EXAMPLE: If in the sample of 400 respondents we have: (1) 60 votes for "red" (2) 120 votes for "blue" (3) 20 votes for "green" (4) 200 votes for "grey" (5) 0 votes for "black" i.e. nobody voted for blacks but we know for sure that such a party exists, then Bayes estimates for proportions with 99% confidence interval can be calculated as: bayes 60 120 20 200 0 / 99%. In the case of 95% confidence interval the last argument can be omitted. bayes 60 120 20 200 0. ******************************** * Macros: Bayes.sps * * Author: Evgeny Ivashkevich * * E-mail: ivashkev@yandex.ru) * * Date : 2005/07/19 * ******************************** preserve. set printback=off. define bayes (!positional !charend ('/') /!positional !charend ('%') !default(95)). !let !n = 0. !let !r = 0. !do !k !in (!1). !let !n = !length(!concat(!blanks(!n),!blanks(!k))). !let !r = !length(!concat(!blanks(!r),!blanks(1))). !doend. input program. numeric id (f2) pr se lb ub (f7.5). vector cov(!r e10.3) #p(!r). string info(A10). variable labels info ' ' id '#' pr 'Proportion' se 'Standard Error' lb 'Lower Bound' ub 'Upper Bound'. compute #i = 0. !do !k !in (!1). compute #i = #i+1. compute #p(#i) = (!k+1)/(!n+!r). !doend. loop #i=1 to !r. compute id = #i. compute pr = #p(#i). compute se = sqrt(#p(#i)*(1-#p(#i))/(!n+!r+1)). compute lb = max(#p(#i)+probit((100-!2)/200)*se,0). compute ub = min(#p(#i)+probit((100+!2)/200)*se,1). loop #j=1 to !r. if (#j =#i) cov(#j) = #p(#i)*(1-#p(#i))/(!n+!r+1). if (#j<>#i) cov(#j) = -#p(#i)*#p(#j)/(!n+!r+1). end loop. if (\$casenum=1) info = concat('P =',string(!2,f4),'%'). if (\$casenum=2) info = concat('N =',string(!n,f5)). if (\$casenum>2) info = ' '. end case. end loop. end file. end input program. execute. report format = list automatic align(center) /variables = id pr se lb ub info /title 'BAYES ESTIMATES FOR PROPORTIONS AND CONFIDENCE INTERVALS'. !enddefine. restore. bayes 60 120 20 200 0 / 99%.