가중 산술 평균

Weighted arithmetic mean

가중 산술 평균은 일반 산술 평균(가장 일반적인 유형의 평균)과 유사하지만 각 데이터 지점이 최종 평균에 동일하게 기여하는 대신 일부 데이터 지점이 다른 데이터 지점보다 더 많이 기여합니다.가중평균의 개념은 기술 통계학에서 중요한 역할을 하며 수학의 여러 다른 영역에서도 보다 일반적인 형태로 발생한다.

모든 가중치가 같으면 가중 평균은 산술 평균과 동일합니다.가중평균은 일반적으로 산술평균과 유사한 방식으로 동작하지만 심슨의 역설에서 포착된 것과 같이 몇 가지 반직관적인 특성을 가지고 있습니다.

기본적인 예

다음과 같은 2개의 학교 수업(하나는 학생 20명, 1개는 학생 30명)과 각 반의 시험 성적:

오전반 = {62, 67, 71, 74, 76, 77, 78, 79, 79, 80, 80, 81, 82, 83, 84, 86, 89, 93, 98}

오후 클래스 = {81, 82, 83, 84, 85, 86, 87, 88, 89, 89, 89, 90, 90, 91, 92, 93, 93, 93, 94, 95, 96, 97, 98, 99}

오전 수업의 평균은 80이고 오후 수업의 평균은 90입니다.두 평균의 가중되지 않은 평균은 85입니다.그러나, 이것은 각 반의 학생 수(20 대 30)의 차이를 설명하지 않기 때문에, 85의 값은 (반과 무관한) 평균 학생 성적을 반영하지 않습니다.평균 학생 성적은 학급에 관계없이 모든 성적을 평균하여 얻을 수 있습니다(모든 성적을 합산하여 총 학생 수로 나눕니다).

또는 각 반의 학생 수에 따라 학급 수단에 가중치를 부여함으로써 이를 달성할 수 있다.클래스가 클수록 "가중치"가 높아집니다.

따라서 가중평균을 사용하면 각 학생의 점수를 몰라도 평균 학생 성적을 찾을 수 있습니다.각 반의 학급별 평균과 학생 수만 있으면 된다.

볼록 조합의 예

상대적인 가중치만 관련되므로 가중평균은 모두 1에 해당하는 계수를 사용하여 표현할 수 있습니다.이러한 선형 결합을 볼록 결합이라고 합니다.

위의 예에서는 다음과 같은 가중치를 얻을 수 있습니다.

그런 다음 다음과 같이 무게를 가합니다.

수학적 정의

형식적으로는 빈( x2, n가 아닌 유한 태플의 가중평균(x 1, x 2, …, n displaystyle 1}, },\에 대응하는 아닌 1, 2, n

다음과 같이 확장됩니다.

따라서 가중치가 높은 데이터 요소는 가중치가 낮은 요소보다 가중 평균에 더 많이 기여합니다.가중치는 음수일 수 없습니다.일부는 0일 수 있지만, 모두 0인 것은 아닙니다(0으로 나눌 수 없기 때문입니다).

공식은 가중치가 최대 1(즉, i i { _ w_{i}'=까지 합이 되도록 정규화될 때 단순화된다. 정규화된 가중치의 경우 가중 평균은 다음과 같다.

_

원래 가중치에 대해 다음과 같은 변환을 수행하면 가중치를 항상 정규화할 수 있습니다.

i { w { i } = sum { w { w _ { i } } { \ \ _ { j=}^{ n } { _ { j }}

일반 x {\1}{n \}{i}}}는 모든 데이터가 동일한 가중 평균의 특수한 경우이다.

데이터 요소가 독립적이고 이 2동일한 랜덤 변수일 경우 가중평균의 표준 인 "x _는 불확실성 전파를 통해 다음과 같이 표시될 수 있습니다.

분산 정의 가중치

x {\i}}이(가) 알려진 분산 {\ _인 서로 다른 확률 분포에서 잠재적으로 파생된 데이터 목록의 가중 평균에 대해 모두 동일한 평균을 가지며, 가중치에 대해 하나의 가능한 선택지가 분산 역수에 의해 주어진다.

이 경우 가중평균은 다음과 같습니다.

가중 평균(역평균 가중치)의 표준 오차는 다음과 같다.

0{ \ _ { i } \ { i } = \ displaystyle{ i= \ _ { }^{2} 으로 합니다.이는 이전 절의 일반 공식의 특수한 경우이다.

위의 방정식을 조합하여 다음을 얻을 수 있습니다.

이 선택의 중요성은 이 가중 평균이 확률 분포가 독립적이고 동일한 평균으로 정규 분포를 따른다는 가정 하에 확률 분포 평균의 최대우도 추정기라는 것입니다.

통계 속성

기대치

가중치 샘플 평균 \ style \ \ { )는 그 자체가 랜덤 변수입니다.기대값과 표준 편차는 다음과 같이 관측치의 기대값 및 표준 편차와 관련이 있습니다.단순화를 위해 정규화된 가중치(무게의 합계는 1)를 가정합니다.

관측치에 기대값이 있는 경우

가중 표본 평균에 대한 기대치가 있습니다.
특히 평균이 i μ \ _}=\인 경우 가중 표본 평균의 기대치는 이 값이 됩니다.

분산

단순 신분증 사건

가중치를 상수로 취급하고 상관 관계가 없는 랜덤 변수의 관측치 n개의 표본을 모두 동일한 분산과 기대(예: d 랜덤 변수의 경우)로 갖는 경우 가중 평균의 분산은 Kish의 설계 효과에 의한 분산의 곱으로 추정할 수 있습니다(증명 참조).

i ( - y ) n- { { _ { y }^{ } ={ _ { - { \ { } = 1 n

그러나 y 관측치에 대한 가정이 강하기 때문에 이 추정치는 다소 제한됩니다.이는 보다 일반적인 대체 추정치의 개발로 이어졌다.

설문 표본 추출 관점

모델 기반의 관점에서, 우리는 서로 i})가 i.i.d 랜덤 변수가 아닐 때 가중 평균의 분산을 추정하는 데 관심이 있다.이 문제에 대한 대체적인 관점은 (치환으로)[1]: 306 균등한 확률로 단위를 선택하는 데이터의 임의 샘플링 설계입니다.

조사 방법론에서 모집단 평균은 모집단의 모든 요소(Y 또는 때로는 T)에 대한 총 y의 추정치를 구해서 이를 모집단 크기( \ N 또는 추정치(style 로 나누어 계산한다.이 문맥에서 y의 값은 일정하다고 간주되며, 변동성은 선택 절차에서 비롯됩니다.이는 무작위성이 종종 y 값으로 설명되는 "모델 기반" 접근법과 대조된다.조사 표본 추출 절차에서는 일부 관측치 i가 표본에 있는 경우 1을 얻고 선택하지 않은 경우 0을 얻는 일련의 베르누이 지표 값(i})을 산출합니다.이는 고정된 표본 크기 또는 다양한 표본 크기 샘플링에서 발생할 수 있습니다(예:포아송 표본 추출).시료에서 선택될 확률은 P ) = {} = \ _ = 1 확률은 P) 이다. { { sample draw } = } \ {{ _ { i } N each each each each each ( n ) 。다음 유도에서는 각 요소를 선택할 확률이 이러한 [2]: 42, 43, 51 확률로 완전히 표현된다고 가정합니다.예를 들어, 일부 요소를 선택해도 다른 요소를 그릴 확률에는 영향을 미치지 않습니다(클러스터 샘플링 설계와 같은 경우에는 해당되지 않음).

각 요소(는 고정되어 있고 샘플에 포함되거나 포함되지 않은 것(에서 랜덤성이 나오기 때문에 랜덤 변수인 두 요소의 곱셈에 대해 자주 이야기합니다.다음 섹션의 혼동을 피하기 위해 를 y = i i i i {\{i} }라고 부릅니다. 기대치 [ i ] i[ ] i \ E [ ' { i } =_ { } _ 및 분산: __{

샘플의 각 원소가 선택 확률의 역방향으로 부풀려졌을 때, 예를 y i = y {\ i {\ i i \ displaystyle { } { i } ={ y _ _ _ { i } { \ _ { } {\a the ( \ displaysty values values values values values values values values values values values values values,, values,,,, the the the the the,,,,,,,,,,,,,, the,,, the i y i { { { y { } } { { [2]: 42, 43, 51, 52 。 위와 같이 표시기 함수에 곱하면 체크 마크를 추가할 수 있습니다.예: i = = i {\ i i {\ i i i ( { ) _ { i } =

설계 기반 관점에서 가중 평균의 분자에 사용되는 가중치는 선택 확률의 역수(즉, 팽창 계수)를 취함으로써 구한다.예: i 1 × i { w { i } = frac {\_ { \ 약 { { \ p _ { }

가중 합계의 분산(총계에 대한 pwr-추정자)

모집단 크기 N이 알려진 경우 Y ^ r N n N { style { \ { Y } } \ {N= { w } { wr } { wr } { WR} { } { wr } y N y N N }을 하여 모집단 평균을 추정할 수 있습니다.

표본 추출 설계가 고정된 표본 크기 n(: pps 표본 추출)인 경우 이 추정기의 분산은 다음과 같습니다.

증명

일반적인 공식은 다음과 같이 개발할 수 있습니다.

모집단 총계는 Y i \ Y = \ _ { i=}^{ } _ { } 로 표시되며 (편향되지 않음)으로 추정할 수 있다.호비츠Thompson Estimator라고도 불립니다.이 추정기 자체는 pwr 추정기(예: p - 교체 추정기 또는 "교체 확률" 추정기)를 사용하여 추정할 수 있습니다.의 표기법에서는 ^ r n i 1 n i y n w n w i ′ \ { hat } { { wr } fr = fr with with with with with with with with with with with with with with with 。 \}{\i}}{\{i}}=\[2]: 51 i}}}.

pwr 추정치의 추정 분산은 다음과 같습니다.[2]: 52

서 w i w \ { i1} { \ { { } _ { } { }

위의 공식은 Sarndal et al.(1992)(1977년 Cochran에서도 제시)에서 따온 것이지만, 다르게 [2]: 52 [1]: 307 (11.35) 기술되었다.왼쪽은 분산 작성 방법이고 오른쪽은 가중 버전을 개발한 방법입니다.

그리고 우리는 위에서 공식에 도달했다.

표본이 무작위 표본 크기를 갖는 경우에 대한 대체 용어는 Sarndal 등(1992)에 다음과 [2]: 182 같이 제시된다.

i = i {\}} = {y_{ {\ _ 또한C ( ) = ( __{_{j}=\_{ 여기서 j \ _ i와 [2]: 36 j를 모두 선택할 확률입니다.그리고 i - j i j}} { } - { \ { _ { } }- { { i } }

선택확률이 상관없는 경우(: i : C , \ \i\i}) 및 각 원소의 확률이 매우 작다고 가정할 경우:

프루프

- 0 \ ( 1 - \ _ { } \ 0 00 이라고 가정합니다.

가중 평균의 분산(비율 평균에 대한 δ 추정치)

이전 섹션에서는 모집단 평균을 알려진 모집단 크기(에 대한 추정 모집단 총계( 의 비율로 추정하는 것을 다루었으며, 그 맥락에서 편차는 추정되었다.또 다른 일반적인 경우는 모집단 크기( N 자체를 알 수 없으며 표본(예:N N 추정은 가중치의 합으로 설명할 수 있습니다. w i ( \ style { 1 n = n = 1 n ˇ i 1 n i 1 n i i style _ { 11 n = 1 n ) i i i i { \ flacclacclac { 1 }}}}} n n n { n frac {n frac { n frac { i }}} __{i}}=\_{ 이전 섹션의 표기법을 사용할 때 유의하는 비율은입니다.예: y i N = = 1 N = 1 wi = i = 1 n wi = i = 1 n i = 1 n wi = 1 n wi = i = i= i = i {\ i {\ i {\ i {\ {\ i {\ i {\ i {\ i {\ i {\ i {\ i {\ i = i {\ i frac { _i}}= frac _}^{}{i}}}{i}}}}= frac frac {i}{i}{n}{n}{n}{n}{n}{n}{n}}}{n}{nn}{n _{{}}{\ _1}}= frac }{n}{ni}}{_i}}}{i}}}}}}}}}}{i_y}}}}}}}}}}{i}}}}}o 간단하게 쓸 수 있습니다. y 1 = w 1 n { style { { w } 1 n{ _ { i=}^{}_ { i } { }{ _ { i }이 값은 y와 w의 특정 값에 대한 추정치가 되지만 통계 속성은 변수 y w 1 i n i= 1 { frac { i bar {} { { w } { w _ { w } { w _ }} { sum { sum } { i } { sum _ sum } }

이를 비율 추정기라고 하며 [2]: 182 R에 대해 거의 치우치지 않습니다.

이 경우 비율의 변동성은 분자와 분모에서 랜덤 변수의 변동성과 상관 관계에 따라 달라집니다.이 분산을 계산할 수 있는 닫힌 분석 형식이 없기 때문에 다양한 방법이 대략적인 추정에 사용됩니다.주로 Taylor 시리즈의 1차 선형화, 점근성 및 부트스트랩/[2]: 172 잭나이프입니다.Taylor 선형화 방법은 일반적으로 작은 표본 크기에 대한 분산을 과소 추정할 수 있지만 통계량의 복잡도에 따라 달라집니다.가중 평균의 경우 표본 크기가 [2]: 176 중간인 경우에도 근사 분산이 비교적 정확해야 합니다.표본 추출이 랜덤 표본 크기를 갖는 경우(포아송 표본 추출에서와 같이), 다음과 같습니다.[2]: 182

2}\}^2w})^22

that i \ \ _ { i} \ approx _ { } w {\ {\ {\{\ } { \ _ { w _ { w _ { i } w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w배우도 같은 추정치를 얻을 수 있습니다.또한 가중치의 합계를 이전부터 알려진 모집단 크기 N과 동일하게 척도화하면 분산 계산이 같아 보인다는 의미도 됩니다.모든 가중치가 서로 같으면 이 공식은 표준 비편향 분산 추정기로 감소합니다.

★★★

Taylor 선형화에서는 2개의 합계(^ ^ { hat { R}} ={ {의 일반 비율 추정치에 대해 R을 중심으로 확장할 수 있으며 다음과 같은 값을 [2]: 178 얻을 수 있습니다.

분산은 다음과 [2]: 178, 179 같이 근사할 수 있습니다.

C^ (^, Z ) ( \ { \ hat { C } ( { \ { } , { \ { } ) )는 Y의 추정합과 Z의 추정합 사이의 공분산 추정치이다.이 값은 두 랜덤 변수 합계의 공분산이므로 지시 변수에 따라 달라지는 공분산 조합을 많이 포함합니다.선택확률이 상관없는 경우(: ij : i ( I, ) \ i j _ =}) 이 용어는 i y y { i } 의 각 요소 i에 대한 n개의 공분산 합계를 포함합니다. ii i i { i }이 공식은 비율 추정기의 분산에 대한 y와 z 사이의 상관 관계 효과를 통합하는 데 도움이 됩니다.

i {}=(를) 정의하면 위와 같이 됩니다.[2]: 182

선택확률이 상관없는 경우(: ij : i ( I, ) \ i j _ =}) 및 각 요소의 확률이 매우 작다고 가정할 때(즉 ( - i ) \ ( - \ _ { } \ 0 ) then then then 、 상기의 값을 다음과 같이 줄인다.

(마지막에 몇 가지 실수까지) 유사한 증명 재작성이 교차 검증된 [3]Thomas Lumley에 의해 제공되었습니다.

가중 평균에 대한 분산 버전(적어도)이 두 개 있다. 하나는 알려진 버전이고 다른 하나는 알려지지 않은 모집단 크기 추정이다.균등하게 더 나은 접근법은 없지만, 문헌은 모집단 추정 버전을 사용하는 것을 선호한다는 몇 가지 주장을 제시한다(인구 크기가 [2]: 188 알려진 경우에도).예를 들어, 모든 y 값이 일정하면 모집단 크기를 알 수 없는 추정치는 올바른 결과를 제공하는 반면 모집단 크기를 알 수 있는 추정치는 변동성이 있습니다.또한 표본 크기 자체가 랜덤인 경우(: 포아송 표본 추출에서), 모집단 평균을 알 수 없는 버전이 더 안정적인 것으로 간주됩니다.마지막으로 표본 추출의 비율이 값과 음의 상관 관계(즉, 큰 관측치를 표본 추출할 수 있는 더 적은 기회)인 경우, 알려지지 않은 모집단 크기 버전이 이를 약간 보상한다.

Gatz et al.(1995)에 의해 부트스트래핑 방법과 비교하여 다음(Taylor 직렬 선형화를 사용한 비율 평균의 분산 추정)이 평균의 표준 오차 제곱에 대한 합리적인 추정이라는 것이 밝혀졌다(화학 성분 측정의 맥락에서 사용될 경우).[4]: 1186

서 w { {} ={_ { } }{ n} 。좀 더 심플하게 하면,

Gatz 등은 위의 공식은 Cochran(1977년)이 발표한 공식에 기초하여 가중 평균을 모집단 [5]크기의 추정기로 나눈 가중 총 추정기의 조합으로 취급할 때 Endlich 등(1988)에 의해 발표되었다고 언급했다.그러나 Endlich 등은 (그들이 그것을 사용했다고 언급했음에도 불구하고) 논문에 이 파생물을 발표하지 않은 것으로 보이며, Cochran의 책에는 약간 다른 [1]: 155 공식이 포함되어 있다.그러나 이는 이전 섹션에서 설명한 공식과 거의 동일합니다.

기반

가중 평균의 분산에 대한 폐쇄적 분석 형식이 없기 때문에, JackknifeBootstraping[1]: 321 같은 복제 방법에 의존하는 것이 문헌에서 제안되었다.

분산이 i {\ _인 상관 관계가 없는 관측치의 경우 가중 표본 평균의[citation needed] 분산은 다음과 같다.

제곱근 x \ \ _ { \ { } }는 가중평균의 표준오차(일반적인 경우)[citation needed]라고 할 수 있습니다.

따라서 모든 관측치가 동일한 분산인 경우, i 2 \ \ { }^{2}= \ _ { i가중 표본 평균은 분산이 된다.

1/ 1 w 2 1 1 {1/n\ _ {i =} { 을 제외한 무게가 0일 때 분산은 최대값 0 에 도달합니다최소값은 모든 가중치가 같을 때(즉, 가중되지 않은 평균) 발견되며, 이 경우 x 0 / x}=\ 즉, 평균 제곱의 표준 오차로 퇴보한다.

항상 정규화되지 않은 가중치를 정규화된 가중치로 변환할 수 있으므로, 이 섹션의 모든 공식은 w i 1 i { } = {\_ {}^{} 를 대체하여 정규화되지 않은 가중치에 적용할 수 있다.

관련 개념

가중 표본 분산

일반적으로 평균을 계산할 때는 해당 평균에 대한 분산표준 편차를 아는 것이 중요합니다.µ {\^{*}}}을 사용하는 경우 가중시료의 분산은 가중시료의 분산과 다르다.

바이어스 가중치 샘플 분산 ^ 2\ \ \ sigma_ { \ { w2}는 바이어스 샘플 분산 ^ \ \ sigma }^과 유사하게 정의됩니다.

여기서, 정규화된 가중치의 경우 1 \ _}=가중치가 주파수 가중치(따라서 랜덤 변수)인 경우,^ {\_ {w id 가우스 관측에 §의 최대우도 추정치임을 알 수[citation needed] 있다.

작은 표본의 경우 모집단 분산에 대해 치우치지 않은 추정기를 사용하는 것이 일반적입니다.정상 비가중 표본의 경우 분모의 N(표본 크기에 해당)이 N - 1로 변경됩니다(베셀 보정 참조).가중치 설정에서는 주파수 가중치의 경우 신뢰성 가중치의 경우 두 가지 다른 편향되지 않은 추정기가 실제로 있습니다.

주파수 가중치

가중치가 빈도 가중치(가중치가 발생 횟수와 동일)인 경우 편향되지 않은 추정치는 다음과 같습니다.

이는 주파수 가중치에 대한 베셀의 보정을 효과적으로 적용한다.

예를 들어 값{2, ,,, 5, 같은 분포에서 추출된 경우 이 세트를 가중치 없는 샘플로 취급할 수도 있고 대응하는가중치의 가중치 샘플{ 4 취급할 수도 있습니다어느 쪽이든 같은 결과를 얻을 수 있습니다.

If the frequency weights are normalized to 1, then the correct expression after Bessel's correction becomes

where the total number of samples is (not ). In any case, the information on total number of samples is necessary in order to obtain an unbiased correction, even if has a different meaning other than frequency weight.

Note that the estimator can be unbiased only if the weights are not standardized nor normalized, these processes changing the data's mean and variance and thus leading to a loss of the base rate (the population count, which is a requirement for Bessel's correction).

Reliability weights

If the weights are instead non-random (reliability weights[definition needed]), we can determine a correction factor to yield an unbiased estimator. Assuming each random variable is sampled from the same distribution with mean and actual variance , taking expectations we have,

where and . Therefore, the bias in our estimator is , analogous to the bias in the unweighted estimator (also notice that is the effective sample size). This means that to unbias our estimator we need to pre-divide by , ensuring that the expected value of the estimated variance equals the actual variance of the sampling distribution.

The final unbiased estimate of sample variance is:

[6]

where .

The degrees of freedom of the weighted, unbiased sample variance vary accordingly from N − 1 down to 0.

The standard deviation is simply the square root of the variance above.

As a side note, other approaches have been described to compute the weighted sample variance.[7]

Weighted sample covariance

In a weighted sample, each row vector (each set of single observations on each of the K random variables) is assigned a weight .

Then the weighted mean vector is given by

And the weighted covariance matrix is given by:[8]

Similarly to weighted sample variance, there are two different unbiased estimators depending on the type of the weights.

Frequency weights

If the weights are frequency weights, the unbiased weighted estimate of the covariance matrix , with Bessel's correction, is given by:[8]

Note that this estimator can be unbiased only if the weights are not standardized nor normalized, these processes changing the data's mean and variance and thus leading to a loss of the base rate (the population count, which is a requirement for Bessel's correction).

Reliability weights

In the case of reliability weights, the weights are normalized:

(If they are not, divide the weights by their sum to normalize prior to calculating :

Then the weighted mean vector can be simplified to

and the unbiased weighted estimate of the covariance matrix is:[9]

The reasoning here is the same as in the previous section.

Since we are assuming the weights are normalized, then and this reduces to:

If all weights are the same, i.e. , then the weighted mean and covariance reduce to the unweighted sample mean and covariance above.

Vector-valued estimates

The above generalizes easily to the case of taking the mean of vector-valued estimates. For example, estimates of position on a plane may have less certainty in one direction than another. As in the scalar case, the weighted mean of multiple estimates can provide a maximum likelihood estimate. We simply replace the variance by the covariance matrix and the arithmetic inverse by the matrix inverse (both denoted in the same way, via superscripts); the weight matrix then reads:[10]

The weighted mean in this case is:

(where the order of the matrix–vector product is not commutative), in terms of the covariance of the weighted mean:

For example, consider the weighted mean of the point [1 0] with high variance in the second component and [0 1] with high variance in the first component. Then

then the weighted mean is:

which makes sense: the [1 0] estimate is "compliant" in the second component and the [0 1] estimate is compliant in the first component, so the weighted mean is nearly [1 1].

Accounting for correlations

In the general case, suppose that , is the covariance matrix relating the quantities , is the common mean to be estimated, and is a design matrix equal to a vector of ones (of length ). The Gauss–Markov theorem states that the estimate of the mean having minimum variance is given by:

and

where:

Decreasing strength of interactions

Consider the time series of an independent variable and a dependent variable , with observations sampled at discrete times . In many common situations, the value of at time depends not only on but also on its past values. Commonly, the strength of this dependence decreases as the separation of observations in time increases. To model this situation, one may replace the independent variable by its sliding mean for a window size .

Exponentially decreasing weights

In the scenario described in the previous section, most frequently the decrease in interaction strength obeys a negative exponential law. If the observations are sampled at equidistant times, then exponential decrease is equivalent to decrease by a constant fraction at each time step. Setting we can define normalized weights by

where is the sum of the unnormalized weights. In this case is simply

approaching for large values of .

The damping constant must correspond to the actual decrease of interaction strength. If this cannot be determined from theoretical considerations, then the following properties of exponentially decreasing weights are useful in making a suitable choice: at step , the weight approximately equals , the tail area the value , the head area . The tail area at step is . Where primarily the closest observations matter and the effect of the remaining observations can be ignored safely, then choose such that the tail area is sufficiently small.

Weighted averages of functions

The concept of weighted average can be extended to functions.[11] Weighted averages of functions play an important role in the systems of weighted differential and integral calculus.[12]

Correcting for over- or under-dispersion

Weighted means are typically used to find the weighted mean of historical data, rather than theoretically generated data. In this case, there will be some error in the variance of each data point. Typically experimental errors may be underestimated due to the experimenter not taking into account all sources of error in calculating the variance of each data point. In this event, the variance in the weighted mean must be corrected to account for the fact that is too large. The correction that must be made is

where is the reduced chi-squared:

The square root can be called the standard error of the weighted mean (variance weights, scale corrected).

When all data variances are equal, , they cancel out in the weighted mean variance, , which again reduces to the standard error of the mean (squared), , formulated in terms of the sample standard deviation (squared),

See also

References

  1. ^ a b c d Cochran, W. G. (1977). Sampling Techniques (3rd ed.). Nashville, TN: John Wiley & Sons. ISBN978-0-471-16240-7
  2. ^ a b c d e f g h i j k l m n o p q Carl-Erik Sarndal, Bengt Swensson, Jan Wretman (1992). Model Assisted Survey Sampling. ISBN 9780387975283.{{cite book}}: CS1 maint: uses authors parameter (link)
  3. ^ Thomas Lumley (https://stats.stackexchange.com/users/249135/thomas-lumley), How to estimate the (approximate) variance of the weighted mean?, URL (version: 2021-06-08): https://stats.stackexchange.com/q/525770
  4. ^ Gatz, Donald F.; Smith, Luther (June 1995). "The standard error of a weighted mean concentration—I. Bootstrapping vs other methods". Atmospheric Environment. 29 (11): 1185–1193. Bibcode:1995AtmEn..29.1185G. doi:10.1016/1352-2310(94)00210-C. - pdf link
  5. ^ Endlich, R. M., et al. "Statistical analysis of precipitation chemistry measurements over the eastern United States. Part I: seasonal and regional patterns and correlations." Journal of Applied Meteorology (1988-2005) (1988): 1322-1333. (pdf)
  6. ^ "GNU Scientific Library – Reference Manual: Weighted Samples". Gnu.org. Retrieved 22 December 2017.
  7. ^ "Weighted Standard Error and its Impact on Significance Testing (WinCross vs. Quantum & SPSS), Dr. Albert Madansky" (PDF). Analyticalgroup.com. Retrieved 22 December 2017.
  8. ^ a b Price, George R. (April 1972). "Extension of covariance selection mathematics" (PDF). Annals of Human Genetics. 35 (4): 485–490. doi:10.1111/j.1469-1809.1957.tb01874.x. PMID 5073694. S2CID 37828617.
  9. ^ Mark Galassi, Jim Davies, James Theiler, Brian Gough, Gerard Jungman, Michael Booth, and Fabrice Rossi. GNU Scientific Library - Reference manual, Version 1.15, 2011. Sec. 21.7 Weighted Samples
  10. ^ James, Frederick (2006). Statistical Methods in Experimental Physics (2nd ed.). Singapore: World Scientific. p. 324. ISBN 981-270-527-9.
  11. ^ G. H. Hardy, J. E. Littlewood, and G. Pólya. Inequalities (2nd ed.), Cambridge University Press, ISBN 978-0-521-35880-4, 1988.
  12. ^ Jane Grossman, Michael Grossman, Robert Katz. The First Systems of Weighted Differential and Integral Calculus, ISBN 0-9771170-1-4, 1980.

Further reading

  • Bevington, Philip R (1969). Data Reduction and Error Analysis for the Physical Sciences. New York, N.Y.: McGraw-Hill. OCLC 300283069.
  • Strutz, T. (2010). Data Fitting and Uncertainty (A practical introduction to weighted least squares and beyond). Vieweg+Teubner. ISBN 978-3-8348-1022-9.

External links