Professional Documents
Culture Documents
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
1. Rakamlar toplam
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
Rakamlar toplam
= toplam; X aile geliri gibi deikendir. Daha sonra toplam aile geliri N gzlemlerinin deeri
i =1
Xi = X 1 + X 2 + ... + XN
dmacpher@coss.fsu.edu
Jump to first page
Rakamlar toplam
Sabit zamanlar deikeni toplam is deikenler toplamna eittir.
k i =1 Xi = kX 1 + kX 2 + ... + kXN
N
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
Rakamlar toplam
ki deiken gzlemlerinin toplam onlarn dier toplamlarna eittir
N i =1
( Xi + Yi ) =i =1 Xi + i =1 Yi
N N
dmacpher@coss.fsu.edu
Jump to first page
Rakamlar toplam
N zerindeki sabit gzlemlerin toplam sabit rn ve N e eittir.:
i =1
k =kN
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
2. Tanmlar
dmacpher@coss.fsu.edu
Jump to first page
Rastgele Deney
Rastgele deney: a process leading to at least two possible outcomes with uncertainty as to which will occur.
Sonular zlebilir
rnein - lmek, madeni para atmek, desteden kart ekmek sastgele deneydir.
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
rnek alan
Bir deneyin tm olaslk sonularnn kurulumu is rnek ktle olarak tanmlanr. rnekler :
madeni para atmak S={H,T}. lm S={1,2,3,4,5,6} 3 sonulu madeni para atmak S={HHH, HHT, HTH, THH, HTT, THT, TTH, TTT}
dmacpher@coss.fsu.edu
Jump to first page
rnek noktalar
rnek noktarnek alann sonularnn herbiridir. rnekler :
Y veya T 1,2,3,4,5 veya 6 YYY
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
Sonular
An event is a subset of the sample space.
rnekler :
Sonu A iki madeni para atldnda yaz gelmesi. Sonu TT sonu A ya aittir.sonu a 3 madeni para atldnda nemli etkiye sahiptir.A={HHH, HHT, HTH, THH}
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
3. Olaslk tanmlar
dmacpher@coss.fsu.edu
Jump to first page
Klasik tanmlar
n olaslk sonularyla deney farzedelim , ve bu sonular are benzer ve ortak sonuludur.
rnek ktlede a sonucu kuralm. P(A) Ann olasldr. M sonular olumludur, daha sonraP(A) m/n dr
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
David Macpherson
8/12/2005
Amprik tanmlar
Klasik tanmda problem - belki deneyin tm sonular eit olmayabilir.
Sonularn numaralar sonlu deildir..
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
10
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
4. Probability Properties
dmacpher@coss.fsu.edu
Jump to first page
11
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
12
David Macpherson
8/12/2005
Eer A, B, ve C ortak zellikli ve ortak ayrntli sonularsa daha sonra bireysel olaslklarn toplam 1dir.
P(A+B+C)=P(A) + P(B) + P(C)= 1
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
13
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
14
David Macpherson
8/12/2005
artl olaslk
farzedelim sonu B nin olasln istiyoruz bilinen A meydana geldimi?
Bu artl olaslktr
P(B\A) = P(AB)/P(A)
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
15
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
Olaslk younluk fonksiyonu (PDF) rastgele deikenler olasln belirler. Rastgele deikenler iin X, pdf f(X) anlamna gelir. pdf x deerlerinin karsndak olaslklarn dalmn gsterir. Pdf nin toplam 1 dir.
dmacpher@coss.fsu.edu
Jump to first page
16
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
The discrete PDF of X is f(x) =P(X=xi) for i=1,2,...n and f(x) = 0 for X xi where P(X= xi) is probability that the discrete random variable takes the value of xi. Bu rastgele deikenler byk harfalerle tanmlanmtr X, Y etc. (and dierleri kk harf, x, y, etc.).
dmacpher@coss.fsu.edu
Jump to first page
17
David Macpherson
8/12/2005
.25
dmacpher@coss.fsu.edu
Jump to first page
.45 .20 .10 .05 <5 5-9 10-14 15-19 >19 Inches of Rain .20
dmacpher@coss.fsu.edu
Jump to first page
18
David Macpherson
8/12/2005
CDF F(X)=P(X x) dir veya rastgele deikenlerin olasl X kk x ten daha byk deer tar.
Forml , F(X)=xf(x)
dmacpher@coss.fsu.edu
Jump to first page
Multivariate PDF
Eer sonular bir rastgele deikenden daha fazla yaylmsa, daha sonra farkl olaslk dalm fonksiyonlarna sahip oluruz. Bivariate PDF iki rastgele deiken yaylr. formul, the discrete joint PDF is f (X,Y) = P(X=x ve Y=y) ve f (X,Y) = 0 when X x and Y y
dmacpher@coss.fsu.edu
Jump to first page
19
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
This is a discrete bivariate PDF. Each cell is a joint probability of college education and income.
dmacpher@coss.fsu.edu
Jump to first page
20
David Macpherson
8/12/2005
Marjinal PDF
f(X) ve f(Y) aremarjinal PDF Xin marjinal PDF x olasl y nin mutlak deeridir.
dmacpher@coss.fsu.edu
Jump to first page
Marjinal PDF
Y $5,000 $10,000 $15,000 Total f(Y) 0.250 0.250 0.500 1.000 X 0 1 Total f (X) 0.542 0.458 1.000
dmacpher@coss.fsu.edu
Jump to first page
21
David Macpherson
8/12/2005
artl PDF
Y nin artl PDF verilmi x deerinin y olasldr f(Y|X) = P(Y=y | X=x) where | X) y nin artl PDF dir..
Gibi hesaplanr: f(Y | X)= f(X,Y)/f(X)
f(Y
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
22
David Macpherson
8/12/2005
statiksel bamszlk
X ve y deikenleri istatiksel bamszdr eer onlarn marjinal PDF leridir.
f(X,Y) = f(X)* f(Y)
dmacpher@coss.fsu.edu
Jump to first page
Let X=H ilk atlan T if tail and let Y=H if head on second coin and T if tail
23
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
6. Beklenen deer
dmacpher@coss.fsu.edu
Jump to first page
24
David Macpherson
8/12/2005
PDFN MOMENT
Dalmn birinci momentibeklenen deer veya anlmdr. kinci moment yaryanstr. Beklenen deer veya popilasyon anlam deeri:
E ( X ) = i =1 Xif ( Xi )
N
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
25
David Macpherson
8/12/2005
2. iki rastgele deikenlerin toplamnn bekleneni rastgele deikenlerin beklenenleri toplamna eittir.
E(X + Y) = E(X) + E(Y)
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
26
David Macpherson
8/12/2005
4. iki rastgele deiken rnn beklenen deeri ratgele deikenlerin beklenen deerlerinin rnne eit deildie genellikle.
E(XY) E(X)E(Y)
Bununlaberaber eer X ve Y bamsz rastgele deikenlersee dorudur.
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
27
David Macpherson
8/12/2005
7. Varyans
dmacpher@coss.fsu.edu
Jump to first page
Varyans
varyans anlamn trafnda dalm gsterir.
var(X)=2 = E[(X-)]2 , yerde = E(X) veya beklenen deer varyans x ve beklenen deeriarasnda karesinin beklenen deeridir. Standart sappma varyansn karesidir..
dmacpher@coss.fsu.edu
Jump to first page
28
David Macpherson
8/12/2005
Varyans
X anlamn etrafnda dalmsa X genitir,
dmacpher@coss.fsu.edu
Jump to first page
Varyans hesaplanmas
Varyans hesaplanmas
Var( X ) = f ( Xi )[ Xi E( X )]2
i =1 N
dmacpher@coss.fsu.edu
Jump to first page
29
David Macpherson
8/12/2005
Variance Example
Varyans hesaplama rnekleri
X -2 0 2 3 f(X) 0.27 0.12 0.26 0.35 [X-E(X)]2 9.18 1.06 0.94 3.88
[X-E(X)]2f(X)
X2 4 0 4 9
E(X) = 1.03 and Var (X) = 421 or Var (X) = 5.27 - (1.03)2 = 4.21
dmacpher@coss.fsu.edu
Jump to first page
Varyansn zellikleri
1. artl varyans sfrdr. 2. x ve Y iki Bamsz deikensex ve y are 2 independent random variables, tonlar toplam veya farklarn varyans zel varyanslarnn toplamna eittir.
var (X+Y) = var(X) + var(Y) var (X-Y) = var(X) + var(Y)
dmacpher@coss.fsu.edu
Jump to first page
30
David Macpherson
8/12/2005
Varyansn zelliklerei
Deikenlere art eklemek onun varyansn deitirmez.
B artlysa var (X + b) = var (X)
4. b artlysab zaman deikenlerinin varyansdeikenlerin varyansnn zaman artnn karesine eittir., var(bX)= b2 var(X)
dmacpher@coss.fsu.edu
Jump to first page
Varyans zellikleri
5.a ve b sartlysa var (aX + B)= a2 var(X) 6. , X ve Y bamsz rastgele deikenlerse a ve b artldr.
var (aX + bY) = a2 var(X)+b2 var(Y)
dmacpher@coss.fsu.edu
Jump to first page
31
David Macpherson
8/12/2005
8 kovaryans
dmacpher@coss.fsu.edu
Jump to first page
Kovaryans
Kovaryans deikenn birlikte hareketlerini ler. Farz edelim 2 rastgele deikene sahibiz , kovaryans:
Cov(X,Y)= E[(X-E(X)) (Y-E(Y))] or Cov(X,Y) = E(XY) -X Y
dmacpher@coss.fsu.edu
Jump to first page
32
David Macpherson
8/12/2005
Kovaryans hesaplanmas
dmacpher@coss.fsu.edu
Eer 2 deiken beraber hareket ediyorsa,kovaryans poztf.farkl Jump hareket ediyorsa negatiftir. to first page
Kovaryans rnei
Ortak dalm farz edelim
Y 3 6 f(X) -2 0.27 0.00 0.27 X 0 0.08 0.04 0.12 2 0.16 0.10 0.26 3 0.00 0.35 0.35 f(Y) 0.51 0.49 1.00
dmacpher@coss.fsu.edu
Jump to first page
33
David Macpherson
8/12/2005
Kovaryans rnei
X -2 0 2 3 f(X ) 0.27 0.12 0.26 0.35 Xf(X) -0.54 0 0.52 1.05 Y 3 6 f(Y Yf(Y) ) 0.51 1.53 0.49 2.94
2 X 4 0 4 9 2 E(X) 1.08 0 1.04 3.15 5.27 2 Y 9 36 2 E(Y) 4.59 17.64 22.23
) E(X) = 1.03 and E(Y = 4.47 2 var(X = 5.27-(1.03) = 4.21 ) var(Y = 22.23-(4.47)2 = 2.25 )
dmacpher@coss.fsu.edu
Jump to first page
Kovaryans rnei
Cov (X,Y) = XYXY f(X,Y) - X Y XYXY f(X,Y) = (-2)(3)(0.27) +(0)(3)(0.08)+(2)(3)(0.16)+(3)(3)(0) +(-2)(6)(0) + (0)(6)(0.04) + (2)(6)(0.10)+(3)(6)(0.35)
= -1.62 + 0.96 + 1.2 + 6.3 = 6.84
34
David Macpherson
8/12/2005
Kovaryansn zellikleri
X ve Y bamsz rastgele deikenler isekovaryanslar sfrdr.
If 2 r.v.s are independent, then:
E(XY) = E(X)E(Y) = X Y kovaryansta yerine koyalm
Kovaryansn zellikleri
Deikenlerin kovaryans onun basit varyansdr.
dmacpher@coss.fsu.edu
Jump to first page
35
David Macpherson
8/12/2005
Korelasyon katsays
eer deikenler beraber hareket ediyorsa kovaryans tanmlar. Korelasyon katsays deikenin ne kadar gl ilikisini gsterir. =cov(X,Y)/X Y Korelasyon katsays rnei:
X =2.05, Y =1.50, cov (X,Y) = 2.24.
= 2.24/(2.05)(1.5) = 0.73
dmacpher@coss.fsu.edu
Jump to first page
Korelasyon katsays kovaryans ile ayn iarete sahip. Korelasyon katsays -1 ve +1 arasndadr
kovaryans llm X ve Y blmlerine baldr, fakat korelasyon katsays deildir. eer = 1 ise anlam iki deiken tam poztf korelasyona sahiptir.
dmacpher@coss.fsu.edu
Jump to first page
36
David Macpherson
8/12/2005
artl beklentiler
artl beklentiler: x in beklene deeri ne ise,y deerleri verilir. E(X|Y=y) = XX f(X|Y=y)
f(X|Y=y) X PDF leri artldr ve x deerlerini toplarz.
dmacpher@coss.fsu.edu
Jump to first page
37
David Macpherson
8/12/2005
9. Toplamdan rnee
dmacpher@coss.fsu.edu
Jump to first page
38
David Macpherson
8/12/2005
Toplamdan rnee
Beklenen deerleri hesaplamak, pdf in varyans,rnek ktlenin tmne veya toplam ktleye ihtiyacmz var. pratikte toplamdan temsil eden rnek kullanrz.
Bu bize anlam varyans etc tahminini verir.
dmacpher@coss.fsu.edu
Jump to first page
rnek anlam
rnek anlam E(X) in tamn edicisidir.
X =
i=1
/ n
dmacpher@coss.fsu.edu
Jump to first page
39
David Macpherson
8/12/2005
rnek varyans
rnek varyans 2n toplam varyanstan tahmin edicisidir.
2 x
i =1
( X i X )2 n 1
n-1 serbestlik derecesidir. N e blersekdaha sonra n-1, we get a biased estimate since only n-1 unconstrained observations exist.
dmacpher@coss.fsu.edu
Jump to first page
rnek kovaryans
rnek kovaryans toplam kovaryansn tahmin edicisidir.
Sample Cov(X,Y) =
(X
X )(Yi Y ) n 1
dmacpher@coss.fsu.edu
Jump to first page
40
David Macpherson
8/12/2005
X )(Yi Y ) /(n 1) S X SY
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
41
David Macpherson
8/12/2005
Normal dalm
simetriktir, bell-shaped distribution. Normal dalm srekli rastgele deiken X in pdf:
f (X ) = 1 X 1 exp 2 2
Normal dalm
dmacpher@coss.fsu.edu
Jump to first page
42
David Macpherson
8/12/2005
olaslklar
Normal dalmn olaslklar:
olaslk(-<X< +) .68 olaslk (-2<X< +2) .95 olaslk(-3<X< +3) .997
dmacpher@coss.fsu.edu
Jump to first page
43
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
44
David Macpherson
8/12/2005
Tabloya bak: p(0 Z 2)= 0.4772 Simetrik dalm: p(-2 Z 0)= 0.4772 Bundan dolay olaslk = 0.4772+0.4772= 0.9544
dmacpher@coss.fsu.edu
Jump to first page
SAT sonular ~N (500,10000) farz edelim herhangi bir sonucun 750 zerinde olma olasl ne? Z=(X1-x)/ x=(750-500)/100 =2.5
Pr(0<Z<2.5)=0.4938 So Pr (Z>2.5) = 0.5-0.4938 = .0062 or .6 percent
45
David Macpherson
8/12/2005
Dorusal birleme
2 normal dalm rastgele deikenin dorusal birleimi onun normal dalmdr.
2 2 If X ~ N(X , X ) and Y ~ N(Y , Y ) and they are independen t
dmacpher@coss.fsu.edu
dmacpher@coss.fsu.edu
Jump to first page
46
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
Sampling of Means
Assume a random sample of X1,X2,,Xn from a population PDF with and 2 .
Random sample - the r.v.s must be independent and identically distributed (i.i.d)
PDF of each Xi is ~ N(, 2)
Let then
dmacpher@coss.fsu.edu
Jump to first page
X = X
/ n
2
X ~ ( ,
/ n)
47
David Macpherson
8/12/2005
Sampling of Means
Konu ile ilgili Z sonular:
Z =
X / n
dmacpher@coss.fsu.edu
Jump to first page
rnek iin Ortalama (X) ~N(75,25)dir.16 rencinin ortalamasnn 78 den byk olma olasl nedir?
X ~ N ( 75 , 25 / 16 ) X Z =
78 75 5 / 16
= 2 .4
48
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
49
David Macpherson
8/12/2005
Ki kare dalm
rastgele deikenlerin varyanslaryla dalm hipotez testi iin yararl dalm. Distribution of these squared terms approximates the chi-squared distribution.
dmacpher@coss.fsu.edu
Jump to first page
Ki kare dalm
Eer Z standart normal deikense (i.e. with =0 and =1), sonra Z2 k serbestlik derecesiyle ki kare dalmna sahiptir.
2 k
Jump to first page
= 2Z
50
David Macpherson
8/12/2005
Ki kare dalm
Ki kare orjinden balar. Dalmn ekli k ya baldr.
Artan k dalm daha simetrik yapar.. k geni oluncaki kare hemen hemen normaldir.
Ki kare dailm
dmacpher@coss.fsu.edu
Jump to first page
51
David Macpherson
8/12/2005
Ki kare dalm
Tablo A-3
k\Q 1 5 10 20 30 40 0.250 1.323 6.626 12.549 23.828 34.800 45.616 0.100 2.706 9.236 15.987 28.412 40.256 51.805 0.050 3.841 11.071 18.307 31.410 43.773 55.758 0.025 5.024 12.833 20.483 34.170 46.979 59.342 0.010 6.635 15.086 23.209 37.566 50.892 63.691 0.005 7.879 16.750 25.188 39.997 53.672 66.766 0.001 10.828 20.515 29.588 45.315 59.703 73.402
verilen 20 serbestlik derecesinde X2 karenin deeri 40 ve daha zeri olma olasl ne?
dmacpher@coss.fsu.edu
S2 Then ( n 1) 2
dmacpher@coss.fsu.edu
~ X
( n 1 )
52
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
13. t-dalm
dmacpher@coss.fsu.edu
Jump to first page
53
David Macpherson
8/12/2005
t dalm
varyans bilinmezken hipotezi test etmek iin t dalmn kullanrz.
Remember that if X ~ N ( , 2 / n), then X ~ N (0,1) / n Suppose only know and don' t know 2 , Z= we only know its estimator S2 .
dmacpher@coss.fsu.edu
Jump to first page
t dalm
Remember S 2 = t= X S/ n
(X
X )2
( n 1)
dmacpher@coss.fsu.edu
Jump to first page
54
David Macpherson
8/12/2005
t dalm
dmacpher@coss.fsu.edu
Jump to first page
t dalmnn zellikleri
Normal gibi, t dalm simetriktir ve geni rnekler iinnormale yaklasktr.
rnek ktle 30dan kkken daha genitir. 30 zerindeki rnek ktlede,t normale yaklasktr..
dmacpher@coss.fsu.edu
Jump to first page
55
David Macpherson
8/12/2005
t Daglmnn zellikleri
t dalmnn anlamstandart normal dalmdaki gibi sfrdr. varyans k/(k-2).
Geni rnek ktleler iin, the varyans standart normal dalmdaki gibi 1 e yaklasr.
dmacpher@coss.fsu.edu
Jump to first page
t dalm tablosu
Table A-2
k\Q 1 5 10 20 30 Infinity 0.500 1.000 0.727 0.700 0.687 0.683 0.674 0.200 3.078 1.476 1.372 1.725 1.310 1.282 0.100 6.314 2.015 1.812 2.086 1.697 1.645 0.050 12.706 2.571 2.228 2.131 2.042 1.960 0.025 31.821 3.365 2.764 2.528 2.457 2.326 0.010 63.657 4.032 3.169 2.845 2.750 2.576 0.002 318.310 5.893 4.144 3.552 3.385 3.090
Given df=20, what is the probability of obtaining a t value of 2.85 or greater = 0.01
dmacpher@coss.fsu.edu
Jump to first page
56
David Macpherson
8/12/2005
t dalm tablosu
%5 anlamllk testi iin,t dalmnn kritik deeri 1.96 ya yaklasr.normal dalmn krtk deeri n geni olunca olur. 30 veya zeri rnek iin, 2 nin kritik deeri akla uygundur.
dmacpher@coss.fsu.edu
Jump to first page
t dalm rnei
25 snav sonucunun rastgele deikenlerine sahibiz anlam:78 ve S2 :16.N 75 anlamllnda rnek ortalama olasl nedir?
t=
dmacpher@coss.fsu.edu
57
David Macpherson
8/12/2005
14. F dalm
dmacpher@coss.fsu.edu
Jump to first page
F Distribution
X ve Y toplam ktleden iki bamsz rastgele deiken sahiibiz X ~ N (X, 2X) and Y~ N (Y, 2Y) Varyanslarnn ayn olup olmadn test etmek istiyoruz..
Tahmn edilmi varyanslar S2X ve S2Y oran S2X/ S2Y ~ F dr her iki ana ktle eitse.
Jump to first page
dmacpher@coss.fsu.edu
58
David Macpherson
8/12/2005
F Distribution
F-dalmn Fm,n ile belirtiyoruz.serbestlik derecesi iki deikenle tanmland zaman, m numerator oluyor ve n serbestlik derecesidir F daima byktr1den.
If varyanslar eitse, F 1 dzeyndedir. F nin daha geni deeri, varyanslar eittir.
Jump to first page
dmacpher@coss.fsu.edu
F Dalm
F dalm is arpktr, ki karede olduu gibi. Sralamas sfr ve sonsuz arasnda. m ve n byrken,f testi normale yaklasr.
dmacpher@coss.fsu.edu
Jump to first page
59
David Macpherson
8/12/2005
Given m = 7 and n = 10, what is the probability of obtaining an F value of 3.14 or greater = 0.05
dmacpher@coss.fsu.edu
Jump to first page
F dalm rnei
Suppose test whether the variance of the SAT math test differs from the variance of the SAT verbal test. 21 rencinin rnei kullanlyor, S2 matematik testinde buluruz ve szl 100,000 ve 80,000. Normal dalan test ktlesi olduunu farz edelim, f istatistii 1.25 (100,000/80,000) ile20 ve20 serbestlik derecesidir..
Jump to first page
dmacpher@coss.fsu.edu
60
David Macpherson
8/12/2005
F dalm rnei
Elde edilen 1.25 in Fdeeri serbestlik dercesinden fazladr ki varyansn eit olduu kural uygulayamayz
If we had come up with a higher number - e.g. 3.0 the probability of obtaining this is less than .01
Ana ktle varyasnn eit olduunu syleyemeyz.
dmacpher@coss.fsu.edu
Jump to first page
dmacpher@coss.fsu.edu
Jump to first page
61
David Macpherson
8/12/2005
yi tahmin nedir?
kriter:
Computational Cost yansz Hzl ve verimli Best Linear Unbiased Estimator tutarl
dmacpher@coss.fsu.edu
Jump to first page
Computational Cost
Some methods involve a lot more time and effort than others. More advanced techniques can often considerable extra effort - is it worth it? Fortunately, many types of models are now routinely available on most statistical packages, so this criteria is less important than it used to be.
Jump to first page
dmacpher@coss.fsu.edu
62
David Macpherson
8/12/2005
yansz
1000 rnek aldmz farz edelim Her rnek iin rnek anlamn tahmin edicisine sahip olmamz gerekir. Bu tahminler ayn almayacaktr, normale yaklasan salmn rneine yaklasacaktr.
dmacpher@coss.fsu.edu
Jump to first page
yansz
So in repeated sampling the mean value of all these estimators tends towards the true population mean,
rnek anlam yansz tahmin edici olarak tanmlanr.nk:
E(X ) =
dmacpher@coss.fsu.edu
Jump to first page
63
David Macpherson
8/12/2005
Verimli
Birden fazla yansz tahmnedene sahip olabiliriz arasndan hangini seebiliriz?
Enkk varyansl rnek ktleninki seilir.. Geni varyansla tahmin ediciden tahmn yapmak daha az istenir,
dmacpher@coss.fsu.edu
Jump to first page
verimli
Tahmin edici sfr varyansa sahipse,parametreleri tam bulabiliriz. Yansz tahmin ediciyi enkk varyansla seilir. En yansz veya en verrimli tahmin edicidir.
dmacpher@coss.fsu.edu
Jump to first page
64
David Macpherson
8/12/2005
Bu ekleme tahmin edicinin dorrusal olmasn snrlar. rnein, rnek anlam dorusal tahmin edicidir.
X = X i / n i=1 This makes it easier to figure out which is the most efficient estimator
n
dmacpher@coss.fsu.edu
Jump to first page
An estimator that is linear, unbiased, and has minimum variance among all linear unbiased estimators is called the best linear unbiased estimator - BLUE. Parametrelerin teknumaral olanlar ile ilgileniyouz.
It is harder to find the BLUE in the the multidimensional case.
dmacpher@coss.fsu.edu
Jump to first page
65
David Macpherson
8/12/2005
tutarl
rnek ktle genil olduunda tahmin edici ne olur?
Tahmin edici tutarl ise rnek ktle daha geni olur, toplam parametrelerin dorru deerine yaklar..
dmacpher@coss.fsu.edu
Consistent
Sometimes we can get consistency but not unbiasedness, where the estimator may not equal the true parameter on average, but will approximate the true parameter as sample size gets large.
dmacpher@coss.fsu.edu
Jump to first page
66
David Macpherson
8/12/2005
16. Estimation
dmacpher@coss.fsu.edu
Jump to first page
Estimators
An estimator is generated when we take a random sample and compute, for example, the mean or variance. Typically, we do not know the mean and variance of the population, so we have to rely on our sample estimates.
dmacpher@coss.fsu.edu
Jump to first page
67
David Macpherson
8/12/2005
Point Estimates
Suppose we take a random sample of 16 students and get an average of 75 and variance 25.
The sample mean of 75 is the point estimate of the population parameter,which is typically unknown.
This point estimate will vary across samples, since the sample mean is an r.v.
dmacpher@coss.fsu.edu
Jump to first page
Interval Estimates
An alternative is that a certain interval contains the true mean.
Remember that if X ~ N ( , 2 / n ), then X ~ N (0,1) / n Suppose don' t know 2 , Z= we only know its estimator S2 .
dmacpher@coss.fsu.edu
Jump to first page
68
David Macpherson
8/12/2005
Interval Estimates
If we use S2 , then we know t = X S/ n which follows the t distributi on with n - 1 d.f.
dmacpher@coss.fsu.edu
Jump to first page
Suppose we want to construct a 95% confidence interval around this mean value of 75
There is a 95% probability that this interval will contain the true .
69
David Macpherson
8/12/2005
2.131(5) 2.131(5) ) = .95 < <X+ 16 16 (75 - 2.66) < < (75 + 2.66)
72.34<<77.66 is the 95% confidence interval for So there is a 95% probability that this interval contains the true mean
dmacpher@coss.fsu.edu
Jump to first page
70
David Macpherson
8/12/2005
Alpha
Sometimes the confidence interval is expressed as:
P(lower limit< <upper limit) =1-
1- is the probability that the random interval contains the true (the confidence coefficient).
dmacpher@coss.fsu.edu
Jump to first page
71
David Macpherson
8/12/2005
Hypothesis Testing
We can evaluate the relationship between the sample estimates and the population parameters. Suppose a random sample has a mean of 75 on an exam.
Hypothesize the true mean()=78.
Is 75 statistically different from 78? Is the 75 due to sampling error? Is the 75 because is not 78? This is hypothesis testing.
Jump to first page
dmacpher@coss.fsu.edu
Hypotheses
Set up a null hypothesis.
Ho: = 78
72
David Macpherson
8/12/2005
73
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
74
David Macpherson
8/12/2005
dmacpher@coss.fsu.edu
Jump to first page
71.32<<78.68 is the 99% confidence interval for Our hypothesis is not rejected - because we have minimized the probability of committing a Type I error.
Jump to first page
dmacpher@coss.fsu.edu
75
David Macpherson
8/12/2005
Type II error = (accept Ho when it is false). The confidence coefficient is 1- The power of the test is 1-
dmacpher@coss.fsu.edu
Jump to first page
76
David Macpherson
8/12/2005
Suppose we take our random sample of 16 students, mean of 75 and variance 25.
dmacpher@coss.fsu.edu
Jump to first page
So, Xbar= 75, S=5, n=25 We don't know , so specify a value for . Set up hypothesis test:
H0: = 78 H1: 78
Calculate t:
t = (75-78)/(5/4) = -2.400
dmacpher@coss.fsu.edu
Jump to first page
77
David Macpherson
8/12/2005
Our t-value is lower than this, so we can reject the null hypothesis that the true is 78 at the 5% level.
dmacpher@coss.fsu.edu
Jump to first page
Suppose we choose an of 1%; the critical t values are + or 2.947. Our t-value is within this range, so we cannot reject the null at the 1% level of significance.
dmacpher@coss.fsu.edu
Jump to first page
78
David Macpherson
8/12/2005
t-test
This test is called the t-test.
We will use it as a test of significance of slope parameters in regression.
Statistically significant means we can reject the null hypothesis. Commonly, we choose 1%, 5% and 10% levels of significance.
dmacpher@coss.fsu.edu
Jump to first page
One-tailed test
Can also do a one-tailed test.
H0: = 78 H1: < 78
Again, we choose 5% level of significance and find the t-value is <-1.753 So again reject the H0 that the mean equals or is greater than 78.
dmacpher@coss.fsu.edu
Jump to first page
79
David Macpherson
8/12/2005
2 test of significance
A random sample of 10 SAT scores has a variance of 84. Test the hypothesis that the true variance is 86. Ho: 2 = 86 H1: 2 86
S2 Remember ( n 1) 2
dmacpher@coss.fsu.edu
~ X
( n 1 )
2 test of significance
Choose 5% significance level
2 = 9(84/86) = 8.79
80
David Macpherson
8/12/2005
F test of significance
Remember S2X/ S2Y ~ Fm-1, n-1 if the variances of the two populations are equal. Assume two random samples of students in two different schools taking a test:
S2X = 9.0 with sample size 51 S2Y = 7.2 with sample size 41
Ho: 2X = 2Y
dmacpher@coss.fsu.edu
Jump to first page
F test of significance
F = 9/7.2 = 1.25
df numerator 50 df denominator 40
81