Chi-Square Test -- 2 X Test of Goodness of Fit 1 (Pseudo) Random Numbers Uniform: values conform to a uniform distribution Independent: probability of observing a particular value is independent of the previous values Should always test uniformity 2 Test for Independence Autocorrelation Test Tests the correlation between successive numbers and compares to the expected correlation of zero e.g. 2 3 2 3 2 4 2 3 There is a correlation between 2 & 3 We won’t do this test software available 3 Hypotheses & Significance Level Null Hypotheses – Ho Numbers are distributed uniformly Failure to reject Ho shows that evidence of non-uniformity has not been detected Level of Significance – α (alpha) α = P(reject Ho|Ho is true) 4 Frequency Tests (Uniformity) Kolmogorov-Smirnov More powerful Can be applied to small samples Chi Square Large Sample size >50 or 100 Simpler test 5 Overview Not 100% accurate Formalizes the idea of comparing histograms to candidate probability functions Valid for large samples Valid for Discrete & Continuous 6 Chi-Square Steps - #1 Arrange the n observations into k classes Test Statistic: X2 = Σ (i=0..k) ( Oi – Ei)2 / Ei Oi = observed # in ith class Ei = expected # in ith class Approximates a X2 distribution with (k-s-1) degrees of freedom 7 Degrees of Freedom Approximates a X2 distribution with (k-s-1) degrees of freedom s = # of parameters for the dist. Ho: RV X conforms to ?? distribution with parameters ?? H1: RV X does not conform Critical value: X2(alpha,dof) from table Ho reject if X2 > X2 (alpha,dof) 8 X2 Rules Each Ei > 5 If discrete, each value should be separate group If group too small, can combine adjacent, then reduce dof by 1 Suggested values n = 50, k = 5 – 10 n = 100, k = 10 – 20 n > 100, k = sqrt(n) – n/5 9 Degrees of Freedom k – s – 1 Normal: s=2 Exponential: s = 1 Uniform: s = 0 10 X2 Example Ho: Ages of MSU students conform to a normal distribution with mean 25 and standard deviation 4. Calculate the expected % for 8 ranges of width 5 from the mean. 11 X2 Example Expected percentages & values <10-15 = 2.5% 15-20 = 13.5% 20-25 = 34% 25-30 = 34% 30-35 = 13.5% 35-40> = 2.5% 5 27 68 68 27 5 12 X2 Example Consider 200 observations with the following results: 10-15 = 1 15-19 = 70 20-24 = 68 25-29 = 41 30-34 = 10 35-40+ = 10 13 Graph of Data 70 60 50 40 30 20 10 0 10 15 20 25 30 35 14 X2 Example X2 Values – (O-E)2/E 10-15 = 15-20 = 20-25 = 25-30 = 30-35 = 35-40+ = Total (5-1)2/5 (27-70) 2/27 (68-68) 2/68 (68-41) 2/68 (27-10) 2/27 (5-10) 2/4 3.2 68.4 0 10.7 10.7 5 98 15 X2 Example DOF = 6-3 = 3 Alpha = 0.05 X2 table value = 7.81 X2 calculated = 98 Reject Hypothesis 16