Statistics Lecture 7: Types of Tests

Lecture 7: Types of Tests How do you know Which test to use? 1.​ What data do you have? 2.​ What level of data do you have? a.​ Discrete: Chi square, logistic, regression b.​ Continuous: T tests, correlation, ANOVA, linear regression 3.​ How is it distributed? a.​ Normal Distribution i.​ More powerful, less chance of type 2 error ii.​ Can draw more conclusions iii.​ Poor with small sample size iv.​ Assumes homogeneous data v.​ Ratio or interval b.​ Non-Normal Distribution i.​ Less powerful, less information ii.​ More chance of type 2 error, decreased power iii.​ Good for small sample sizes iv.​ Nominal or ordinal v.​ Less affected by outliers 4.​ Is there a relationship between the groups/samples? 5.​ What is the purpose of your test? 6.​ How many groups or variables are you examining? Parametric vs. Nonparametric Tests Parametric -​ -​ -​ -​ -​ -​ -​ Assuming normal (gaussian) distribution More powerful = less chance of type 2 error Can draw more conclusions Poor with small sample size Errors with non-normally distributed data Assume homogeneous data Typically ratio or interval level Nonparametric -​ -​ -​ -​ -​ -​ -​ Not normally distributed or unknown distribution Less powerful - uses less information Decreased power: more chance of type 2 error Good for small sample sizes Can be used for normal or non-normal distributions Typically ordinal or nominal level Less affected by outliers Types of Tests Parametric Tests -​ Independent t-test (unpaired t-test) Nonparametric Tests -​ Mann-Whitney U test -​ -​ -​ Paired sample t-test (repeated measures) One-way independent ANOVA Pearson correlation coefficient or regression -​ -​ -​ Wilcoxon signed rank test (Wilcoxon matched pairs signed rank test) Kruskal-Wallis test Spearman’s Rank Order Correlation Dependent Data -​ Data is has a relationship (such as genetics or home environment) -​ Can be two measures done on the same person (i.e a before and after) -​ Tests: paired t-test (if normal distribution) or Wilcoxon signed rank test (if non-normal) Independent Data -​ Data does not have a relationship (different people) -​ Student's t-test (aka independent samples t-test, independent t-test) -​ Null: the two sample means for the groups are equal -​ Used for determining if one treatment/group/outcome is better -​ Used to compare the means between groups Dependent and Independent Question Examples: -​ Exam scores in two sections in N341→ independent -​ Number of days of work missed last month by nurses on a surgical unit and a medical unit → independent -​ Duration of hospital stay in two groups of patients admitted with the same diagnosis → independent -​ The pain scores of a group of patients who were in a cross-over study comparing a new pain medication to standard care → dependent -​ The quality of life scores of two sets of twins before and after a teaching intervention → dependent Need to Know if Paired and Distribution: -​ Paired / Dependent -​ Non normally distributed = Friedman’s test -​ Normally distributed = repeated measures ANOVA -​ Unpaired / Independent -​ Non-normally distributed = Kruskall-Wallis test -​ Normally distributed = ANOVA ANOVA -​ Anova is used to compare the means of more than two groups (3+) -​ Null hypothesis: the means of groups 1, 2, and 3 are equal -​ Alternate hypothesis: the means of groups 1, 2, and 3 are different from each other -​ Produces F ratio -​ Examining the ratio of variance between groups and variance within groups -​ Post-hoc tests needed VIDEO CHOOSING A STATISTICAL TEST Paired t-test looks at data connection and correlation between two events (ex. Weight before and after) 2-sample t-tests look at two data points ignoring their correlation