-1)^2}{n} n2(ν1)2n(ν1)2

其他相关:
Covid19:传染病模型SIR/SEIR的估计与拟合
Covid19:基于网络的传染病模型Virus spread on network
Covid19:传染病模型SIR/SEIR on Network
Covid19:传染病模型模拟Agent based simulation for SIR/SEIR models.

欢迎关注二幺子的知识输出通道:
avatar

已标记关键词 清除标记
相关推荐
©️2020 CSDN 皮肤主题: 数字20 设计师:CSDN官方博客 返回首页