我对以下单方面的Cantelli版本的Chebyshev不等式感兴趣:
基本上,如果您知道总体均值和方差,则可以计算观察到某个值的概率的上限。(至少这是我的理解。)
但是,我想使用样本均值和样本方差,而不是实际总体均值和方差。
我猜想,由于这会带来更多不确定性,因此上限会增加。
是否存在类似于上述的不等式,但是使用样本均值和方差?
编辑:Chebyshev不等式(不是单面)的“样本”类似物,已经制定出来。在维基百科页面有一些细节。但是,我不确定它将如何转化为我上面提到的单面案例。
我对以下单方面的Cantelli版本的Chebyshev不等式感兴趣:
基本上,如果您知道总体均值和方差,则可以计算观察到某个值的概率的上限。(至少这是我的理解。)
但是,我想使用样本均值和样本方差,而不是实际总体均值和方差。
我猜想,由于这会带来更多不确定性,因此上限会增加。
是否存在类似于上述的不等式,但是使用样本均值和方差?
编辑:Chebyshev不等式(不是单面)的“样本”类似物,已经制定出来。在维基百科页面有一些细节。但是,我不确定它将如何转化为我上面提到的单面案例。
Answers:
是的,我们可以使用样本均值和方差得到类似的结果,也许在此过程中会出现一些小小的意外。
首先,我们需要对问题陈述进行一些细化,并提出一些假设。重要的是,很明显,我们不能希望用右侧的样本方差代替总体方差,因为后者是随机的!所以,我们重新调整我们的注意力就相当于不等式
第二,我们假设我们有一个随机样本和我们感兴趣的一个上界的类似量 ,其中是样本均值和是样品标准偏差。
前进半步
请注意,已经通过将原来片面切比雪夫不等式,我们得到的是 ,其中σ2=V一- [R (X1),这是较小的比的原始版本的右手侧。这很有道理!来自样本的随机变量的任何特定实现都倾向于(略微)接近其所贡献的样本均值,而不是总体均值。正如我们将在下面看到的,在更一般的假设下,我们将用S替换σ。
单面切比雪夫的示例版本
要求:令是一个随机样本,使得P(S = 0 )= 0。然后,P(X 1 - ˉ X ≥ 吨小号)≤ 1特别是,边界的样本版本比原始人口版本更严格。
注意:我们不假定具有有限的均值或方差!
证明。这个想法是适应原始的单边切比雪夫不等式的证明,并在此过程中采用对称性。首先,设置为标记方便。然后,观察到 P(Ý 1 ≥ 吨小号)= 1
现在,对于任何,在{ 小号> 0 }, 1 (ÿ 我 ≥ 吨小号) = 1 (ÿ 我 + 吨Ç 小号≥ 吨小号(1 + c ^ )) ≤ 1 ((ÿ 我 + 吨Ç 小号)2 ≥ 吨2(1 + c ^ )2 š 2
然后
That pesky technical condition
Note that we had to assume in order to be able to divide by in the analysis. This is no problem for absolutely continuous distributions, but poses an inconvenience for discrete ones. For a discrete distribution, there is some probability that all observations are equal, in which case for all and .
We can wiggle our way out by setting . Then, a careful accounting of the argument shows that everything goes through virtually unchanged and we get
Corollary 1. For the case , we have
Proof. Split on the events and . The previous proof goes through for and the case is trivial.
A slightly cleaner inequality results if we replace the nonstrict inequality in the probability statement with a strict version.
Corollary 2. Let (possibly zero). Then,
Final remark: The sample version of the inequality required no assumptions on (other than that it not be almost-surely constant in the nonstrict inequality case, which the original version also tacitly assumes), in essence, because the sample mean and sample variance always exist whether or not their population analogs do.
This is just a complement to @cardinal 's ingenious answer. Samuelson Inequality, states that, for a sample of size , when we have at least three distinct values of the realized 's, it holds that
Then, using the notation of Cardinal's answer we can state that
Since we require, three distinct values, we will have by assumption. So setting in Cardinal's Inequality (the initial version) we obtain
Eq. is of course compatible with eq. . The combination of the two tells us that Cardinal's Inequality is useful as a probabilistic statement for .
If Cardinal's Inequality requires to be calculated bias-corrected (call this ) then the equations become
and we choose to obtain through Cardinal's Inequality