CRIM 320 Lecture Notes - Lecture 5: Situation Two, Statistical Power, Level Of Measurement

• School: Simon Fraser University
• Department: Criminology
• Course: CRIM 320
• Professor: Stephanie Dawson

Document Summary

Chi-square: what happens when normality cannot be assumed, continued. What do you do? (when cell is lower than 5) Two options to deal with statistical chi-square limitation: collapse categories, conservative vs. non-conservative, drop offending category (category with low cell count, conservative vs ndp. Smaller sample size, more categories -> reduce the statistical power of the test, must harder to find true relationship. Limitations of chi-square: expect cell counts must be greater than 5, often can"t tell us about the relative strength of relationship, statistical vs substantive importance (substantive = why we care) Phi: only for 2 by 2 tables, n= sample size, cohen"s convention. Cramer"s v: use when table is larger than 2 by 2. Like chi-square, is a measure of association of nominal data. Instead of association, lambda treats one variable as dependent and the other as independent. Unlike chi-square (and its extensions), it is a pre (proportional reduction in error) measure.