Statistical Analyses: Chi-square test Psych 250 Winter 2013 Types of Measures / Variables • Nominal / categorical – Gender, major, blood type, eye color • Ordinal – Rank-order of favorite films; Likert scales? • Interval / scale – Time, money, age, GPA Main Analysis Techniques Variable Type Example Commonly-used Statistical Method Nominal by Nominal blood type by gender Chi-square Scale by Nominal GPA by gender t-test GPA by major Analysis of Variance weight by height GPA by SAT Regression Correlation Scale by Scale Question Do men and women differ in the % that choose jail time vs. probation only? Main Analysis Techniques Variable Type Nominal by Nominal Example Commonly-used Statistical Method Chi-square (categorical by categorical) blood type by gender Scale by Nominal GPA by gender t-test GPA by major Analysis of Variance weight by height GPA by SAT Regression Correlation Scale by Scale Stat Analysis / Hypothesis Testing 1. Form of the relationship 2. Statistical significance Variables: Categorical by Categorical • Form of the relationship: Cross-tab = two-way table • Statistical Significance: Chi Square [ if n very small Fisher’s exact test ] Example: cross-tab Males Probation Jail n = 24 n = 16 16 4 8 12 n = 20 Females n = 20 Example: cross-tab Males Probation Jail n = 24 n = 16 16 4 n = 20 Females n = 20 80% 8 20% 12 40% 60% Example • Men more likely to choose probation in the sample • Can we infer men in general more likely to choose probation? Statistical Significance Statistical Significance • Q: Is this a “statistically significant” difference? • Can the “null hypothesis” be rejected? Null hypothesis: there are NO differences between men and women in sentencing sample Sample n = 40 inference M: 80% probation F: 40% probation Universe n=∞ M: ?% probation F: ?% probation sample Sample n = 40 inference M: 80% probation F: 40% probation Universe n=∞ Null Hypothesis: M% = F% Logic of Statistical Inference 1. If the Null Hypothesis is True… … what are the expected frequencies for Men and Women in any sample? 2. Do the frequencies in my sample (n = 40) differ from the expected frequencies? Testing Null Hypothesis: Expected Frequencies Males Probation Jail n = 24 n = 16 exp: 12 exp: 8 exp: 12 exp: 8 n = 20 Females n = 20 Observed & Expected Frequencies Males n = 20 Females n = 20 Probation Jail n = 24 n = 16 16 4 exp: 12 8 exp: 12 exp: 8 12 exp: 8 Logic of Statistical Inference • What is the probability of drawing the observed sample (M = 16 probation vs. F = 8 probation) from a universe with no differences? • If probability very low, then differences in sample likely reflect differences in universe • Then null hypothesis can be rejected; difference in sample is statistically significant Statistical Significance • If probability of obtaining my sample is less than 5 in 100, the null hypothesis can be rejected, and it can be concluded that the difference also exists in the universe. p < .05 • The finding from the sample is statistically significant Strategy • Draw an infinite number of samples of n = 40, and graph the distribution of their male vs. female probation %-s Samples of n = 40 Universe n = ∞ M: 80% F: 40% M: 60% F: 50% Null Hyp: M = 60% probation F = 60% probation M: 70% F: 70% M: 50% F: 65% Chi Square Distribution M% = F% 2.5% of area F% > M% 2.5% of area M% > F% Statistical Significance • If probability of obtaining my sample is less than 5 in 100, the null hypothesis can be rejected, and it can be concluded that the difference also exists in the universe. p < .05 • The finding from the sample is statistically significant Testing Null Hypothesis: Sample with small difference Males n = 20 Females n = 20 Probation Jail n = 24 n = 16 13 7 exp: 12 11 exp: 12 exp: 8 9 exp: 8 Sample N = 40 M = 65% F = 55% sample Universe N=∞ p < .05 ? M = 60% probation F = 60% probation Males n = 20 Females n = 20 Probation Jail n = 24 n = 16 13 7 exp: 12 11 exp: 8 9 exp: 12 Chi Square p = .519 exp: 8 Small Difference • p = .519 • Over 50% of samples drawn from null hypothesis universe will have differences this large (65% vs. 55%) • Difference is not statistically significant Testing Null Hypothesis: Sample with large differences Males n = 20 Females n = 20 Probation Jail n = 24 n = 16 16 4 exp: 12 8 exp: 12 exp: 8 12 exp: 8 Sample N = 40 M = 80% F = 40% sample Universe N=∞ p < .05 ? M = 60% probation F = 60% probation Males n = 20 Females n = 20 Probation Jail n = 24 n = 16 16 4 exp: 8 exp: 12 8 12 exp: 12 Chi Square p = .010 exp: 8 Report Findings • “Men were found to choose probation more frequently than women: 80% of the time vs. 40% of the time (df = 1, χ2 = 6.67, p. < ,05).” • “Men chose probation 80% of the time, and women only 40% of the time, a difference which was statistically significant (df = 1, χ2 = 6.67, p. < ,05).”