This preview shows pages 1–4. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Unit 3: Experiments with More Than One Factor Sources : Chapter 3. • Paired comparison design (Section 3.1). • Randomized block design (Section 3.2). • Twoway and multiway layout with fixed effects (Sections 3.3 and 3.5). • Latin and GraecoLatin square design (Sections 3.6 and 3.7). • Balanced incomplete block design (Section 3.8). • Splitplot design (Section 3.9). • Analysis of covariance (ANCOVA) (Section 3.10). • Transformation of response (Section 3.11). 1 Sewage Experiment • Objective : To compare two methods MSI and SIB for determining chlorine content in sewage effluents; y = residual chlorine reading. Table 1: Residual Chlorine Readings, Sewage Experiment Method Sample MSI SIB d i 1 0.39 0.36 0.03 2 0.84 1.35 0.51 3 1.76 2.56 0.80 4 3.35 3.92 0.57 5 4.69 5.35 0.66 6 7.70 8.33 0.63 7 10.52 10.70 0.18 8 10.92 10.91 0.01 • Experimental Design : Eight samples were collected at different doses and contact times. Two methods were applied to each of the eight samples. It is a paired comparison design because the pair of treatments are applied to the same samples (or units). 2 Paired Comparison Design vs. Unpaired Design • Paired Comparison Design : Two treatments are randomly assigned to each block of two units. Can eliminate blocktoblock variation and is effective if such variation is large. Examples : pairs of twins, eyes, kidneys, left and right feet. (Subjecttosubject variation much larger than withinsubject variation). • Unpaired Design : Each treatment is applied to a separate set of units, or called the twosample problem. Useful if pairing is unnecessary; also it has more degrees of freedom for error estimation (see page 5). 3 Paired t tests • Paired t test : Let y i 1 ,y i 2 be the responses of treatments 1 and 2 for unit i,i = 1 ,...N . Let d i = y i 2 y i 1 , ¯ d and s 2 d the sample mean and variance of d i . t paired = √ N ¯ d/s d The two treatments are declared significantly different at level α if  t paired  > t N 1 ,α/ 2 . (1) 4 Unpaired t tests • Unpaired t test : The unpaired t test is appropriate if we randomly choose N of the 2 N units to receive one treatment and assign the remaining N units to the second treatment. Let ¯ y i and s 2 i be the sample mean and sample variance for the i th treatment, i = 1 and 2. Define t unpaired = ( ¯ y 2 ¯ y 1 ) / radicalBig ( s 2 2 /N ) + ( s 2 1 /N ) . The two treatments are declared significantly different at level α if  t unpaired  > t 2 N 2 ,α/ 2 . (2) Note that the degrees of freedom in (1) and (2) are N 1 and 2 N 2 respectively. The unpaired t test has more df’s but make sure that the unittounit variation is under control (if this method is to be used). 5 Analysis Results : t tests t paired = . 4138 . 321 / √ 8 = . 4138 . 1135 = 3 . 645 , t unpaired = 5 . 435 5 . 0212 radicalbig (17 . 811 + 17 . 012) / 8 = . 4138 2 . 0863 = 0 . 198 . The p values are Prob (  t 7  > 3 . 645) = . 008 , Prob (  t 14  > . 198) = . 848 .....
View
Full
Document
This note was uploaded on 12/12/2010 for the course STAT 425 taught by Professor Ma,p during the Fall '08 term at University of Illinois, Urbana Champaign.
 Fall '08
 Ma,P

Click to edit the document details