Presentation on theme: "STA291 Statistical Methods Lecture 22. Really common problem Want to make an inference (estimation or hypothesis test) about the difference between two."— Presentation transcript:
Really common problem Want to make an inference (estimation or hypothesis test) about the difference between two (comparable) parameters: o Compare average ROI for two investment firms o See whether a candidate’s favorability increased or decreased after a particular occurrence o Find which analgesic provides the best pain relief 2
Matched-Pair Experimental Design Common analysis situation: variability among subjects (much) greater than difference between treatments To control for variability among subjects, we use each subject as her/his/its own control – boils down to observation of difference as variable of interest 3
Assumptions (Simple) random sample Especially with small n, normality of differences Paired data – natural connection between individual observations in the two data sets 4
Notation***** Some texts introduce new notation: ( y 1i, y 2i ) d i = y 1i – y 2i 5 i th matched pair i th difference (what will be our i th obsn.) d-bar, average difference standard deviation of the differences
More notation********** Even more new notation: H 0 : d = 0 6 Null hypothesis Test statistic; with df = n-1 Confidence interval for the mean difference
Example: Car Dealership Suppose your are the owner of a car dealership and you want to test the average difference two of your sales people, Tyrone and Shannon, are willing to give in discounts per car to customers. You randomly select 30 cars and ask each sales representative how much he would give in discounts to each of the 30 cars. You find the sample difference to be $64.40 with a standard deviation of $146.74. Test the hypothesis that the mean discount each sales rep. would give is the same and give a 95% percent confidence interval for the difference.
Looking back o Two-sample problems in general o Difference of means: matched- pair data o Assumptions o Hypothesis testing o Confidence intervals