This preview shows pages 1–10. 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 DocumentThis 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 DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: ’ & ST3241 Categorical Data Analysis I Models For Matched Pairs Dependent Proportions and Conditional Models 1 ’ & Example: Rating of Performance • For a poll of a random sample of 1600 votingage British citizens, 944 indicated approval of the Prime minister’s performance in the office. • Six months later, of these same 1600 people, 800 indicated approval. 2 ’ & Example Second Survey First Survey Approve Disapprove Total Approve 794 150 944 Disapprove 86 570 656 Total 880 720 1600 3 ’ & Dependent Categorical Data • To compare categorical responses for two samples when each sample has the same subjects or when a natural pairing exists between each subject in one sample and a subject from the other sample. • The responses in the two samples are then statistically dependent . • The pairs of observations are called matched pairs . • A twoway table having the same categories for both classifications summarizes such data. 4 ’ & Comparing Dependent Proportions • Let n ij = the number of subjects making response i at the first survey and response j at the second. • In the example, the sample proportions approving are 944 / 1600 = 0 . 59 and 880 / 1600 = 0 . 55 . • These marginal proportions are correlated, and statistical analyses must recognize this. • Let π ij = probability that a subject makes response i at survey 1 and response j at survey 2. 5 ’ & Dependent Proportions • The probabilities of approval at the two surveys are π 1+ and π +1 , the first row and first column totals. • When these are identical, the probabilities of disapproval are also identical, and there is marginal homogeneity . • Note that, π 1+ π +1 = ( π 11 + π 12 ) ( π 11 + π 21 ) = π 12 π 21 • Marginal homogeneity is equivalent to equality of offmaindiagonal probabilities; that is π 12 = π 21 . • The table shows symmetry across the main diagonal. 6 ’ & Inference For Dependent Proportions Use δ = π +1 π 1+ Let d = p +1 p 1+ = p 2+ p +2 . • From the results on multinomial distributions, cov ( p +1 ,p 1+ ) = ( π 11 π 22 π 12 π 21 ) /n • Thus, var ( √ nd ) = π 1+ (1 π 1+ ) + π +1 (1 π +1 ) 2( π 11 π 22 π 12 π 21 ) • For large samples, d has approximately a normal distribution. • A confidence interval for δ is then d ± z α/ 2 ˆ δ ( d ) 7 ’ & Inference For Dependent Proportions • Here ˆ δ 2 ( d ) = [ p 1+ (1 p 1+ ) + p +1 (1 p +1 ) 2( p 11 p 22 p 12 p 21 )] /n = [( p 12 + p 21 ) ( p 12 p 21 ) 2 ] /n (1) • The hypothesis of marginal homogeneity is H : pi 1+ = pi +1 (i.e. δ = 0 ). • Wald test statistic is: z = d/ ˆ σ ( d ) 8 ’ & McNemar’s Test • Under H , an alternative estimated variance is ˆ σ 2 ( d ) = p 12 + p 21 n = n 12 + n 21 n 2 • The score test statistic is z = d ˆ σ ( d ) = n 21 n 12 ( n 21 + n 12 ) 1 / 2 • The square of z is a chisquared distrbution with df = 1 ....
View
Full
Document
This note was uploaded on 11/20/2011 for the course STATISTICS ST3241 taught by Professor Manwai's during the Spring '11 term at National University of Singapore.
 Spring '11
 ManWai's

Click to edit the document details