Transcript Slide 1
Single Factor Research data Central tendency variablity correlation Nominal mode range phi Ordinal median range Spearman rho Interval/ Ratio Skewed median range Convert to ranks Spearman’s rho Interval/ratio mean Standard deviation Pearson r regression Type of design Parametric Non parametric One sample Population sd known Z test none One Sample Population sd unknown One sample t-test none Two independent samples Independent t test Mann-Whitney U Rank sums test, one way chi-square Two related samples Related (paired t test) Wilcoxon test Three or more independent samples (1 factor) Between subjects ANOVA Post hoc test – HSD or protected t Kruskal Wallis H Or one way chi squared Post hoc test Rank sums test Three of more related samples ( 1 factor) Within subjects ANOVA Post hoc test – HSD or protected t Friedman Post hoc Nemenyi’s test • If I want to test the efficacy of a new drug on blood pressure then I have some choices of design…. • Two independent samples • Control vs drug groups (random assignment) • Comparison is independent t- test • Or Mann -Whitney U • Non-equivalent groups • Eg Sick vs Healthy • Male vs female…. • Comparison is independent t- test • Or Mann -Whitney U • Matched design. • Pairs of twins sign up for study • One twin is in control group the other in drug group • Comparison is paired t-test • Or Wilcoxon • Repeated design • Take a baseline measure of BP for each subject • Then give drug • Repeated (paired) t-test • Wilcoxon CONTROL GROUPS – provide baseline for comparison No treatment • Compare treatment vs no treatment • (no treatment only difference is variable of interest) Standard control for : • Situational • Task or • Instructional difference Placebo • Standard for drug trials • Or obvious clinical treatments • Inert substance given for comparison • Tests the protocol • Tests the beliefs of treatment Waiting List • If no treatment or placebo are unethical or not possible then assign subjects to waiting list. • Must be done randomly (not first come first serve) Yoked • Executive Monkey Experiment • Two monkeys ‘yoked’ together • Executive controls off switch the other is helpless • Both get shocked equally No treatment Placebo Waiting List Yoked Perception of Treatment none Believe treated Believe will be none Beliefs effect Beliefs irrelevant Beliefs relevant Beliefs Relevant Beliefs Irrelevant yes no yes yes yes yes yes Different amount of IV possible No treatment baseline IV - control Only two levels Advantage • Simple • Direction of difference obvious Disadvantage • Deceptive appearance of linearity • Single hypothesis test