PSY 2116 Lecture Notes - Lecture 2: Skewness, Central Limit Theorem, Missing Data
5 Mar 2016
School
Department
Course
Professor
Document Summary
Run missing values analysis generate new data to replace. If less than 5% left missing like demographic. Demographic variables: assign missing code and leave it missing. Nominal/ordinal: if value is more than 5% of missing data on particular variable, make new category e. g. not disclosed) can be created to represent scores. Detect via graphs (histogram, box plot) or statistically using z-distribution, winsorizing. In z-score outlier that is bigger than +/- 1. 96 is outlier. Types of outliers: univariate: unusual scores in observed variables, bivariate: result from combo of two variables, combo of multiple variables. Substitute outlier score with next closest value that is not an outlier. If sample is large, outliers are more of an issue. Detection of normality: eyeball test (histogram) sample larger than 300, skewness + kurtosis - divide values by their standard error a. b. c. If it exceeds cut-off value conclude non-normal data: formal normality test. Test if variance in diff groups the same.
