/* Demonstration of using -sim_ci- to demonstrate the concept of confidence intervals: */ version 9.2 global URL = "http://www2.jura.uni-hamburg.de/instkrim/kriminologie/Mitarbeiter/Enzmann/Software/" * If missing, install scheme lean: cap which scheme_lean.hlp if _rc net install gr0002_3, from("http://www.stata-journal.com/software/sj4-3") * Define program -sim_ci-: qui do "${URL}sim_ci.do" * Some examples (always using the default of 10 cases per sample): sim_ci sim_ci, samp(100) sim_ci, samp(100) seed(35) tab sim_ci, mean(15) sd(5) samp(100) seed(35) sim_ci, mean(15) sd(5) samp(100) seed(179) name(sim01, replace) scheme(lean1) sim_ci, mean(15) sd(5) samp(100) seed(754582) name(sim02, replace) scheme(lean1) sim_ci, samp(100) mean(15) sd(5) seed(83) name(sim03, replace) scheme(lean1) sim_ci, samp(100) mean(15) sd(5) onesided seed(83) name(sim04, replace) scheme(lean1) sim_ci, samp(100) mean(15) sd(5) level(90) seed(83) name(sim05, replace) scheme(lean1) * ------------------------------------------------------------------------------ /* Note: The following demonstration takes time because there are 16 simulations of 10,000 samples (with 10 to 72 cases): */ set matsize 10000 * a) t-distribution, two-sided, sigma=1: sim_ci, n(10) samp(10000) nograph sim_ci, n(18) samp(10000) nograph sim_ci, n(36) samp(10000) nograph sim_ci, n(72) samp(10000) nograph * b) t-distribution, one-sided, sigma=1: sim_ci, n(10) onesided samp(10000) nograph sim_ci, n(18) onesided samp(10000) nograph sim_ci, n(36) onesided samp(10000) nograph sim_ci, n(72) onesided samp(10000) nograph * c) normal-distribution, two-sided, sigma=1: sim_ci, n(10) z samp(10000) nograph sim_ci, n(18) z samp(10000) nograph sim_ci, n(36) z samp(10000) nograph sim_ci, n(72) z samp(10000) nograph * d) normal-distribution, one-sided, sigma=1: sim_ci, n(10) z onesided samp(10000) nograph sim_ci, n(18) z onesided samp(10000) nograph sim_ci, n(36) z onesided samp(10000) nograph sim_ci, n(72) z onesided samp(10000) nograph * =========================================================== /* Example of a simulation of 1,000,000 samples using 10,000 different random seeds (n = 10, mu = 15, sigma = 5) To run, delete comment indicators of line 93 and 66. NOTE: The simulation is time consuming (about 5 minutes)! */ /* * ---------------------------------------------------------- set linesize 110 sca sumprop = 0 sca minprop = 100 sca minseed = 0 sca maxprop = 0 sca maxseed = 0 forvalues i = 1/10000 { // 10,000 times 100 (samp) = 1,000,000 qui sim_ci, mean(15) sd(5) samp(100) seed(`i') nog if minprop > r(prop_sig) { sca minprop = r(prop_sig) sca minseed = `i' } if maxprop < r(prop_sig) { sca maxprop = r(prop_sig) sca maxseed = `i' } sca sumprop = sumprop + r(prop_sig) di as res %5.1f 100*r(prop_sig) "% of 100 (cumulated: " /* /// */ %7.3f 100*sumprop/`i' " % of " %10.0fc `i'*100 /* /// */ ") sample means differ significantly from mu (max = " /* /// */ %5.1f 100*maxprop " %)" } di as res _n "minprop = " minprop " (seed = " minseed /* /// */ "), maxprop = " maxprop " (seed = " maxseed ")" * ---------------------------------------------------------- */