90 likes | 226 Views
Clase 4. Summary…. Experiments should be comparative Genuine replication ( many times) Blocking ( pairing ) to avoid errors Randomizations should be part of the experiment Only checking original data can prevent bad data
E N D
Summary… • Experimentsshouldbecomparative • Genuinereplication (many times) • Blocking (pairing) toavoiderrors • Randomizationsshouldbepart of theexperiment • Onlychecking original data can preventbad data • Wachoutfor IID, exchangeability and linear dependenceviolations
Confidenceintervals • Values of thestatistic (variance, mean, etc) thatmakethesamplevaluesignificant at thestatedlevel of probability.
Chi-square • NIID • Then • So, forunknown mean • Ex. s2=13, n=6,=0.05 Rough estimate of Variation in stdev (%):
F (Fisher) distribution • Two populations with σ1, σ2, normal, samples n1 and n2. Sample variances s1, s2 (DOF n1,2-1) • The ratio What’s all that for?? ex. Fertilizer pp. 78; randomization
Confidencelimits • Since • In terms of confidence limits Ex. pp104
Howaboutdiscretepdfs? • Binomial • Ex. ‘tricky’ penny: 20 tosses, 15 heads Pr(y>14)=Pr(y=15)+Pr(y=16)+Pr(y=17)+… =0.015 +0.005 +0.001 +0.000 +… Ex. Pp106 ,107
Poissondist. • Binomial, p0, nlarge, finite • Varianceisalso • If >5, Poisson normal (y- )/ and ify1…yk are observationsfromindependentdist. withmeans1,… k Ex pp111